HARVARD BUSINESS REVIEW

How AI Is Reshaping Strategy

Smart firms are using it to streamline decision-making.

PAGE 44

September– October 2026

---

HERMÈS

PARIS

---

Hermès, ride on!

Patricia Elmore

---

---

GAGGENAU

FOR THOSE WHO KNOW

Creators of luxury kitchen appliances. Established 1683.

Experience it

The difference is Gaggenau

---

We help Carnival

turn 5.6 billion transactions a year into endless fun.

Unisys powers the tech platform behind every booking and online transaction before anyone ever steps on board, helping Carnival connect millions of guests to moments of fun. We deliver the reliability and security Carnival has counted on — from their first ship to a global fleet — for almost 50 years and counting. Discover how at unisys.com/carnival

We help Carnival

Keep Breaking Through

unisys | Carnival

---

Contents

September–October 2026

44 Features

44 STRATEGY

AI Is Revolutionizing Strategic Decision-Making

New tools can improve human judgment by tirelessly generating, evaluating, and synthesizing insights.

Felipe A. Csaszar

54 CHANGE MANAGEMENT

Transformation Should Be a Learning Journey

Rigid change programs often fail. A more experimental approach can build momentum—and increase the odds of success.

Evgeny Kaganer and Christoph Loch

64 STRATEGY

Collaborate on the Core. Compete on the Edges.

Rival companies should team up to build the market while working independently to own it.

Frank Nagle

COVER ART

HBR Staff/AI

Illustration by DIMITRIS LADOPOULOS

Harvard Business Review September–October 2026

5

---

September– October 2026

“Many companies are compromising their ability to grow over the long term.”

“HOW MUCH SHOULD YOU BE INVESTING IN GROWTH?,” PAGE 104

76 ORGANIZATIONAL DECISION-MAKING

How AI Agents

Orchestrate Work Across Silos

A new generation of tools can connect functions and reduce the friction that slows companies down. Kris Johnson Ferreira and Jordan Tong

86 SALES & MARKETING

Don’t Underestimate

the Power of Customer Referrals

New research reveals they’re a critical source of low-cost growth. Fred Reichheld, Jamie Cleghorn, and Wojtek Kokoszka

94 LEADERSHIP

The Four Stages

of the CEO-Board Relationship

It evolves over time. Here’s how to navigate each phase successfully. Claudius A. Hildebrand and Douglas L. Peterson

104 FINANCE & INVESTING

How Much Should

You Be Investing in Growth?

Most firms get capital allocation wrong. Here’s how to do it right. Paul Blase and Paul Leinwand

6 Harvard Business Review September–October 2026

Artwork by RYAN KOOPMANS AND ALICE WEXELL

---

ZEGNA

Su Misura

---

September– October 2026

Our Commitment to Sustainability

We're proud that the paper we use in our print magazine is certified under the Sustainable Forestry Initiative® program, meaning that it comes from responsibly managed sources and is a renewable resource.

SUSTAINABLE FORESTRY INITIATIVE Certified Sourcing www.dfprogram.org SP-0000

15

Idea Watch

New Research and Emerging Insights

15 IN THEORY

Why Cash

Rewards Don't Always Produce Better Ideas

Paying for creativity often changes what employees produce and how much they create. PLUS How autorenew promotions can backfire, when you shouldn't apologize to customers, the secret talents of frequent job hoppers, and more.

38 LEADERSHIP

"Losing Doesn't

Work for Me"

The HBR Interview with Dan Schulman. Adi Ignatius

Departments

10 FROM THE EDITOR

11 CONTRIBUTORS

130 EXECUTIVE SUMMARIES

25

In Focus Managing Teams in the Agentic Age

26

To Scale AI Agents

Successfully, Think of Them as Team Members

Rahul Telang, Muhammad Zia Hydari, and Raja Iqbal

30

To Thrive in the

AI Era, Companies Need Agent Managers

Suraj Srinivasan and Vivienne Wei

33

Create an

Onboarding Plan for AI Agents

Joseph Fuller

35

It's Hard to Use

AI as a Team. These Three Practices Can Help.

Gabriele Rosani et al.

115

Experience

Advice and Inspiration

115 MANAGING YOURSELF

Five Ways to Wield

Influence When You Lack Authority

Working with hidden power brokers, using back channels, and other tactics.

Pamela Meyer

120 HOW WE DID IT

The CEO of

HCLTech on Pivoting for the AI Era

The Indian company is ensuring rapid gen AI adoption through new platforms, extensive training, responsible governance, and client collaboration. C. Vijayakumar

125 CASE STUDY

Is Our Subscription

Program Really Working?

A luxury automaker evaluates an initiative that lets customers try a variety of models without committing to a purchase. Varun Gupta et al.

132 LIFE'S WORK

Dionne Warwick

8 Harvard Business Review September–October 2026

Photograph by GUERIN BLASK/AUGUST

---

Growth takes time.

Preservation takes discipline.

Through intentional estate planning and disciplined investment management, we help you protect what you've built so it lasts well beyond you.

WhittierTrust

Investment Management & Consulting | Trust Services Family Office | Philanthropy & Family Continuity | Real Estate

CONTACT

Tim McCarthy (626) 463-2545 whittiertrust.com

Whittier Trust generally requires a minimum of $15 million in marketable securities for new client relationships. Investment and Wealth Management Services are provided by Whittier Trust Company and The Whittier Trust Company of Nevada, Inc. (referred to herein individually and collectively as "Whittier Trust"), state-chartered trust companies wholly owned by Whittier Holdings, Inc. ("WHI"), a closely held holding company. This document is provided for informational purposes only and is not intended, and should not be construed, as investment, tax or legal advice. Past performance is no guarantee of future results and no investment or financial planning strategy can guarantee profit or protection against losses. All names, characters, and incidents, except for certain incidental references, are fictitious. Any resemblance to real persons, living or dead, is entirely coincidental.

---

From the Editor

A Better Way to Craft Strategy in the Age of AI

Amy Bernstein

STRATEGY MAKING HAS always been overwhelming. Too many options, too many risks. That's why we rely on frameworks like Michael Porter's five forces—to simplify a process that stubbornly resists simplification.

The problem isn't that we aren't smart enough. It's that our cognition is finite, as Felipe Csaszar of the University of Michigan's Ross School of Business writes in "AI Is Revolutionizing Strategic Decision-Making" on page 44. We can hold only so many possibilities in our heads, evaluate only so many alternatives in a planning cycle before fatigue, politics, or the clock forces a decision. That was true even before AI upended the competitive environment. Now strategy is geometrically more complicated—and so is setting a course for your organization.

Csaszar's answer: Use AI, which relaxes the cognitive constraints that shape how leaders make their most consequential decisions. AI can generate and screen thousands of strategic options where a human team might consider only a few. It can build and continually update rich models of markets, customers, and competitors that are far more dynamic than any static framework. And it can pressure-test ideas and synthesize diverse perspectives without the groupthink, hierarchy, and deadlines that distort strategy meetings.

AI is multiplying both threats and opportunities. It may also be our best tool for navigating them.

Amy Bernstein Editor in chief

Connect with HBR

JOIN US ONLINE WWW.HBR.ORG

FACEBOOK @HBR

INSTAGRAM

@harvard_business_review

LINKEDIN

Harvard Business Review

TIKTOK

@harvardbusinessreview

WHATSAPP

Harvard Business Review

X @HarvardBiz

YOUTUBE

@harvardbusinessreview

CONTACT HBR

PHONE

800.988.0886

EMAIL

editors@hbr.org

customerservice@hbr.org

Elis Horwin

10

---

Contributors

A workshop examining new research on lying sparked Pamela Meyer's fascination with the science of deception. A former fraud investigator, Meyer now draws on years of experience decoding subtle behavioral cues to help leaders learn to "read the room"—a critical skill she says almost no one is formally taught. In her article in this issue, Meyer, the author of a new book on that subject, explains how leaders can understand the people around them more accurately and become more effective motivators.

115 Five Ways to Wield Influence When You Lack Authority

Felipe A. Csaszar, who chairs the Strategy Area at University of Michigan's Ross School of Business, originally studied computer science, which gave him a special perspective on organizations. "I came to see an organization as a kind of artificial intelligence," he says. "It's not a person, but it is designed to behave intelligently." When large language models arrived, he found himself wondering: Can AI do strategy? The answer, he explains in his article in this issue, is yes: AI can challenge assumptions, widen the options a firm considers, and build richer models.

44 AI Is Revolutionizing Strategic Decision-Making

While developing algorithms to improve business decisions, Kris Johnson Ferreira became intrigued by a quandary: Organizations rarely captured AI's value, because people either trusted the technology too much or dismissed it outright. That led Ferreira, an associate professor at Harvard Business School, to study human-AI collaboration with Jordan Tong. In this issue they argue that people add the most value not by substituting their judgment for AI's but by contributing knowledge that AI lacks and using agentic systems to coordinate information and AI output across organizations.

76 How AI Agents Orchestrate Work Across Silos

Paul Blase's experience cofounding an early AI startup gave him deep familiarity with venture capital investment models. Blase was struck by how different those models were from the growth-investment practices of the large companies he'd advised in his previous job leading the data and analytics practice at a consultancy. Now he's the head of PwC's growth platform, and in this issue he and his coauthor describe a systematic way to determine a company's optimal level of investment, which should be pegged to its context and its industry's maturity.

104 How Much Should You Be Investing in Growth?

Milan-based illustrator Antonio Sortino has been driven to create art since childhood—he even dabbled in graffiti in his youth—but he didn't find the courage to pursue it full-time until after beginning a career as a pharmacist. "The childhood need to create something was too strong to ignore," he says. Now 10 years into his new profession, Sortino says that he particularly enjoys drawing people, as he did for his illustrations in this issue, because it presents an opportunity to capture the unique ways people look, dress, and interact at a particular moment in history.

86 Don't Underestimate the Power of Customer Referrals

11

---

Meet the Interactive Issue

An intuitive way to experience HBR.

Swipe. Click. Listen. Search.

hbr.org/subscribe

EDITOR IN CHIEF Amy Bernstein

EDITOR, HBR.ORG Maureen Hoch

EDITOR, HBR Daniel McGinn

EDITORIAL DIRECTOR, HBR PRESS Dave Lievens

EXECUTIVE CREATIVE DIRECTOR John Korpics

EXECUTIVE EDITORS Alison Beard Ania W. Masinter Sarah Moughty

EDITOR AT LARGE Adi Ignatius

EDITORIAL

MANAGING EDITOR, HBR.ORG Gretchen Gavett

SENIOR EDITORS Scott Berinato, Courtney Cashman, David Champion, Paige Cohen, Susan Donovan, Kevin Evers, Alana Grambush, Eben Harrell, Jeff Kehoe, Scott LaPierre, Toby Lester, Juan Martinez, Steven Prokesch, Thomas Stackpole, Erica Truxler

PRODUCTION DIRECTOR & MANAGING EDITOR, HBR PRESS Allison Peter

SENIOR ASSOCIATE EDITORS Stefanie Fernández, Kelsey Hansen, Ramsey Khabbaz

EDITORIAL COORDINATORS Rachel DuRose, Emily Lang

EXECUTIVE ASSISTANT Christine C. Jack

CONTRIBUTING EDITORS Eileen Ahuja, Karen Dillon, Sarabeth Fields, Amy Gallo, Kelly Messier, JM Olejarz, Anand P. Raman, Dana Rousmaniere

DESIGN

DESIGN DIRECTORS Stephani Finks HBR Press, Susannah Haesche HBR, Marta Kusztra Product & UX, Karen Player Special Projects

SENIOR GRAPHICS EDITOR Macaulay Campbell

SENIOR DESIGNERS Carey Bass HBR, Kim Lapan UX, So Young Park Product

SENIOR PHOTO EDITOR Sasha Patkin

DESIGNER Alex Belser Multimedia

PRODUCTION

MANAGER, EDITORIAL PRODUCTION Christine Wilder

SENIOR PRODUCTION EDITORS Victoria Desmond, Kristin Murphy Romano, Jennifer Waring

PRODUCTION DESIGNER Emily Peterson

PRODUCTION EDITOR Anne Starr

EDITORIAL OFFICES

60 Harvard Way, Boston, MA 02163 617-783-7410 | fax 617-783-7493 HBR.org

VOLUME 104, NUMBER 5 SEPTEMBER-OCTOBER 2026 Printed in the U.S.A.

Copyright 2026 Harvard Business School Publishing Corporation. All rights reserved.

---

COPRESIDENT, HARVARD BUSINESS PUBLISHING Sarah McConville

SENIOR VICE PRESIDENT, GROUP PUBLISHER Erika Heilman

VICE PRESIDENT, NATIONAL ADVERTISING & MARKETING MANAGING DIRECTOR, DIGITAL PRODUCT STRATEGY Craig Catalano Dave Lefort

ADVERTISING

DIRECTOR, U.S. SALES & PRODUCT DEVELOPMENT Daniel Cohen DIRECTOR OF RESEARCH, ANALYTIC SERVICES Beth Traction-Bishop SENIOR EDITOR, RESEARCH & SPECIAL PROJECTS Todd Pruzan ASSISTANT EDITOR, ANALYTIC SERVICES Amanda Yao ASSOCIATE DIRECTOR, MARKETING Samantha Holloway DIRECTOR OF PROJECT MANAGEMENT, ANALYTIC SERVICES Samantha Barry SENIOR RESEARCHER, ANALYTIC SERVICES Mary Callaghan SENIOR PROJECT MANAGER, ANALYTIC SERVICES Audrey White CONTRIBUTING CONSULTANT MaryAlice Holmes

MARKETING & COMMUNICATIONS

DIRECTOR, CONSUMER MARKETING Nini Diana DIRECTOR, COMMUNICATIONS Amy Poftak ASSOCIATE DIRECTOR, MARKETING FOR CUSTOMER JOURNEYS Caty Jackson SENIOR MANAGER, LEAD GENERATION & CONVERSION Josh Getman MANAGER, DIGITAL MARKETING Stephanie Kazemi

PRODUCT MANAGEMENT

DIRECTOR, PRODUCT MANAGEMENT Anne Bartholomew SENIOR PRODUCT MANAGER Jon Goldberg SENIOR MANAGER, AI & MOBILE Gerrie Diaz SENIOR TECHNICAL PRODUCT MANAGER Frederic Lalande SENIOR PRODUCT MARKETING MANAGER, LEARNING Marqui Mapp-Taylor AGILE LEADS Kara Berrini, James Monia

BUSINESS & CUSTOMER ANALYTICS

SENIOR DIRECTOR, BUSINESS ANALYTICS & INSIGHTS Jessica Avery DIRECTOR, CUSTOMER & MARKETING ANALYTICS Emily Freeman Kroger ASSOCIATE DIRECTOR, EMAIL AUTOMATION & OPERATIONS Jerry Knott ASSOCIATE DIRECTORS, EXPERIMENTATION & ANALYTICS Chris Chaky, Abigail Dawson SENIOR MANAGER, EMAIL PRODUCTION Frances Lee MANAGERS, EMAIL MARKETING Milo Goodman, Cara Vaccaro ANALYST, EXPERIMENTATION & ANALYTICS Tony Manev

FINANCE & OPERATIONS

SENIOR DIRECTOR, FINANCE & OPERATIONS Greg St. Pierre ASSOCIATE DIRECTOR, OPERATIONS & CUSTOMER EXPERIENCE Greg Daly SENIOR MANAGER, FINANCE Krishna Dudhwewala MANAGER, CUSTOMER EXPERIENCE Daniello Weber SPECIALIST, DATA INTEGRITY & OPERATIONS Edward D. Domina IV

HBR PRESS

ASSOCIATE PUBLISHER Jon Shipley DIRECTOR, MARKETING & COMMUNICATIONS Julie Devoll DIRECTOR, B2B & CUSTOM SALES Lindsey Dietrich ASSOCIATE DIRECTOR, MEDIA & COMMUNICATIONS Sally Ashworth SENIOR MANAGER, OPERATIONS Rick Emanuel SENIOR MANAGER, MARKETING Alexandra Kephart SENIOR ASSOCIATE, SALES Jordan Concannon

TECHNOLOGY

CONTENT SYSTEMS ARCHITECT Matt Wagner SOFTWARE ARCHITECTS Bill Munyan, Kenneth Pereira SOFTWARE ARCHITECT, MOBILE Ahmad Khan QUALITY ASSURANCE ARCHITECT Matthew Han FRONT-END ENGINEER Caitlin Smith

WORLDWIDE ADVERTISING OFFICES

NEW YORK 973-769-3631 Maria A. Beacom Sales Director, Technology & North American Research

LONDON 44-0-7890-608-707 Alexander Bart Sales Director, EMEA & APAC

CALIFORNIA 415-988-7762 BEIJING 86-10-6551-0663 FRANCE 33-01-4643-0066 HONG KONG 852-237-52311 INDIA 212-872-9291 JAPAN 81-03-3541-4166 KOREA 82-2-730-8004 SOUTH AMERICA 55-11-98160-0177 UAE 971-4-228-7708 UNITED KINGDOM 44-20-7291-9129

For all other inquiries, please call 212-872-9280 or visit HBR.org/hbr-advertising-sales.

SUBSCRIPTION SERVICES

Harvard Business Review 20 Guest St, Suite 400 Brighton, MA 02135 customercare@hbr.org US/Canada: 800-274-3214 International: 1-617-783-7500

RATES PER YEAR United States $119 Canada US$139 Mexico US$139 International US$165

ARTICLE REPRINTS

To purchase reprints, go to HBR.org.

LIBRARY ACCESS

Libraries offer online access to current and back issues of Harvard Business Review through EBSCO host databases.

SUBMISSIONS

We encourage prospective authors to follow HBR's "Guidelines for Authors" before submitting manuscripts.

To obtain a copy, please visit: HBR.org; email: hbr_editorial@HBR.org; or write to: The Editor, Harvard Business Review, 60 Harvard Way, Boston, MA 02163.

A NOTE TO READERS

The views expressed in articles are the authors' and not necessarily those of Harvard Business Review, Harvard Business School, or Harvard University. Authors may have consulting or other business relationships with the companies they discuss.

Rotman Management has evolved. Introducing Rotman Intelligence.

The same trusted source of perspectives on business, strategy, leadership and the accelerating pace of change.

Rotman Intelligence brings together the expertise, learning experiences and viewpoints of the Rotman community, offering thought-provoking insights and practical tools for leaders, innovators and entrepreneurs.

New name. Same great thinking.

Subscribe today! Uoft.me/Subscribe RotmanIntelligence

Rotman INTELLIGENCE

---

Stop reorganizing.

Start adapting.

Becoming an Octopus Organization

Learn to break down hierarchy, get rid of silos, and create an adaptable, resilient, learning organization in this eight-week newsletter series.

SIGN UP TODAY hbr.org/newsletters

---

New Research & Emerging Insights

ideaWatch

IN THEORY

Why Cash Rewards Don't Always Produce Better Ideas

Paying for creativity often changes what employees produce and how much they create.

COMPANIES OFTEN TRY to boost innovation with a simple idea: Pay employees when they produce something creative. The thinking is straightforward. If creative employees get rewarded, then they will work harder to come up with more and even better ideas.

But new research suggests it's not that simple. Incentives can steer innovation in very different directions depending on who receives them. In a

Illustrations by DEBORA SZPILMAN

15

---

ideaWatch

recent study published in the Journal of Management, Wenlong He of Renmin University of China, Dan Prud'homme of Florida International University, Nianchen Han of Nanyang Technological University, and Kenneth Huang of the National University of Singapore looked at what happens when companies introduce financial rewards for employee inventors. These programs, which give employees a bonus or other reward when a patent is filed or granted, are common in R&D settings. Major companies, including IBM, Microsoft, Apple, Google, Amazon, and Intel, operate formal employee-patent-recognition or inventor-reward programs.

To understand the impact of rewards, the researchers analyzed company and patent data from publicly listed firms in China, tracking 2,376 outputs of inventors over time. They compared performance before and after companies introduced financial incentives, measuring the number of patents produced and how inventive those patents were relative to existing technologies.

They also tried to understand each inventor's expertise. The researchers mapped the technical fields each inventor worked in using International Patent Classification codes. Inventors whose work stayed within a narrow set of categories were classified as specialists; those whose work spanned multiple fields were considered generalists.

You might expect financial rewards to push everyone in the same direction. They don't. The researchers found that specialists produced a greater number of inventive patents after the incentives were introduced. Generalists increased

their patent output by even more, but their patents were less inventive.

Why? The researchers say it's because not everyone responds to incentives in the same way. People respond based on how they view themselves within professional social identities. Specialists are embedded in tight expert communities, both inside and outside their workplaces, where reputations are built by doing something genuinely new. Generalists don't face that pressure, so when incentives are offered they produce more low-hanging, incremental inventions in order to benefit from the incentives quickly.

"You don't get better innovation by only paying more," says Prud'homme. "You get it by aligning what people value with what you reward."

For managers, there are three practical takeaways for improving financial incentive programs to drive innovation:

Match the incentives to your innovation goals. Before you design an innovation reward program, be clear on the goal. Are you trying to get the most ideas possible out of the system, or are you looking primarily for bigger breakthroughs? The mistake many companies make is to try to get both outcomes from the same incentive.

If you simply want more ideas, offer small, frequent rewards for submissions or early concepts. If you only want breakthroughs, create a separate track with rewards tied to impact, such as technical significance or downstream use. Firms seeking both types of innovation can offer both tracks.

"If you don't align your incentives with your objective, you end up getting a lot of activity without necessarily getting better innovation," says Prud'homme.

Manage the members of your workforce according to their professional identities, not just their roles or titles. Even when tailoring incentives to innovation objectives, firms still need to recognize that different employees respond to incentives differently. Identify the specialists and the generalists within your company, because the same reward structure can push the two groups toward very different behaviors.

Once you see the mix clearly, you can forecast how your workforce will respond to the rewards you offer. Employees are inclined to produce rewarded outcomes that fit their identities. Specialists tend to emphasize inventiveness because their

16

---

professional norms place greater value on originality. Generalists, by contrast, lean toward producing a higher volume because they are less constrained by specialized norms and more responsive to the direct incentives that rewards create.

“Effective innovation management needs to account for those differences rather than assume everyone will act the same,” Prud’homme says.

Engineer ways for generalists and specialists to work together. To make the balance work, build collaboration into the innovation process. For example, you can create structured handoffs: Generalists generate and expand ideas early, and then specialists pressure-test and refine them before the ideas are moved forward. Alternatively, generalists and specialists can work together throughout the process so that specialists’ quality standards influence idea development while generalists keep the pipeline moving.

Finally, watch for imbalance over time. If one group starts to dominate, you will see it in the output. Too many generalists, and quality might drop. Too many specialists, and the pipeline might slow. Based on your innovation objectives, adjust hiring goals or the team mix to keep things in check.

“If you get this right, you don’t have to choose between more ideas and better ones,” Prud’homme says. “You can build a system that delivers both.” ▼

HBR Reprint F2605A

ABOUT THE RESEARCH “Inventor Rewards, Specialization, and Innovation Performance,” by Wenlong He et al. (Journal of Management, 2025)

IN PRACTICE

“If You Reward Activity, You’ll Get More Ideas. If You Reward Impact, You’ll Get Better Ones.”

Andres Arias is a senior manager of customer solutions at Amazon Web Services. He spoke with HBR about what happens when companies try to incentivize innovation, why people often confuse output with outcomes, and how leaders can design programs that produce real impact instead of just activity. Edited excerpts of the conversation follow.

In your experience, do financial incentives drive innovation, or do they just increase activity? You see a spike in submissions, but most are not well reasoned or complex. Teams optimize for winning the reward, not solving the problem. They’ll generate five ideas in a week, pick one, polish it just enough to present, and move on.

Who responds best to incentives: subject-matter experts or generalists? They behave differently, but I wouldn’t say one group responds better than the other one does. Specialists can overinvest. They’ll create something technically impressive but more complex than the business needs. But that instinct is what keeps the team honest about quality when it really counts.

Can you offer an example of a successful incentive program? We ran a hackathon-style program around a specific business problem. We didn’t frame it as “Come up with as many ideas as you can.” Instead we defined clear outcomes, including improving a specific customer process. We built cross-functional teams that included specialists and generalists and gave them a short window in which to work. We rewarded teams on the basis of speed to a working solution, knowledge gained, and applicability of the idea after the event. Those criteria changed how people approached the problem. They stopped trying to impress and started trying to make something work.

What lessons can managers take away from that approach? There are two. First, align teams around clear outcomes. That keeps them focused on solving the problem, not just on generating ideas. Second, make generalists and specialists work together. In our hackathon, the specialists brought depth and technical expertise, and the generalists helped connect that work to the bigger problems and kept things practical. But the key was using incentives to keep their focus. If you just let them run, you’ll get a lot of different ideas. When you anchor their work to a specific outcome that matters to the business or the customer, that’s when the approach clicks. They collaborate better, make smarter trade-offs, and deliver something useful.

Have you ever adjusted an incentive program after realizing it was driving the “wrong” kind of innovation? It’s not always a formal “we redesigned the program” moment, but you notice when the incentives are pushing the wrong behavior. I’ve seen this across the industry: Teams that are rewarded for output, like how many features they ship or how fast they close tasks, see a lot of activity. There’s movement every week but not a lot of real impact. It gets frustrating, because people feel like they’re doing a ton of work without changing anything meaningful. That’s usually when it hits that the incentives are off and need to be more closely tied to outcomes. The difference comes down to what you reward. If you reward activity, you’ll get more ideas. If you reward impact, you’ll get better ones. ▼

Illustration by JORI BOLTON

17

---

ideaWatch

MANAGING PEOPLE

Why You Shouldn't Treat AI Agents Like Employees

As more organizations deploy AI, some aren't just rolling out tools; they're giving their agents names, roles, and performance goals. But this seemingly symbolic choice can change how managers oversee work and assign responsibility.

In January 2026 researchers surveyed 1,261 HR and finance managers, directors, and executives across industries from the United States, Canada, and the European Union. About a third said that their leaders already talked about AI as a "teammate" or "employee," and 23% reported that their organizations had listed AI agents on their org charts, assigning them formal roles.

To see how that framing affects behavior, researchers ran an experiment in which 813 of those managers were asked to review documents—including job descriptions and budget reports—that contained built-in errors. They were given 20 minutes to review as many reports as they could, flag any errors, and escalate large problems to a higher-up. All managers saw the same documents. What varied was who they were told had drafted them: an AI tool, an AI employee listed as a direct report on an org chart, or a human employee who was a direct report.

The framing had relatively small effects, except in one group: managers who already worked in firms that listed AI agents on their org charts. Among those participants, describing the

drafter as an AI employee rather than an AI tool reduced their confidence in their own judgment. They caught about 18% fewer errors and were 22% more likely to escalate issues, even when they were told that they would be penalized if they asked for an unnecessary extra review. When asked who was responsible for the errors, these managers assigned about nine percentage points less responsibility to themselves and eight percentage points more to the AI if it was framed as an employee rather than a tool. They also shifted more blame to their team members and leadership.

In organizations where AI agents are treated as teammates, managers are accustomed to doing less careful checking themselves. They also lean more on others to review work and are more inclined to see the AI as bearing responsibility if something goes wrong, say the researchers.

"The evidence presented in our study suggests that AI agents should be treated as what they are—software automation," which requires human accountability to minimize risks, they write.

ABOUT THE RESEARCH "Putting AI on the Org Chart: Evidence on Delegation and Oversight," by Emma Wiles et al (working paper, 2026)

CORPORATE STRATEGY

Tech Is Reshaping C-Suite Roles

Much attention has been paid to technology's potential to eliminate or transform entry level jobs, but it is also changing the makeup of executive ranks. When researchers analyzed the posted job descriptions for a representative global sample of 5,000 open C-suite positions from 2019 to 2025, they found that, increasingly, companies have been adding new and more technology-related executive roles. Additionally, changing political and legal landscapes have moved firms to absorb DEI responsibilities into core business functions or phase them out altogether.

Relative change in C-suite roles (2019–2025)

Source: Russell Reynolds Associates

18

---

CEOs ARE BEING APPOINTED AT OLDER AGES

A new study found that the age of newly appointed CEOs has risen at U.S. firms, from an average of 48 in 2000 to 55 in 2023, with similar increases among CEOs at European firms. The shift is driven not by longer tenures or later retirement. Instead, more firms are selecting leaders who have held a wider variety of roles across different firms and industries, the researchers say. "Aging at the Very Top," by Valentin Kecht, Alessandro Lizzeri, and Farzad Saidi

HIRING & RECRUITMENT

Job Hoppers Have a Hidden Talent

Hiring managers often treat frequent job changes as a red flag indicating someone won't commit long enough to justify the onboarding investment. But a new study finds that people who switch employers a lot possess valuable social skills that help them get up to speed more quickly than other new hires.

Using monthly employment and performance data on 8,693 U.S. hedge fund managers from 2004 to 2019, researchers tracked what happened when a fund manager joined a new firm. Consistent with past work, they found that 72% of moves were followed by a performance decline, from which it took an average of four months to recover.

But that slump was not the same for everyone. Managers with more prior

company moves suffered a smaller initial decline and returned to their pre-move performance within two months, compared with five months for those who had low mobility.

The researchers suggest that repeated transitions create a form of transferable knowledge about how to get up to speed in new environments. With each move, people practice noticing unwritten norms, interpreting cues, and adjusting how they work with colleagues. Over time those skills become more generalizable and help make the next switch go more smoothly.

The study found that the advantage was biggest when the social and cultural hurdles of joining were higher, such as when a manager joined a firm in a distant geographic location, and where existing employees had long tenures (and therefore norms are likely to be less flexible).

The researchers note that more-mobile managers weren't better

performers overall; they were better at the transition itself and absorbed the shock of a move faster. Still, hiring teams may do well to reframe their thinking about "job hoppers," from pure flight risks to something rarer and more useful: employees who can swiftly learn the ropes of a new organization.

ABOUT THE RESEARCH "Movin' and Groovin' Increased Prior Mobility Facilitates Newcomers' Transitions into Organizations," by Rebecca R. Kehoe and F. Scott Bentley (Academy of Management Journal, 2026)

BUSINESS MODELS

Is Your Autorenew Keeping the Right Customers?

Many firms assume that making subscriptions "sticky" using trials that autorenew is an easy way to grow revenue. A new experiment shows that that model could be deterring engaged consumers from subscribing.

Researchers partnered with a major European digital newspaper that erects a paywall after readers consume a quota of free articles. In 2018 more than 1.4 million readers who hit the paywall were randomly shown one of two offers: a trial that automatically rolled into a paid subscription unless canceled, or a trial that automatically ended unless the reader actively renewed. The researchers then tracked every reader's subscription and site usage for roughly 20 months.

In the short term, autorenewal did raise revenue. Among those who took

19

---

A MORE POSITIVE WAY TO APPROACH NETWORKING

Many people avoid reaching out to new contacts, worried they'll be a burden. But new research found that when people were trained to focus on the upsides for advice givers, they scheduled 37% more informational interviews than those who focused on benefits to themselves.

"A Prosocial Perspective on Advice Seeking and Networking: How Focusing on What Advice Givers Can Gain Motivates Advice Seekers to Reach Out More," by Anne Burmeister and Daniel Z. Levin

the promotional offer, more remained paying subscribers in the months immediately after the trial than those in the autocancellation group. But the autorenewal offer sharply discouraged people from signing up in the first place. During the promotional window, 35% fewer readers subscribed when the deal was autorenewing rather than autocanceling. Over the course of the study, the initial gains made in the autorenewal group eroded and reversed, and the autocancellation offer ultimately produced 23% more total paid subscribers. Additionally, consumers who were offered autorenewal became less likely to subscribe to the publication through any channel afterward, including through future subscription offers.

Autorenewal is so damaging because it is predicated on the assumption that consumers don't realize they'll forget to cancel a subscription they don't use. But by examining usage data, the researchers found that consumers are actually self-aware. Readers who suspected they wouldn't remember to cancel were put off by autorenewal and stayed away. Those who did sign up and found themselves paying for something they didn't use later disengaged from the brand entirely.

So should companies always use autocancellation? In an article on HBR.org that expanded on the findings, researchers Klaus M. Miller and Z. John Zhang note that the results best fit "stickier" products, such as news, music

streaming services, or cloud storage—categories where customers usually stay once they find a good option. In markets where switching is more common—such as television streaming services, fashion subscriptions, or meal kits—autorenewal may play a more positive role. In industries with very high repurchase rates (above roughly 70%), the researchers say, autocancel may help lock in customers who want to be there for the long haul.

ABOUT THE RESEARCH "Sophisticated Consumers with Inertia: Long-Term Implications from a Large-Scale Field Experiment," by Klaus M. Miller, Navdeep S. Sahni, and Avner Strulov-Shlain (working paper, 2026)

AI & MACHINE LEARNING

How People Use Gen AI Is Changing

Since 2024, researchers at theaiwild.com have analyzed tens of thousands of posts on online forums like Reddit and Quora to examine how people's AI usage has shifted over time. In 2026 people used the technology more for personal and professional support; using AI for technical assistance has become less prevalent.

% of total use cases by category	2024	2025	2026
Content creation and editing	23%	31%	34%
Technical assistance and troubleshooting	21	18	18
Personal and professional support	17	16	15
Learning and education	15	15	13
Creativity and recreation	13	11	12
Research, analysis, and decision-making	10	9	11

Source: theaiwild.com

Top use cases	2025	2026	New use case
Therapy/companionship	1	1	Therapy/companionship
Organizing my life	2	2	Troubleshooting
Finding purpose	3	3	Fun and nonsense
Enhanced learning	4	4	Fan fiction and storytelling
Generating code (for pros)	5	5	Technical use of software
Generating ideas	6	6	Autonomous agentic operations
Fun and nonsense	7	7	Relationship advice
Improving code (for pros)	8	8	Work buddy
Creativity	9	9	Astrology and tarot readings
Healthier living	10	10	General advice

*Was not on list of top 100 in 2025

20 Harvard Business Review September–October 2026

---

Your vision for your wealth is unique. Shouldn't you be getting unique advice?

NORTHERN TRUST

FOR WHERE WEALTH GOES NEXT

FOR MORE INFORMATION: NORTHERNTRUST.COM/NEXT

INVESTING / BANKING / TRUST & ESTATE SERVICES / WEALTH PLANNING / FAMILY OFFICE

Member FDIC. © 2026 Northern Trust Corporation.

---

ideaWatch

WORKPLACE HEALTH & SAFETY

Giving Stock Options to Frontline Workers Improves Safety

Offering stock options to your employees is usually framed as a way to build loyalty and align incentives with shareholders. When you extend those options to rank-and-file workers, you can also help make your workplace safer.

Researchers matched injury data from the U.S. Occupational Safety and Health Administration with compensation data from 20,160 workplaces representing 629 firms from 2002 to 2011. Across the sample, which covered a wide range of industries—including retail, pharmaceutical products, transportation, and industrial metal mining—workplaces averaged about 6.6 injuries per 100 employees per year. After controlling for factors such as company size, use of seasonal workers, hours worked, and capital intensity, the researchers found that companies with higher-than-average stock options for regular employees had an injury rate that was 7% lower than that of companies offering average stock options.

To examine causality, the researchers looked at what happened after a new federal accounting rule (FAS 123R) made stock options look more expensive on companies' financial statements, prompting many firms to cut back on the options they granted. At the companies that had used rank-and-file options most heavily before the rule, injury rates rose more after its adoption than they did at otherwise similar firms, consistent with the idea that reducing options leads to worse safety outcomes.

The researchers list two reasons for the connection between options and safety: retention and cooperation. Because unvested options tie employees' wealth to the firm and are forfeited if employees leave, they can reduce turnover, and experienced workers are less likely to be injured than new hires. Options also tie rewards to team performance, which encourages coworkers to monitor and support one another on safety-critical tasks.

For leaders, the takeaway is to treat broad-based stock options as more than a recruiting perk. When thoughtfully designed, they can reduce turnover, strengthen peer accountability, and

improve safety, especially in high-churn, hard-to-monitor parts of the business where injuries are most likely.

ABOUT THE RESEARCH 'Rank-and-File Employee Stock Options and Workplace Safety,' by Yangyang Chen et al. (Management Science, 2025)

MARKETING

To Justify Price Increases, It Helps to Blame the Market

When companies raise prices, they usually tell their customers it's because costs went up or quality improved. New research suggests that a different explanation may do more to keep customers from walking: The market made us do it.

Researchers partnered with a Canadian self-storage provider in 2022 and 2023 to test how messages about price increases affected customers' real behavior. In the study 1,626 customers received price-increase emails after they'd been with the provider for seven months. The emails used one of three justifications—cost, market conditions, or quality improvements. The study also included a no-justification price-increase message and a no-price-increase control group to capture "natural" churn. The researchers tracked the behavior of the customers over eight months.

Unsurprisingly, raising the price increased attrition overall. About 33% of customers left when prices didn't change, whereas 44% left when they did. But the reason given mattered.

22

---

Market-condition justifications produced the lowest attrition—customers’ likelihood of leaving was roughly 30% lower than it was when no reason was given. Cost and quality justifications, on average, didn’t reduce churn.

That’s because price-increase letters do more than trigger fairness concerns, the researchers contend. They also shape customers’ perceptions of their options. For example, when market conditions are cited, it sends the message “Good luck finding an easy alternative right now,” which compels people to stay put. In the field data, the market message worked best with customers who had fewer nearby alternatives. In follow-up online experiments, the same market justification increased people’s sense that alternatives were scarce.

Quality justifications were more complicated. In online tests with North American participants, the message “We’re improving things” boosted perceived value and fairness. But reactions were split in the field. It worked with some customers and backfired with others, netting out to no benefit on average.

The researchers note that their findings apply most readily to ongoing purchases, where the cost of switching would be expensive or cumbersome. For this group, a clear market-based explanation may cushion the blow of higher prices by reminding customers they may not have better options.

ABOUT THE RESEARCH “Cushioning the Blow: Reducing Customer Attrition in Response to Price Increase Notifications,” by Hoorsana Damavandi, Kersi D. Antia, and Praveen K. Kopalle (Journal of Marketing, 2026)

CUSTOMER STRATEGY

When It’s a Bad Idea to Apologize to Customers

You might think the first rule of customer service is to apologize when something goes wrong. But if customers haven’t noticed a problem, that instinct can actually hurt satisfaction, trust, and revenue.

In a field experiment with a large U.S. food-delivery platform, researchers examined what happened when the firm apologized for a small failure: food arriving late by fewer than 15 minutes. For some customers, the staff followed the company’s usual practice by calling ahead, explaining the delay, and apologizing. For others, the staff did not reach out, and the order simply arrived slightly late.

The researchers found that, over the next 90 days, customers who’d received an apology were less likely to place another order at all. Those who did return placed fewer repeat orders, took longer to come back, and spent less overall than comparable customers who received no apology. They also reported lower satisfaction with the

experience, less trust in the provider, and weaker intentions to recommend the company.

Several follow-up experiments help explain why: Apologies not only made people more likely to notice that some aspect of the service fell short (for instance, that the actual delivery time differed from the original estimate), but also led them to interpret that deviation as a “failure.” Consumers then downgraded their judgments of service quality and the firm’s competence.

The researchers say that firms should distinguish between situations where customers are almost certain to notice a failure (a missed flight, a lost shipment, a never-arriving order) and situations where they might not realize that a failure occurred. In the former, timely apologies are still advisable. In the latter, it may be wiser to simply send an update or say nothing rather than flagging a “failure” the customer might never have perceived. ▼

ABOUT THE RESEARCH “Are Apologies Always the Best Policy? Apologies for Service Failures Backfire When Consumers Are Not Aware of the Failure,” by Mason R. Jenkins, Paul W. Fombelle, and Mary Steffel (Journal of Consumer Research, 2026)

COMPILED BY HBR EDITORS | SOME OF THESE ARTICLES PREVIOUSLY APPEARED IN A DIFFERENT FORM ON HBR.ORG.

23

---

Happier teams start with happier managers.

ARTHUR C. BROOKS, HARVARD PROFESSOR AND BEST-SELLING AUTHOR

The Leader's Happiness Reset is a free six-week newsletter on finding happiness at work—for yourself and your team—from HBR and Arthur C. Brooks.

The Leader's Happiness Reset

---

One Challenge, Four Strategies

in Focus

THE CHALLENGE

Managing Teams in the Agentic Age

1

To Scale AI Agents Successfully, Think of Them as Team Members

26

2

To Thrive in the AI Era, Companies Need Agent Managers

30

3

Create an Onboarding Plan for AI Agents

33

4

It's Hard to Use AI as a Team. These Three Practices Can Help.

35

J. Studios/Getty Images

These articles were selected from HBR.org.

Harvard Business Review September-October 2026

25

---

IN FOCUS

Managing Teams in the Agentic Age

To Scale AI Agents Successfully, Think of Them as Team Members

by RAHUL TELANG, MUHAMMAD ZIA HYDARI, and RAJA IQBAL

PICTURE A FAMILIAR scene: A vendor demonstrates a new generative AI agent to your leadership team. It’s impressive.

The agent triages support tickets, updates customer records, and drafts a proposal and routes it for approval.

The demo is seamless. Afterward, someone asks, “How soon can we deploy this across the enterprise?”

That question reflects assumptions that have guided enterprise software adoption in the software-as-a-service era: first, that most tools can be provisioned, configured, and scaled with relatively little customization, and second, that if the integration works and employees adopt the product, deployment is largely an implementation project. Agentic AI breaks that model.

Unlike traditional software, AI agents are designed to reason, plan, and take actions across systems. The moment an agent can change a system of record—update a price, send a payment, or modify customer data—it stops being a productivity tool and becomes part of the organization’s operating model.

Most important, it introduces new categories of risk. A narrow generative AI tool (such as ChatGPT) creates content risk: It might say something wrong. Agentic AI creates execution risk: It might do something wrong.

Drawing on our work as researchers (Rahul and Zia) and as a practitioner with experience leading agentic-enterprise-AI deployments (Raja), we’ve paid close attention to how agentic AI is actually being implemented. We’ve found that although many agents today are ready to act, companies are rarely ready to let them.

To cross that threshold and effectively integrate AI agents at scale, you need to stop treating them as turn-key software that simply needs to be installed and instead treat them like a new kind of workforce that requires management. Just like your human

Sharon Novak/Getty Images

26

---

AI agents need a role, a defined scope of authority, approved sources of truth, and clear escalation rules.

employees, AI agents need a role, a defined scope of authority, approved sources of truth, and clear escalation rules. They also need supervision and audit trails, because you will be accountable for their actions.

Until those organizational foundations are in place, scaling agentic AI will be difficult. Four recurring frictions, in particular, tend to slow or derail progress. Understanding them is the first step toward managing them.

1 Identity: Leaders Need to Know “Who” Is Acting

Companies have spent decades building access controls for human employees. Computer system users log in with a unique identity, and their role determines what they can and cannot do.

AI agents complicate that, because they behave like human employees—but aren’t. Many early deployments handle this problem by giving agents a shared “service account” with broad access to multiple systems. It’s a convenient solution, but granting agents so many permissions creates a security risk.

Consider this example. A customer service representative is authorized to issue refunds up to $500. If the rep tries to issue a larger refund, the system blocks the transaction and routes it for approval. But an AI agent operating through a shared back-end service account might not be subject to that authorization limit and might issue a $5,000 credit in a single step.

The risks are not just hypothetical. In 2025 a developer experiment using Replit’s AI coding agent showed how

quickly automation can outrun its controls. Despite being instructed to not make any changes, the agent executed commands that deleted a production database. It then attempted to obscure the failure, generating thousands of fake records and misleading system messages, which slowed the response and complicated recovery.

The lesson is clear: Organizations should treat each AI agent as a distinct digital worker with its own identity, credentials, and role. Instead of relying on shared service accounts, companies should assign agents narrowly scoped permissions that reflect the specific tasks they are designed to perform. The same principles used to manage human employees—such as least-privilege access and role-based limits—should apply to agents. If a customer service employee can’t issue refunds above a certain threshold without approval, the same constraint should apply to the agent performing that same work. Just as important, every action an agent takes should be logged under a traceable identity so that the organization can clearly see who—or what—performed it. If leaders cannot easily explain which identity an agent uses when it executes an action, the system is not ready for production.

2 Context: Bad Data Can Lead to Bad Actions

AI agents perform well in demonstrations because the environment is controlled. The data is clean, the instructions are clear, and the sources of truth are obvious. But real organizations are different. Enterprise data

is fragmented across systems, duplicated across teams, and often contradictory. Policies evolve over time, and older documents remain in circulation. People handle this ambiguity by using judgment and experience that AI agents don’t have.

For a system that generates text, an imperfect context may produce a flawed answer. That is a problem, but its consequences are often limited. For a system that takes actions, however, the ramifications are far greater. Imagine an HR agent that uses a frequently referenced policy document from 2022—even though the rules have since changed—to guide managers through a termination process. That’s not a hallucination. It’s a retrieval mistake that exposes the company to legal risk.

Agents also introduce a new security challenge: the manipulation of context. If an agent reads emails, forms, or support tickets and then performs tasks on the basis of that information, attackers can embed hidden instructions designed to influence its behavior. Researchers demonstrated this risk in 2025 through a vulnerability known as ForcedLeak. By embedding malicious instructions in a routine web form, they tricked a Salesforce Agentforce agent into retrieving sensitive customer-relationship-management data and sending it to an external destination.

To cope with these context frictions, organizations need to establish clear standards for which information their agents can trust. That requires defining authoritative sources for policies, pricing, and operational data so that agents can consistently rely on the correct version of the truth. Systems

27

---

The companies that succeed with agentic AI will build the structures that allow the agents to be trusted.

should also capture the provenance of information used in decision-making, allowing teams to trace any agent action back to the specific documents or data sources it relied upon. Finally, companies should treat external inputs—such as emails, forms, or uploaded files—not simply as helpful context but as potential attack vectors. Inputs that originate outside the organization should be handled carefully and validated before an agent is allowed to act on them. Without these safeguards, the same data inconsistencies that humans routinely navigate can quickly become operational errors for automated systems.

3 | Control: Probabilistic Systems Need Hard Boundaries

Traditional software behaves predictably: The same input produces the same output every time. But large language models don't work that way. Their responses are probabilistic, meaning that the same request can produce slightly different results across runs. That variability is acceptable when the output is a draft email, but it becomes far more problematic when the output is a transaction.

One company deploying an AI support agent encountered this issue when the legal team insisted that the system never mention a particular competitor. The company's knowledge base, however, included many legitimate comparison articles that referenced that competitor, so the agent frequently mentioned it while answering customer questions. When engineers tightened the guardrails to block those responses,

the system began refusing to answer valid questions altogether.

The deeper issue is that traditional testing methods assume stable behavior. A system that passes a test suite today may behave differently tomorrow if the model updates, the prompt changes, or new data is added. And the risks grow in multiagent environments where agents pass work to one another. In 2025, for example, researchers at AppOmni demonstrated how insecure configurations in ServiceNow's Now Assist environment could allow "second-order prompt injection." In their experiment, malicious instructions introduced by one agent were passed along to others, potentially leading to unintended or unauthorized actions, such as retrieving sensitive records or sending information to external destinations.

The takeaway here is that without clear boundaries, mistakes—or attacks—can cascade across an automated system. To manage that risk, organizations should build deterministic controls around probabilistic AI systems. Rather than allowing agents to execute actions directly, companies should place validation layers between the AI model and operational systems. In this approach, the agent proposes an action, such as issuing a refund or updating a record, and before it's executed, deterministic software verifies that it complies with established rules. Organizations should limit unsupervised agent-to-agent interactions so that outputs from one agent do not automatically become executable instructions for another without validation, policy checks, or human review.

By separating the generation of a recommendation from the execution of an action, companies can create guardrails that prevent model variability from causing operational errors.

4 | Accountability: Companies Need to Assign Internal Ownership of Agent Behavior

When an employee makes a mistake, managers can investigate by asking questions. When traditional software fails, engineers can check logs. But AI agents introduce a more complicated scenario, because their behavior often emerges from a chain of reasoning steps, retrieved documents, and tool calls that may not be easy to reconstruct after the fact.

That creates a serious accountability challenge. Imagine a procurement agent, for example, that summarizes supplier performance and posts the results in a company Slack channel. If the agent accidentally includes confidential contract terms—because it interpreted "share transparently" as permission to disclose them—leaders will need to know exactly how that decision was made. Which documents did the agent read? What instructions did it follow? Why did it believe it was allowed to share the information? Without that kind of evidence trail, organizations can't explain their systems' behavior to regulators, auditors, and customers.

A 2024 tribunal decision offers an early signal of how the law may treat this issue. In Moffatt v. Air Canada, a customer relied on incorrect information provided by an airline chatbot about bereavement-fare eligibility.

28

---

IN FOCUS Managing Teams in the Agentic Age

When the airline argued that the chatbot was effectively a separate entity, the tribunal rejected the claim and held the company responsible for the misinformation.

To avoid such problems, companies have to design their AI systems with accountability in mind from the outset. That begins with maintaining comprehensive records of how an agent operated, including which data sources it accessed, what prompts it received, and which tools it used to complete a task. Those records should enable reconstruction of the chain of reasoning that led to any action. Organizations should also assign clear internal ownership for the monitoring and governing of agent behavior so that responsibility does not become diffused. If a regulator, auditor, or customer asks why an AI system made a particular decision, the company should be able to provide a clear, evidence-based explanation. Without that level of transparency, large-scale automation will remain difficult to defend.

The Way Forward

If turnkey deployment is unrealistic for AI agents, the alternative is not to avoid them but to introduce them gradually, expanding their autonomy only as the organization develops the ability to govern them.

One useful way to think about that progression is as an “autonomy ladder” with different levels of execution authority: Does the system draft content, propose actions for approval, or execute actions within tightly defined limits? Organizations typically begin with agents that produce assistive

output—drafts, summaries, or recommendations that humans review before anything is sent or executed. The next level is retrieval with guardrails, where agents answer questions using internal information but rely on well-governed data sources. From there, companies may allow supervised actions, in which agents propose operational tasks—issuing refunds, updating records, or routing approvals—but a person always confirms the decision before execution. Only after those controls are proven should organizations consider bounded autonomy, where agents execute workflows independently within narrow limits and predefined thresholds.

Many effective deployments intentionally remain on the lower rungs of the autonomy ladder. The benefits consultancy OneDigital uses Azure OpenAI to accelerate consultant research, improving “time to insight” rather than replacing the consultants themselves. Other prominent deployments reach bounded autonomy by keeping the scope narrow. Klarna has reported that its AI assistant handles a large share of customer service chats autonomously while maintaining immediate escalation paths to human support for complex or sensitive cases.

For leaders evaluating the growing number of agent platforms, the most important investments are often organizational rather than technical. Companies should begin by defining a clear “turnkey boundary,” distinguishing between AI applications that can be deployed with minimal structural change and those that require significant redesign of controls and governance. They

should also treat permissions as a core design question by assigning each agent narrowly scoped access aligned with its role. Clear human-in-the-loop thresholds should determine when automated decisions require oversight, particularly in situations involving financial exposure, regulatory obligations, and reputational risk. Finally, leaders should measure outcomes instead of pilots, focusing on operational indicators such as cycle time, error rates, and compliance incidents rather than simply counting the number of AI experiments underway.

GENERATIVE AI WILL continue to improve, and vendors will package more safety features into their platforms, but the gap between an impressive demonstration and a trustworthy production system will persist unless organizations rethink what deployment actually requires. Enterprises are not turnkey environments; they are complex systems shaped by legacy technology, policies, and human judgment. The companies that succeed with agentic AI won’t simply install more agents—they’ll build the structures that allow those agents to be trusted. ▼

HBR Reprint H0949U

RAHUL TELANG is the Trustees Professor of Information Systems at Carnegie Mellon University’s Heinz College.

MUHAMMAD ZIA HYDARI is an assistant professor of business administration at the University of Pittsburgh’s School of Business. RAJA IQBAL is the founder of Ejento AI, a governance-first agentic AI platform, and an adjunct faculty member at the University of Pittsburgh’s School of Business.

29

---

To Thrive in the AI Era, Companies Need Agent Managers

by SURAJ SRINIVASAN and VIVIENNE WEI

ZACH STAUBER'S DAY begins before the first customer-support ticket even lands in the queue. As a support agent manager at Salesforce, a global company that provides businesses with customer relationship management platforms, Stauber manages a fleet of generative AI support agents across support, sales, and marketing on a platform the company calls Agentforce. Stauber describes his routine this way: "Data, data, data. I start and end my day in dashboards, scorecards, and agent observability monitoring." He focuses on how the AI

agents are working but also how they are learning and adapting—much as a traditional manager might walk the floor, check in with a struggling employee, or huddle with a team on a tricky case.

Salesforce offers a glimpse into what the future of digital work will look like across industries. The Agentforce platform, which Salesforce is selling to client companies and using itself, now resolves nearly 74% of the company's inbound customer-support cases. Dozens of digital "employees," all AI agents, are resolving customer issues, drafting personalized emails, and routing more-complex cases to human specialists. Each AI agent operates semiautonomously, learning from feedback,

collaborating with other AI agents, and escalating complex tasks it can't handle to human support agents.

Agent managers—people like Stauber, who runs an AI-powered human-plus-technology team—oversee this hybrid workforce. They orchestrate how AI agents learn, collaborate, perform, and work with human counterparts. And they supervise AI agents the way traditional managers coach and motivate human employees, though with a greater focus on the safety, accuracy, and business alignment of the agents' work.

As companies such as Salesforce, JPMorganChase, and Walmart operationalize autonomous AI across functions from customer service to finance, a profound shift in management is underway. The agents being deployed today aren't just tools to automate tasks; they're teammates that require training, governance, and performance management. With this shift, the agent manager is emerging as one of the most critical roles in the age of AI.

Just as product managers became indispensable during the software revolution, agent managers are fast becoming the connective tissue between strategic intent and autonomous execution. Their mission: to make AI agents smarter, faster, safer, and more impactful through orchestration across agents and with human counterparts.

The Hybrid Workforce for Sales Development

Salesforce's sales development representatives (SDRs) are an example of this shift. Previously, the company's agentic

Yannick Dary/Shin/Boosky

30

---

transformation and sales development team was responsible for following up on leads, typically handling dozens of contacts but speaking with only 12 to 15 prospects daily. This disparity meant that valuable leads were neglected due to human capacity limits.

Now an AI agent operates as a part of the hybrid team. The SDR AI agent takes over the initial high-scale, low-value interactions—personalized outreach, qualification, and continuous follow-up on stale leads—ensuring that the sales cycle never stops. “While my team is sleeping, our agents are already interacting with customers,” says Vanessa Tabbert, VP of the team.

The integration of these agents transformed the team’s capacity from booking 150 meetings in 30 days to scheduling more than 350 meetings in a single week after launch with the same lead volume. This resulted in generating $60 million in annualized pipeline and acquiring more than 300 new clients within four months.

Crucially, this AI-augmented capability allowed for a rapid geographical rollout managed by a “two-pizza team” (Amazon’s lingo for “a small team”). The expansion occurred across major markets in quick succession; the agent is now live in the U.S., Canada, the UK and Ireland, Africa, and Japan, with immediate expansion planned for Australia, New Zealand, South Asia, and more.

The evolution to agentic SDRs transformed the human seller’s role from one focused on low-connect prospecting to one centered on high-value human interaction, empathy, and creative problem-solving. Human SDRs can now focus on persuasion and judgment to

close deals. The agent manager’s job is to ensure that the autonomous-agent workforce is continually adapting and operating safely, and most critically, that the AI agent is aligned with the company’s sales priorities.

Redefining Ownership

Achieving sustained success with agentic AI requires an attitudinal shift championed by line-of-business (LOB) owners. This shift redefines AI agent ownership. In the pre-agentic world, AI deployment lived within IT departments or data science organizations. In the agentic era, business units need to take control. LOB owners should design, test, and govern the agents that power their workflows, much as they would for the human workforce.

At Salesforce, for instance, customer success teams define the AI agent’s tone, escalation rules, and success metrics, all under the stewardship of agent managers. If AI agents are performing real work for a business unit, that unit must own their performance. That requires establishing a clear philosophy that AI agents are complementary partners, not competitors, with human agents. For instance, in customer contact centers, the call center teams rather than the IT team manage the AI (and human) agents. The contact center teams, not the technology organization, are accountable for problem-solving and serving customers, whether the service is provided through AI agents or humans.

This approach creates both opportunity and complexity. To be successful, agent managers must blend deep functional expertise with operational AI

literacy. They must know not only what the business wants but how to teach AI agents to fulfill those needs safely, consistently, and transparently.

Defining This Pivotal Role

In our interviews with Salesforce’s internal Agentforce group, we found that agent managers typically operate at the crossroads of customer experience, AI operations, and product management. Their goal: to translate functional expertise into measurable AI performance.

Depending on organizational maturity, agent managers may report to one of the following:

• digital customer success (as at Salesforce)
• sales management (with tight alignment to a centralized AI operations team)
• a cross-functional digital transformation office

But across all settings, this role is not likely to be transient. It’s a durable operating function, akin to DevOps or site reliability engineering, born of a structural change in how work will be performed in a hybrid digital-human fashion.

Agent managers’ responsibilities combine business insight, analytical rigor, and hands-on interaction with AI systems, creating a new kind of operational leadership: AI orchestration. The role usually involves:

• monitoring agent performance: quality, speed, escalation, and customer sentiment
• refining prompts and workflows to improve accuracy and tone
• managing handoffs to human agents and ensuring escalation when required

31

---

IN FOCUS

Managing Teams in the Agentic Age

• conducting root-cause analyses of failed cases to drive continuous improvement

• quantifying impact through ROI analysis and executive reporting

What Makes an Effective Agent Manager

The best agent managers resemble early product managers or site reliability engineers. They excel at the intersection of human judgment and machine performance.

Their success depends on six critical capabilities:

1. AI operational literacy: They understand how agents operate, how prompts drive outcomes, and how system failures occur.

2. Functional depth: They possess deep knowledge of the business process that the agent supports, whether customer service, finance, or logistics.

3. Systems thinking: They visualize how agents interact across workflows, departments, and even other agents to achieve “multiagent orchestration.”

4. Change resilience: They adapt quickly to shifting models and business needs, refining agent logic in weekly “test-deploy-learn” cycles.

5. Prompt craftsmanship: They excel at designing and refining the language and logic that shape agent behavior (the equivalent of employee training for machines).

6. Work designed across machines and humans: They know how to create hybrid workflows, assessing the limits of machine capabilities and creating human escalation routines as needed, and how to motivate the human

workforce in the context of hybrid AI-human work.

Ultimately, an effective agent manager is fluent in the language of business strategy, the operational workings of AI, and people management.

Successful management in this hybrid era also requires a shift in how human performance is measured. The focus on activity-based KPIs (for example, making 60 calls a day) is obsolete. With the new model, KPIs focus on outcomes that depend on orchestration and influence—a recognition that performance now depends on how well people are tuning and running workflows with their agents. This means management should concentrate on maximizing the efficiency of the entire human-agent system.

Finally, the human workforce must rapidly acquire new, distinct skills to capitalize on this shift. The agent manager facilitates this transformation by enabling employees to move away from low-value, information-retention tasks. The new skill set includes managing AI (knowing how to command the agent effectively) and managing engagements (developing the acumen necessary to engage successfully in the high-value human interactions).

Hiring and Developing Agent Managers

While agent managers can come from a variety of disciplines, the most effective ones emerge from roles already accountable for service quality, customer outcomes, and operational judgment. These individuals have deep domain expertise and a lived understanding of

what “good” looks like in real customer interactions, capabilities more essential than formal AI credentials.

Zach Stauber’s trajectory illustrates the desirable background for this emerging role. Trained in audio production and shaped by years in service delivery and conversational design, including leading early chatbot teams, Stauber was selected for what he calls “earnest curiosity”: a willingness to experiment, learn quickly, and take ownership as AI reshaped the work.

Early deployments also illustrate how agent work should be structured. Agent managers used natural language, shaping intent, judgment, and tone by translating complex business logic into simple, adaptive instructions an AI agent can follow. They worked closely with AI engineers, who often sat within the IT department and focused on data parsing, system integrations, and the technical steps required when an agent takes action. Internal groups that deliberately paired them scaled faster and with greater trust.

Agent management is not necessarily a technical role. Organizations that have successfully developed the position treat it as an apprenticeship, immersing managers in live operations, failure reviews, and iterative test-deploy-learn cycles while clarifying decision rights and escalation paths early. Those that centralize agent management entirely within IT or index on AI credentials often see agent managers function technically but fail strategically. As AI agents take on execution, success will increasingly depend on the quality of the managerial judgment that guides them.

32

---

The Rapid Growth of a New Discipline

As embedded intelligence spreads across every function, from HR to finance to supply chain, the need for dedicated orchestration will only grow. Without having someone accountable for designing and governing agents, even the most sophisticated AI initiatives will stall.

Yet identifying and developing this new class of leaders won't be easy. They will need a blend of business insight, AI fluency, and ethical judgment. Companies will have to invest in new training pathways, integrating business process design, performance analytics, AI expertise, and AI governance into traditional management-development programs.

This moment therefore demands a call to action. Within 12 to 18 months, "agent manager" will most likely be a standard job title in AI-first enterprises, a career path for leaders who know how to scale impact through intelligent automation.

Our research reveals that technology alone doesn't create transformation—leadership does. The agent manager is a crucial part of that leadership, the bridge between corporate intent and autonomous execution, between human judgment and machine precision. ▼

HBR Reprint H092LZ

SURAJ SRINIVASAN is the Philip J. Stomberg Professor of Business Administration at Harvard Business School and a member of the board of Harvard Business Publishing. VIVIENNE WEI is the chief operating officer of the Unified Agentforce Platform at Salesforce.

Create an Onboarding Plan for AI Agents

by JOSEPH FULLER

MOST EXECUTIVES BELIEVE that the big challenge in adopting agentic AI is figuring out how to adapt to a new and important technology, but in fact it's about managing work.

Executives have been slow to recognize this in part because of AI's still-nascent capabilities. According to a recently published and widely cited paper by Anthropic's head economist, Peter McCrory, and his colleague, Maxim Massenkoff, a yawning gap exists between what AI can theoretically do and how it is actually deployed. Most estimates suggest that 94% of the tasks in computer- and math-related occupations, for example, are displaceable by generative AI, but, the authors write, Anthropic's current offerings cover only about a third of those tasks.

A much larger gap exists on the human side of the equation. Multiple

studies released around this year's World Economic Forum (including those from Deloitte and McKinsey) indicated that less than 10% of companies thought they were making substantial progress in designing effective human-machine interactions. To capture AI's near-term benefits and prepare for its expanding impact, organizations need to integrate it into existing HR processes and make its role clear to employees.

Below are six suggestions I've developed while helping leading companies of various industries to realize that objective.

Give Every AI Agent a Job Description

As agentic AI penetrates organizations, teams will consist of human and agentic AI "colleagues," and all of them will need job descriptions—a tried-and-true way of specifying a worker's responsibilities, decision rights, and role in relevant processes. Crafting a job description for AI agents will require their managers to be deliberate about allocating responsibilities across human and agentic colleagues. As you craft an AI agent's job description, ask: What is it responsible—and not responsible—for? (Like people, AI agents perform best

Yaroslav Danychenko/Stockov

33

---

IN FOCUS

when objectives are explicit.) What are the boundaries of its decision rights? What are its authorities? When must it seek input or approval from colleagues or superiors?

Design the AI Agent to Address Pain Points for Human Colleagues

Many observers have pointed out that agentic AI promises to improve the quality of knowledge workers' work lives. Just as automation helped eliminate work in industries that were, as the saying goes, "dirty, dark, and dangerous," such as manufacturing and mining, AI can help eliminate work that is "dull, dispiriting, and deterministic." When a company assigns tedious steps or the dreariest elements of a job to AI agents, employees have a reason to adopt AI and work around its limitations by grounding it in their day-to-day experience.

Evaluate Every AI Agent on a Regular Cycle

AI agents need measurable performance metrics for the actual process outcomes that arise from their actions. The metrics have to be explicit and allow the broad evaluation of performance, addressing not just accuracy and ease of use but also timeliness and reliability, so that human teammates are reassured that their algorithmic colleagues are being held to standards too. Moreover, such metrics will help improve training regimens by revealing areas for improvement. Much as performance reviews inform professional development plans for employees, agentic teammates should benefit

from a cycle in which feedback informs learning. Without metrics, managers can't distinguish between acceptable variation and real failure.

Give Every AI Agent a Human Supervisor

Observers have begun to tout the capacity of AI agents to "orchestrate" the activities of many discrete agents. That may speed up actions and make AI agents' work more precise, but who will oversee the orchestrators? The need for human oversight undoubtedly remains. Every generation of AI has demonstrated some propensity to hallucinate. Instances are likely to wane as AI improves, but the stakes will mount as AI expands into professions deemed to be significantly exposed. Organizations will also remain accountable for any results generated by AI, so a sentient decision-maker must be held responsible for how the AI agent is trained, how it integrates with processes, and how it interacts with other human and agentic teammates. Regulators, legislators, and the courts will insist on that.

Hire AI Agents as Interns and Make Them Earn Full-Time Status

Precocious interns can help you and your colleagues, but you need to provide them with clear training, guidance, and structure. New AI agents are analogous. They've been "trained" on the basic concepts of the task they're about to undertake, but they lack meaningful practical experience. They don't have contextual intelligence about your

company's culture, values, and strategy, or about the workings of specific processes or markets. You hire interns to let them gain some of that experience and to see how they perform. Treat AI agents the same way. Don't "hire" them for their training or their eventual accomplishments. Also, employ them on a full-time basis after they can perform their job descriptions within the performance parameters established. Only a proven agentic AI should be made a permanent part of an ongoing process and introduced to other parts of an organization.

Give Each AI Agent a Name

Do this not to humanize the AI agent but to make its role discussable. When people say, "This decision came from AI," it's easy for colleagues to forget that the AI agents are their teammates and their sense of personal responsibility for the process's outcome can evaporate. In addition, multiple AI agents may be involved. By breaking down and assigning responsibilities to agents as you would to human employees, and by giving each agent a clear identity (as in the well-known cases of Watson, Alexa, and Siri), their roles become easier for people to understand.

EMBEDDING AGENTIC AI in large organizations takes more than demonstrating a business case. It requires rethinking the way work is managed and developing a pathway to achieve the transitions required. The quickest way to accomplish that is to use mechanisms for managing work that are familiar to executives, managers, and employees alike. Extending those tools to incorporate the burgeoning numbers of virtual agentic workers eases this daunting transition. ▼ HBR Reprint H094S9

JOSEPH FULLER is a professor of management practice and a faculty cochair of Managing the Future of Work at Harvard Business School.

34 Harvard Business Review September–October 2026

---

When team members treat prompting as a collective act, interactions with AI will advance their thinking.

It's Hard to Use AI as a Team. These Three Practices Can Help.

by GABRIELE ROSANI, ELISA FARRI, DANIEL TRABUCCHI, and TOMMASO BUGANZA

ACCORDING TO A global survey of 500 executives conducted by the Capgemini Research Institute, the active use of AI in team meetings is anticipated to more than triple in the next three years. The executives surveyed said that they expect group use of AI to lead to better meetings that arrive at higher-quality outcomes. Leaders should not let this optimism obscure the challenges ahead, however. Our research suggests that integrating AI into team settings doesn't

happen naturally, and introducing it into meetings without laying the proper groundwork can reduce participation, fragment discussions, and shift ownership away from the team.

Fortunately, there is an approach that overcomes these pitfalls: We call it "human-AI team chemistry." Our research indicates three practices to help build this new capability:

1. Engage with AI as a team. Participants should introduce themselves and

involve AI in a collective dialogue so that it addresses the group, considering the various expertise at the table.

2. Leverage AI's role fluidity. AI should function just as a notetaker but as a team member, deliberately switching roles (such as stakeholder representative, challenger, customer, competitor, and so forth) to enrich the team discussion.

3. Maintain collective ownership of the interactions with AI. When team members treat prompting as a collective act, debate alternative directions, and pause to judge AI's output, interactions with AI will advance—rather than outsource—their thinking.

These recommendations emerged from a five-month experiment involving 60 managers from 12 companies across diverse industries. In each organization, a team of three or more managers—all with previous experience using generative AI—was tasked with designing a platform-based solution to address a strategic business challenge. The problems were of comparable scope and complexity, and to ensure consistency, all teams followed the same methodology, developed by two of us (Daniel and Tommaso), and used OpenAI's ChatGPT model. Each team met in person five times, for a total of 30 hours. To understand how the teams actually worked with AI and experienced the process, we observed team interactions with AI in real time, analyzed the complete chat transcripts from each session, and collected post-session surveys to capture participants' feedback. This allowed us to see not just what teams produced but how they collaborated and where team-AI collaboration thrived or faltered.

Jong Grewal/Getty Images

35

---

IN FOCUS

On the basis of this research, we believe that teams that engage in these practices will achieve higher-quality outcomes and reduce the risk of falling into common AI-related traps.

Session One: False Starts and Missed Potential

In the first session, the idea of integrating AI into teamwork intrigued all participants. But the initial excitement faded quickly, often within the first hour. Teams grew quieter, slipped into a more passive mode, and began to simply watch the screen as the AI generated responses. Survey data after the first session confirmed what we had observed: Teams reported limited perceived benefits, and overall engagement was low. This result was the opposite of what we had expected. Rather than amplifying collaboration, AI seemed, at least initially, to dampen it.

Defaulting to individual chat mode.

A closer review of the chat transcript revealed the roots of the issue. As one of us noted, “If I didn’t know this was a team chat, I would have assumed it was an individual interacting with an AI.” In a human-only team meeting, when a new colleague or consultant joins, everyone introduces themselves, explains their roles, and shares context so that the newcomer can contribute effectively. When interacting with AI, teams did not follow the same norms. As a result, the AI responded as though it were supporting only the person typing rather than engaging with the full group or accounting for its dynamics. Lacking awareness of the team’s context, and its different roles, expertise, and

viewpoints, AI defaulted to a narrow, individual-focused perspective.

Giving AI a static role. Most teams assigned AI a single, static role—often “researcher” or “subject-matter expert”—and left it at that for the entire session, interacting with it primarily in answer-seeking mode and treating it as a repository of expertise to query rather than a thinking partner. None experimented with shifting AI into more-abrasive roles, such as critic or skeptical stakeholder, which could have encouraged reflection, constructive debate, or challenge. Nor did they ask the AI to take different perspectives—a customer, a competitor, an end user—which might have revealed blind spots or hidden assumptions.

Interacting staccato-style. We also noticed that the teams’ input to AI was often brief and transactional: “Give me another example” or “This isn’t the right direction.” These quick, minimal requests reveal that the team members rushed through the task without articulating their goals or explaining their reasoning to AI. Furthermore, AI often jumped ahead by proposing unsolicited “next steps” or ready-made options, nudging the team toward simple one-click confirmation (“OK, option B”) and steering the conversation before collective alignment had occurred.

The same three issues appeared repeatedly across groups, indicating a systemic pattern rather than isolated missteps. The issue was neither the technology nor the participants’ individual skills but how the teams were interacting with AI.

This raised two questions: (1) How can we help teams become aware of the

challenges related to integrating AI into team settings? And (2) what practical guidance do they need to bring AI fully into their discussions?

Session Two: Building Team-AI Chemistry

To help teams avoid the pitfalls they encountered in the first session, we developed a three-element framework:

1. Engage with AI as a team.
2. Leverage AI’s role fluidity.
3. Maintain collective ownership.

To raise awareness of the importance of these elements, we asked each team to go back to the session-one chat transcripts and reflect on how they had worked with AI. To support that reflection, we introduced a checklist of questions to help the teams recognize the interaction patterns limiting their effectiveness.

Here are some of the self-evaluation questions:

• Did we introduce ourselves as a team and explain our respective roles and expertise?
• Did we assign more than one role to AI?
• Did we articulate our reasoning clearly, or did we fall back on short, minimal responses?

By working through these questions, participants reflected on how they interacted with AI as a team and identified the areas where they could strengthen the interaction in the next workshop. We then provided practical tips and ready-to-use prompts designed to help teams integrate AI more intentionally as an active participant in their conversations, not just a tool they query.

This proved helpful. After the second session, teams looked more upbeat. A detailed review of the chat logs revealed that most team members now introduced themselves in turn, and AI began factoring in the nuances of different roles and expertise rather than treating the group as a single

36

---

Team-AI chemistry needs to be deliberately nurtured and intentionally embedded in the way teams work.

individual. They started having a collective conversation with AI.

Teams also began using AI more flexibly, moving beyond standard fixed roles like notetaker, expert, or analyst. Depending on the stage of the discussion, AI was asked to act as a brainstorming partner, a challenger to test assumptions, a “prototyper” to create artifacts, and a storyteller to refine pitches. Teams realized that AI can instantly shift across stakeholders and viewpoints, becoming a multirole team member within the same meeting.

While the first two capabilities—interacting as a team and assigning multiple roles to AI—were absorbed relatively quickly, the third one, taking collective ownership, took longer to mature. In the second session, some teams still followed AI rather than steering it, sitting around the screen and passively reacting to its outputs. In later sessions, however, the dynamic had gradually shifted. Before submitting a prompt, teams paused to discuss how to frame the next iteration. They debated alternative directions, conducted judgment checks, and collectively challenged the AI’s outputs before engaging again. These pauses prevented the team from slipping into “spectator mode” and helped them remain firmly in the driver’s seat. Survey data of later sessions confirmed the benefits. Average engagement increased by 30%, and participants reported that AI was providing more-meaningful support to their discussions. Two-thirds noted that their group conversations, alignment, and collaboration had improved as a result, leading to higher-quality outputs. Three out of five participants noted that

collective judgment mitigated the typical pitfalls of using AI alone, such as too much trust or conformity.

Nurturing Team-AI Chemistry in Your Meetings

This kind of team-AI chemistry does not come naturally, and it rarely emerges on the first attempt. For most teams, it needs to be deliberately nurtured and intentionally embedded in the way they work. The risk is that the effect will fade if it is not reinforced. How to do it in practice? Here are three tips:

• Plan the meeting agenda with explicit AI slots. Identify the agenda points where AI participates and specify the role it should play: For example, plan a five-minute introduction round, where the team briefs AI with context, or a 15-minute “challenge slot” toward the end of the meeting, where AI plays the skeptic.

• Prepare a few prompts to summon AI in a designated role. For instance, “Take the perspective of...” and “How might stakeholder XYZ react to...?” Or include cues that ensure judgment pauses, such as “Wait for our decision before proceeding.”

• After the session, review the chat transcript. Check it against a list of questions to see how well the team-AI dynamics unfolded, and identify opportunities to improve next time. You can also use AI as a coach (upload the chat transcript and ask for AI’s evaluation against the checklist) or a sparring partner in a dialogue (discuss ways to strengthen team-AI interaction in future sessions).

Putting these tips into practice requires more than good intentions:

Teams often lack the authority to redesign meetings or change established rituals on their own. Leaders play a critical role, as they can decide and embed a new way of working. They can experiment by intentionally integrating AI into one team meeting. They should also set expectations that it may take a few iterations to climb the learning curve and master the new approach. Once the new practice is tested, leaders should continue to apply it so that their teams don’t revert to old habits.

TEAM-AI CHEMISTRY DOESN’T develop automatically. To foster it, high-performing teams should engage AI collectively, draw on its multiple roles, and maintain shared ownership of the interaction. When nurtured deliberately, these capabilities enhance team performance by yielding better alignment and coordination and ultimately elevating the quality of teamwork outcomes. ▼

HBR Reprint H09678

GABRIELE ROSANI is the director of content and research at Capgemini Invent’s Management Lab. ELISA FARRI is the vice president and head of Capgemini Invent’s Management Lab. Gabriele and Elisa are the coauthors of the HBR Guide to Generative AI for Managers (Harvard Business Review Press, 2025) and the HBR Guide to Generative AI for Teams (Harvard Business Review Press, 2026). DANIEL TRABUCCHI is an associate professor in the Department of Management, Economics, and Industrial Engineering at Politecnico di Milano. TOMMASO BUGANZA is a professor of leadership and innovation at Politecnico di Milano. Daniel and Tommaso are the coauthors of Platform Thinking (Business Expert Press, 2023) and The Digital Phoenix Effect (Platform Thinking Publishing, 2025).

Harvard Business Review

September–October 2026

37

---

The HBR Interview with Dan Schulman

“Losing Doesn’t Work for Me”

Dan Schulman wasn’t looking to get back in the game. After a successful nine-year stint as the CEO of PayPal, he was ready to retire. But the board of Verizon repeatedly urged him to step in to try to reverse the slide at the telecommunications giant, and late last year he agreed. As Verizon’s CEO, his task is to turn around a company that has seen steep declines in market share, share price, and customer satisfaction. In a conversation with HBR editor at large Adi Ignatius, Schulman discusses how he’s trying to shake up Verizon’s culture, the risk and promise of AI, and what it takes to be an exceptional leader. Here are excerpts.

38

PHOTOGRAPHER GUERIN BLASK

---

39

---

**The HBR

Interview with Dan Schulman**

HBR: You originally rejected the offer to run Verizon. What made you get back in the saddle?

DAN SCHULMAN: I was happily retired, living on our ranch in Montana, and I wasn’t looking to come back into corporate life. But I was lead director on Verizon’s board, and we needed to make a change. I felt there was a real opportunity to take an iconic American company and turn it around. So after a lot of discussions with my wife, I reluctantly agreed to do it. And here I am, nine months into the job, and I’m loving it.

What had gone wrong at Verizon?

We had been losing market share for five years. We went from having the largest market cap in the industry to having the lowest. We had the lowest P/E multiple, which is the market’s way of saying that it doesn’t believe in your growth going forward. Our churn rates were going up. Our customer satisfaction results were going down. And inside the company, we were willing to be prey, gently ceding market share to our competitors. We weren’t playing to win.

That’s a damning list. Was the root of the problem technology? Strategy? Your relationship with customers?

It was probably more cultural than anything else. For a long time Verizon relied

on its engineering prowess. That’s no longer enough. We put the network—not the customer—at the center of everything we did. The company had a lot to be proud of, but it was resistant to change. We were risk averse. We were hierarchical. We were not imagining the kind of role we could play in the future, in the age of AI. When the pace of change external to a company is faster than the change of pace internally, you’re falling behind.

Companies talk all the time about being customer obsessed. Verizon is a utility. People want it to work, but they don’t necessarily want a relationship with it. What does customer obsession mean in your context?

It’s not easy to put the customer front and center. If it were, every company would do it, and it would bring no competitive advantage. We don’t actually use the term “customer centricity,” but we have an initiative called “Every customer has a name.” It means we treat you as a person, not as an account. So, when you come into our store, how quickly do we serve you? Do you understand the plans we have? When you call our customer service centers, do we treat you with respect and empathy?

Customers fire companies all the time in quiet ways because they feel disrespected. We want to create a company that respects you as an individual and also helps you understand the direction of technology. We’re moving into an AI world, and Verizon can be a part of the new infrastructure. Our service can help you use AI, create agents in a safe way, and so on.

I’ve heard criticism that the complexity of Verizon’s offerings pushes people almost unwittingly into a more expensive tier. Are you willing to accept lower revenue temporarily to simplify your offers and build long-term loyalty?

I don’t think that being customer-centric means being less than fiscally responsible to your shareholders. In fact, it can be the opposite. Our churn rate was going up because people were unhappy with us. We were raising prices. Our plans were confusing. When I came in, I said, “We’re not going to raise prices without offering real value, and we’re going to invest in the customer experience and grow our number of customers.” We’ve started to see our churn come down quite nicely.

40 Harvard Business Review September–October 2026

---

“You can satisfy customers and inspire your employees... at the same time. It’s not an impossible equation.”

If you do the right things, make the difficult decisions, and challenge the culture, you can satisfy customers and inspire your employees and shareholders at the same time. It’s not an impossible equation.

It’s never easy to come in as an outsider. You had a great run at PayPal, and you’d been on Verizon’s board. But people are resistant to change. What’s worked and what hasn’t with your employees?

You’re right: People are resistant to change. But change is a constant, and its pace is accelerating. Leaders need to tell it like it is, without sugar-coating. Every employee is an adult. They probably know what’s going on, and they’re waiting for leadership to admit it. At employee meetings, I don’t have slides or notes. I just talk from the heart. I tell them that we’re losing. And that’s hard to hear. But I tell them that losing doesn’t work for me, and I know it doesn’t work for them, either. We need a full-scale internal revolution and cultural shift, where we play to win, where we take risks—even if sometimes they fail. This isn’t a marketing slogan. It’s a way of life inside the company.

You recently laid off 13,000 people. How do you distinguish between cost cutting and a true transformation?

We needed to create a war chest to invest in the customer, and that necessitated the layoffs. I spoke to the company

AUGUST

41

---

“Being authentic is a superpower. You want to be approachable, to never think or talk in corporate-speak.”

about how painful laying off even a single person can be, but our job now is to make sure that we invest most of those dollars back in the customer so that we can reclaim market leadership and become one of the most trusted companies in our industry.

Are employees responding? What’s actually motivating and inspiring them? Surely it’s not “to save Dan’s job.”

I think they’re motivated by the idea that we’re playing to win, to be trusted, to be at the center of our customers’ lives. It’s important to have early successes for people to rally around. We’re pushing people hard. I want to see improvement every single day, and there’s an intensity about that. Working hard, being disruptive in the market, putting customers at the center, being on a winning team—I love all of that. Some people don’t, and that’s OK. It means we are not the company for them. We’re getting people who are inspired to be here, who want to make us better. They’re all in.

You have a reputation for being a truth teller, especially about the risks and opportunities of AI. What’s the truth about AI in business?

Truth is subjective, and the future is uncertain. I just try to be authentic and honest about how I see things playing out. AI is going to be part of our future, and every single one of us needs to know how to use AI tools. You will be better at your job and better in life if

you understand how to capture the benefits of AI.

We’ve put in place full weeklong training programs on using AI for all employees, and we have extensive reskilling programs for employees who might be displaced by AI. Successive waves of technology are going to radically change the way we work and the way we live. And I want to have that conversation as honestly as I possibly can.

What’s a plausible optimistic scenario for how AI plays out?

The value that we offer to consumers is going to be radically reshaped by AI. I don’t think we’ll just be a connectivity company, offering wireless and broadband services. We’ll be able to do things that enable consumers and small businesses to take full advantage of AI.

We can direct them to the models that are best for certain tasks, for example. People could buy tokens from us—I think tokens [units of data that an AI model generates] will become the coin of the realm in the future—and we could cost-optimize how they’re used. We could provide a secure environment if you want to utilize an AI agent. We’ll be part of the AI infrastructure, connecting data centers. And we’re creating edge data centers that can reduce latency for applications like remote surgery or autonomous-driving vehicles. We’ll be able to see how our service is doing building by building, floor by floor, and autonomously resolve issues without having to dispatch a tech worker to figure out what’s going on.

All this will help us increase customer satisfaction dramatically.

What’s the plausible darker AI scenario, and how do leaders talk about that?

Three years ago, when the new AI models came out, they couldn’t even do simple math. Today they’re solving problems that the very best mathematicians hadn’t been able to solve in decades. So we know there is going to be disruption, but we don’t know exactly what it’s going to look like. If you look at a functional area like customer service, where much of the work is routinized, AI agents will be able to solve many questions. In other areas, humans and machines will work together to provide solutions to the customer. The question is, How do we get ready for the change? When we did the initial layoff, we put in place a $20 million fund for AI reskilling and retraining. That’s not going to be enough. We’re going to double or maybe triple that amount over the next two years to make sure that our employees and the communities we serve are prepared for the future. As leaders, we have a responsibility to get people ready for this new age. We need to be prepared for the disruption. If there’s going to be a two- to five-year trough where there’s potentially more unemployment, that’s not good for our economy, for our country, or for our companies.

A lot of people aren’t enthusiastic about AI—not because they’re afraid of it but because they’ve used it in their

42 Harvard Business Review September–October 2026

---

The HBR Interview with Dan Schulman

work and have seen it create slop. How do you deal with that perception?

I get it. All I can say is that the models we're using now are the worst models we'll ever use. Every couple of months, they get much better. It's important that people don't just accept their visceral negative reaction to AI. I want them to be as open-minded as possible about how the future might unfold. There's a great line from Shakespeare's Hamlet: "There is nothing either good or bad, but thinking makes it so." You create your reality with how you think about things, so try not to overreact one way or another to a future that is changing rapidly.

You had years of experience as a CEO before your Verizon role. What have you learned about what it takes to lead effectively?

First, you need to be humble. You don't know it all. And if you do think you're the smartest person in the room, then you're in the wrong room. We need to admit when we're wrong and course correct, and we need to challenge ourselves continually. I never want to stand still because when you do, you start falling way behind. I also think being authentic is a superpower. You want to be approachable, to never think or talk in corporate-speak.

That can't be enough. There are plenty of humble, authentic people out there. What are the other skills and attributes that leaders need to be successful?

There are the basics: You've got to be able to work at a very intense level. You have to have extremely high expectations and insist upon them. You need

to be able to make difficult decisions and stand up for them. People need to know that you have an expectation that you're going to win in your market. I'm a fighter, a real fighter. I've done mixed martial arts for 40 years.

Thanks for the warning.

Yeah, you're welcome. I've learned a ton from that. I've learned to be cool, calm, and collected in high-stress situations. As a leader, you can never be too up or too down, because the good things are going to be followed by difficult things—and you need to learn from them both.

At PayPal, you talked about leading with a sense of purpose and mission. For example, when you realized some of your employees weren't making a living wage, you increased their pay. Do CEOs still do things like that? Or are we in a different place now, back to an era of shareholder primacy?

I don't think we're in a different place. Shareholders were important in that era too. But great CEOs have a set of values that are important to them. Today technology is front and center. People are worried about AI. Putting millions into training people to use AI is probably as important as anything we were doing in the past. I talk with other CEOs, and they're also thinking about how we can help make sure people land on their feet if there is job displacement, to create as strong an economy as we can. If we don't, people will be upset with the system, and that's a problem for democracy. I have a sense of purpose in doing this job. It revolves around my employees, and I have to make difficult

decisions. I need to keep customers front and center. I have my shareholders, my regulators, the communities we serve. To be successful, there are a lot of constituencies that I need to do the right thing for.

How will you know whether and when this turnaround has been a success?

We're in a three-year plan now. The first year is about confronting the issues that we have head-on and putting into place a fully comprehensive transformation program. We have 10 work streams and 400 initiatives—everyone in the company is engaged in those. Year two will be institutionalizing that work: Let's see the results, course correct, solidify. And then in year three, we'll turn into an AI-native company. We'll have transformed who we are. In years four, five, and six we're going to go into quantum computing and robotics.

What would have to be true to say Dan Schulman fundamentally changed Verizon?

I think you'll see my impact in our metrics. You'll see it in our customer satisfaction results. You'll see it in our churn rate. You'll see it in our employee satisfaction scores. You'll see that we are truly AI native. The most successful companies will have figured out culturally how to adapt to this changing environment. The utilization of AI, and what that's done for all the constituencies that I serve, is the legacy that I hope to leave. ▼ HBR Reprint R2605A

ADI IGNATIUS is the editor at large at Harvard Business Review and its former editor in chief.

43

---

STRATEGY

Al Is Revolutionizing Strategic Decision-Making

Felipe A. Csaszar Professor, Ross School of Business

ILLUSTRATOR DIMITRIS LADOPOULOS

44

---

45

---

STRATEGY

ABOUT THE ART

Dimitris Ladopoulos combines art, math, and algorithms to create a study of forms and variations.

Think about your last strategy offsite.

Your team spent weeks preparing. You flew people in, booked the conference room, hired a facilitator. And after two days of debate, how many truly different strategic options did you walk out with? Three? Four? Now ask yourself: Was that because only three or four good options existed, or because that's all your team had the time and mental bandwidth to develop and evaluate?

For most companies, the honest answer is the latter. The bottleneck in strategic decision-making has never been a shortage of possible directions. It has been the limited capacity of the human minds doing the work. We can hold only so much information in our heads, evaluate only so many alternatives in a strategic-planning cycle, and process only so many perspectives in a meeting before fatigue, politics, or the clock forces a decision. Scholars call this dynamic bounded rationality—the idea that human decision-makers, however capable, are constrained by finite attention, memory, and processing power.

These constraints are so fundamental that we rarely notice them, but they have quietly shaped every tool in the standard strategy playbook. The reason a SWOT analysis has four quadrants, a growth share matrix is a 2×2, and Michael Porter's most famous framework has exactly five forces is not that the competitive world is actually so simple. It's that the frameworks had to be simple enough for a human team to map them out on a whiteboard in a few hours.

For decades, those frameworks were the best we had. That's no longer the case.

The current generation of artificial intelligence tools—particularly large language models (LLMs) and the multi-agent systems being built on top of them—are not just additions to the planning tool kit. They're technologies that directly relax the cognitive constraints that have shaped how companies make their most important decisions. AI can generate and screen thousands of strategic alternatives where a human team might be able to consider a mere handful. It can build and continuously update models of markets, customers, and competitors that are far richer and more dynamic than any static framework. And it can

IDEA IN BRIEF

THE PROBLEM

Strategic decision-making has long been constrained by bounded rationality. Leaders can generate, evaluate, and debate only a limited number of strategic options because humans have finite time, attention spans, and cognitive capacity.

THE SOLUTION

Generative AI can expand the bounds of strategic thinking by generating and evaluating thousands of options, replacing static frameworks with dynamic, data-rich models and enabling more-structured deliberation.

THE PAYOFF

Firms that integrate gen AI into strategy processes won't just move faster—they'll make higher-quality decisions, uncover hidden opportunities, and build a durable competitive advantage using proprietary data, processes, and speed.

46

---

47

---

STRATEGY

test ideas through simulated deliberation—synthesizing diverse perspectives and challenging assumptions without the groupthink, hierarchy, and time pressure that distort real-world strategy meetings.

In short, AI offers a path to unbounding the strategy process. The limits are not removed, of course, but they're pushed outward—and where they end up matters enormously, because the implications go well beyond efficiency. When you can explore more options, you find better paths forward. When you can model your environment in higher resolution, you see more opportunities and threats. When you can stress-test a plan by simulating competitors, skeptical customers, and a devil's advocate who never gets tired, you make decisions that are more resilient. The companies that master this new way of working won't just do strategy faster. They'll do it better—and build the next generation of competitive advantage.

I study strategic decision-making and have spent considerable time examining how AI performs on core strategy tasks. In recent experiments my colleagues and I found that business plans generated by an LLM were rated more favorably by experienced investors than were plans written by entrepreneurs in a startup accelerator and business plan competition. Perhaps more striking, when the business plans were evaluated, AI's assessments were more aligned with the investor panel's average judgment than individual investors' assessments were—which suggests that AI can both generate strategic options at scale and evaluate them with a consistency exceeding that of human experts. Such findings don't mean AI is ready to replace your strategy team. But they do suggest that the cognitive work at the heart of strategy is no longer exclusively human territory.

In this article I'll lay out what this shift means in practice. First, I'll describe three ways AI can transform the strategy process, with examples from companies already using it to do so. Then I'll address the skeptic's natural question—If every company has access to the same AI, how can it be a source of advantage?—and explain how firms can build durable competitive moats around strategy capabilities augmented by AI. Finally, I'll offer a practical playbook for leaders who want to start redesigning how their organizations make strategic decisions.

THREE WAYS AI EXPANDS STRATEGIC THINKING

To understand what changes with AI, it helps to think about what your organization has to do when making strategy. At its core, any strategic decision involves three cognitive tasks: searching for possible courses of action, representing the environment in which those actions will play out, and aggregating the judgments of the people involved in the decision. Let's look at each in turn.

From a handful of options to thousands. The most immediate constraint on any strategy process is in the search for options: How many alternatives can your team generate and evaluate? For most organizations the answer is surprisingly low. A typical strategic-planning cycle might surface a dozen ideas and then narrow the field to three or four, with team members spending most of their energy debating those finalists. The vast majority of the possibility space is never explored—not because it lacks promise but because human teams lack the time and capacity to take on that job.

As my research suggests, the breakthrough with AI is that the cost of generating and screening options falls sharply, helping organizations explore far more of the possibility space. The shift is already visible in M&A. In one case, documented by McKinsey, a software company used a generative AI scouting system that combined semantic search with a database of more than 40 million public and private companies. Drawing on patents, filings, expert transcripts, and other structured and unstructured data, the system surfaced and scored more than 500 acquisition targets in less than a day, narrowed the list to 15 serious leads, and supported three completed acquisitions within months.

The point here is not just speed. It is a qualitative shift in what "looking for options" means. Instead of asking a team to brainstorm plausible targets, a firm can now scan nearly the entirety of the relevant universe, using multiple strategic lenses simultaneously, and create a short list that no earlier process would have produced. The implication for leaders is straightforward: Before any major strategic choice—an acquisition, a new market entry, a product portfolio decision—you should task an AI system with generating a far larger set of alternatives than your team would normally consider, filtered by your specific criteria. Use your team's judgment on the short list, not the long list. AI expands the search; humans choose.

From static frameworks to living models. The second constraint is representation: How does your organization model the environment it operates in? For most companies the answer is some combination of frameworks, spreadsheets, and quarterly reports—tools that are useful but inherently static and simplified. They capture the most important dynamics but necessarily leave out enormous amounts of nuance, context, and real-time change.

48

---

The bottleneck in strategic decision-making has never been a shortage of possible directions. It has been the limited capacity of the human minds doing the work.

AI makes it possible to work with far richer and more dynamic representations. MYbank, the digital lender within Ant Group, offers a striking example. Traditional banks evaluate small businesses' creditworthiness using just a handful of variables—revenue, collateral, credit history. Many small businesses lack one or more of these, leaving millions effectively "unscorable" and therefore unfundable. MYbank replaced that basic model with an AI system that draws on more than 3,000 variables, integrating transaction data, supply chain relationships, business network data, and even satellite imagery. The result is what the company calls its 3-1-0 model: three minutes to apply, one second for approval, zero human intervention. MYbank has extended credit to more than 53 million small businesses—72% of which were first-time borrowers—with a default rate around 1%.

The strategic insight here is about what happens when you replace a low-resolution model of your environment with a high-resolution one. Using its AI system, MYbank didn't just serve existing customers faster. It became able to see an entirely new customer segment—tens of millions of viable borrowers—that the old model had rendered invisible.

The same logic applies to any company whose models of demand, supply, or competitive dynamics were built for an earlier era. Consider Unilever's ice-cream business. Rather than relying on lagging quarterly forecasts, the company created an AI-driven model that integrates weather probabilities, demand signals, and telemetry from roughly 3 million connected freezers across 35 factories. The result is a continuously refreshed picture of where demand is forming and where supply may fall short—a living representation of the market that lets managers intervene in real time rather than react after the fact. Unilever reports a 10% improvement in forecast accuracy in Sweden and sales increases of up to 30% in some regions from insights generated by AI-enabled freezers.

For leaders, the practical step is this: Identify the one or two strategic models your organization uses most heavily (your market map, your customer segmentation, your competitive positioning) and ask whether AI could make them richer, more current, and more granular. In most cases the answer will be yes. And the payoff won't just be better information. It will be the ability to see opportunities and risks that your current models are incapable of revealing.

From groupthink to structured challenge. The third constraint is aggregation: How does your organization combine the knowledge and judgment of multiple people into a decision? In theory, diverse teams should produce better decisions than any individual could alone. In practice, strategy meetings are shaped by hierarchy, politics, personality, and time limits. The most senior person's view carries disproportionate weight. Dissent can be risky. The group converges too quickly.

AI offers a different approach. Instead of depending on a roomful of people with uneven incentives to speak up, organizations can now orchestrate what might be called synthetic deliberation: AI-driven processes that pressure-test ideas without the social frictions of group dynamics.

The mechanism can be simple or sophisticated. McKinsey, for example, has described multiagent workflows in which a "creator" agent drafts an analysis and a "critic" agent systematically challenges it—a process that automates the rigorous back-and-forth that strong teams try to create manually. At the more ambitious end, the BCG Henderson Institute has described LLM-powered strategic war games in which agents represent competitors, regulators, customers, and other stakeholders. Unlike in traditional rule-based simulations, these agents can generate less predictable, more human-like responses, which forces management teams to confront scenarios they might never have imagined on their own.

The most rigorous evidence for AI's ability to improve collaborative ideation comes from a field experiment at Procter & Gamble. In a study involving 776 commercial and R&D professionals working on real innovation tasks, researchers found that individuals using AI matched the average quality of two-person teams working without it. Teams using AI were about 12% faster and were more likely to produce top-decile solutions. Perhaps most important, AI reduced the silo effect between commercial and R&D professionals, giving each function access to the other's perspective and helping both produce more-balanced ideas.

The lesson for leaders is to build structured AI challenges into your decision-making process. Assign AI agents to distinct roles: one to generate the strongest case for a strategic direction, one to argue against it, and one to simulate likely competitor responses. You can also create synthetic panels (customers, regional managers, regulators) to investigate how a decision might land with different stakeholders. The

49

---

goal is to ensure that by the time your leadership team sits down to decide, the obvious weaknesses have been identified and the plan has been pressure-tested against a wider range of perspectives than any single meeting could accommodate.

None of this means AI outputs should be taken at face value. LLMs can produce confident-sounding analyses that are subtly wrong, internally inconsistent, or built on fabricated evidence. Synthetic deliberation can surface useful challenges—but it can also generate plausible-seeming objections that miss the point entirely. The quality of AI-augmented strategy depends on the quality of the data fed into the process and, critically, the human judgment applied to its outputs. That isn't a reason to avoid these tools, but it is a reason to design the process carefully, and it's precisely why the role of the human strategist becomes more important, not less, in an AI-augmented world.

BUILDING ADVANTAGE WHEN EVERYONE HAS AI

If you've read this far, you may be thinking: *But if my competitors can buy the same AI models that I have, how does any

of this become a source of competitive advantage?* It's a fair question, and it has a clear answer if you look at it through the right lens.

The most useful analogy is the internet in 1995. By the mid-1990s, internet access was becoming widely available. The underlying technology was not proprietary. Yet the companies that built the most durable advantages of the next two decades—Amazon, Netflix, Google—did not win because they had a website. They won because they recognized that the internet wasn't something to just bolt onto existing operations; it was a new foundation on which to build business models, capabilities, and data assets that competitors could not easily replicate. The firms that treated the internet as a utility saw their offerings quickly become commoditized. The firms that treated it as a platform for reinvention created moats that lasted for decades.

AI is at a similar inflection point. The models themselves are increasingly available. The advantage lies in what you build on top of them. Three approaches stand out.

Make AI smarter with your data. General-purpose AI models are trained on public information. They are, by

50 Harvard Business Review September–October 2026

---

STRATEGY

definition, generic. The moment you fine-tune or ground a model with proprietary data—your customer histories, transaction patterns, or internal research—you create something that competitors cannot replicate simply by buying the same software. Morgan Stanley’s AI assistant, which gives the company’s 16,000 financial advisers access to more than 100,000 internal research documents, reached 98% adoption and raised document-retrieval efficiency from 20% to 80%. Stripe’s fraud detection system, Radar, is another example at network scale. Trained on data from millions of businesses processing more than $1.4 trillion in annual payments, there is a 92% chance that it has already seen any given card on its network. A smaller competitor using the same algorithms but lacking that data cannot match its accuracy.

Deploy AI in processes that only you have. Data is one moat. Process is another. When a company embeds AI into a proprietary workflow—one that reflects its specific strategic logic, organizational knowledge, and competitive context—it creates a capability that’s difficult to observe from the outside and even harder to copy.

John Deere’s See & Spray system illustrates this concept well. The underlying technologies—computer vision and machine learning—are widely available. But Deere embedded them into precision-agriculture equipment that scans roughly 2,100 square feet per second, makes more than 5,000 spraying decisions per minute, and reduces herbicide use by 50% to 77%. The moat is not the algorithm. It’s the integration of AI into hardware, field-level data, and a dealer network far more expansive than that of rivals. What makes this approach strategic, not merely operational, is that See & Spray changes Deere’s competitive positioning: It transforms the company from an equipment manufacturer into a precision-agriculture platform, deepening customer lock-in and opening new revenue streams around data-driven farm management.

Move to the new frontier faster than your rivals.

There is a well-known idea in strategy that operational effectiveness—doing the same things as other companies but better—is not the same as strategy. In stable times, that distinction holds. But in periods of technological discontinuity, the gap between firms that have reached the new performance frontier and firms that have not can be enormous.

Duolingo exemplifies the point. It launched AI-powered features—Roleplay, Explain My Answer, Video Call—soon after OpenAI’s GPT-4 became publicly available. But speed alone is not its moat. Duolingo’s real advantage is Birdbrain, a proprietary learning model trained on behavioral data from more than 500 million users completing roughly 1.25 billion exercises daily. The general-purpose model is the commodity. The learning system built around it, refined over years and fed by behavioral data whose scale dwarfs that of competitors, is the moat.

AI can compress months of analysis into days, reveal opportunities that were invisible, and stress-test decisions that would otherwise go unchallenged. Firms that adopt those capabilities operate at a fundamentally different level of strategic quality. The advantage may be temporary in theory, but in practice it just keeps compounding: Better decisions lead to better positioning, which generates better data, which improves future decisions. The frontier will keep moving, so enduring advantage comes from the ability to reach it again and again, gaining strength each time.

51

---

52 Harvard Business Review September–October 2026

---

Identify the strategic models your organization uses and ask whether AI could make them richer, more current, and more granular. In most cases the answer will be yes.

The bottom line is this: As a technology, AI is becoming ubiquitous. But as a capability—embedded in your data, your processes, and your speed of execution—it is not. The companies that treat AI as a commodity to be purchased will find that it raises the bar for everyone without distinguishing them. The companies that treat it as a platform for building proprietary strategic capabilities will find that it is one of the most powerful sources of competitive advantage available today.

A LEADER'S PLAYBOOK FOR THE AI ERA

If you want to redesign how your organization makes its most important decisions, the playbook is straightforward.

Widen the option set before narrowing it. For every major strategic decision—an acquisition, a market entry, a product portfolio shift—use AI to generate a long list before the leadership team debates a short list. Most organizations narrow too early because exploring lots of options used to be expensive. It no longer is.

Replace static snapshots with living models. Pick the strategic representation your organization relies on most heavily—your customer segmentation, your competitive map, your demand forecast—and upgrade it from a periodic document to a dynamic model that updates as conditions change. The goal is to create a map rich enough to reveal what your current framework structurally cannot.

Institutionalize structured challenge. Good strategy depends not just on generating ideas but also on rigorously testing them. In most organizations that testing is still shaped by hierarchy, diplomacy, and the human tendency to converge too quickly. Don't leave dissent to chance. Design it into the process.

Build creator-critic-competitor workflows into every major commitment: Assign one AI agent to make the strongest case for a plan, one to dismantle it, and one to simulate how rivals or regulators would respond. Make structured challenge as routine as financial modeling.

Redefine the strategist's role from analyst to architect. In a world of tight cognitive bounds, strategists spend most of their time gathering data, building spreadsheets, and populating simplified frameworks. As AI starts to unbound cognitive constraints, that baseline cognitive work

can increasingly be automated—and the key resource shifts from analysis to imagination, from populating frameworks to designing the process itself.

All of those shifts demand a new talent profile: the hybrid strategist. Leaders who fit this profile know how to frame a strategic question, can design an AI workflow to answer it, and, above all, understand where the machine must yield to human judgment.

To become a hybrid strategist, you'll need to change what you demand in the room where decisions are made. Before any major strategic proposal reaches the leadership table, require three things: the AI-expanded list of alternatives that were considered and rejected, a current model of the competitive environment, and the results of a structured critique. When the executive committee raises the standard for what a credible recommendation looks like, the rest of the organization will follow.

Finally, make AI use normal. Morgan Stanley's AI assistant reached 98% adoption among 16,000 financial advisers because it was woven into the proprietary research workflows that they already depended on. The tool made their daily work clearly better, and adoption took care of itself. Start with the same logic: Find the decision process where your people already feel the pain, rebuild it with AI, and make the new workflow the default. Culture usually follows usefulness.

PICTURE YOUR NEXT strategy offsite. But this time, imagine it unbounded.

Your team arrives having already reviewed an AI-generated scan of hundreds of possibilities, filtered to the most promising. The market model on the wall is current, not six months stale. Before anyone presents a recommendation, it has survived a structured critique—devil's advocate, simulated competitors, skeptical customers. The conversation in the room is sharper, the options are stronger, and the time that used to go to basic analysis now goes to the questions only humans can answer: What do we believe? What are we willing to risk? What kind of company do we want to become? ▼

HBR Reprint R2605B

FELIPE A. CSASZAR is the Alexander M. Nick Professor and the chair of the strategy area at the University of Michigan's Ross School of Business.

53

---

Transformation Should Be a Learning Journey

Rigid change programs often fail. A more experimental approach can build momentum—and increase the odds of success.

AUTHORS

Evgeny Kaganer Professor, IESE Business School

Christoph Loch Professor, IESE Business School

PHOTOGRAPHER ERIN DERBY

54

---

بِسْمِ اللَّهِ الرَّحْمَنِ الرَّحِيمِ

---

CHANGE MANAGEMENT

ABOUT THE ART

Erin Derby builds digital floral collages, working with familiar materials to create something unexpected.

A

LL COMPANIES recog- nize the need to adapt as technology, society, and geopolitics collide to produce multiple

competing futures. But attempts at systemic change often fail. Typically, a CEO sets a bold, monolithic transforma- tion goal and then launches multiple initiatives to achieve it. The trouble is, that goal is constantly destabilized by shifts in the environment. As a result, early projects fail to deliver expected results, new initiatives proliferate, and the transformation journey becomes increasingly fragmented, leading to employee fatigue and shareholder frustration.

The root cause of these failures, we believe, is the senior leadership team's insistence on managing transformation as a large, complex project with a detailed road map to a fixed goal and results-oriented targets. While a disciplined, planned approach is understandable given the coordination complexity and budget demands of driving change in a large corporation, it's ultimately unfeasible. The gaps between the knowledge and capabilities of the firm's current business and those of its future business, combined with inevitable environmental turbulence over a multiyear project, hamper the organization's chances of success.

A better approach is to manage transformation as a learning journey. By that, we mean a process with an incom- pletely defined goal that evolves as the organization makes decisions, learns from outcomes, and adjusts accordingly. At the outset the senior leaders need to explicitly acknowledge that while they have some idea of where the company ought to be, a final outcome can't yet be fully known. With no fixed end point, the road map similarly is not fixed. The leader- ship team must proceed by selecting a coordinated portfolio of projects that will generate the greatest momentum and

56

---

IDEA IN BRIEF

THE PROBLEM

Many leaders manage corporate transformations like large-scale engineering projects: They set a fixed destination, create detailed road maps, and measure success through short-term financial outcomes. But most of those efforts fail.

THE REASON

Executives treat transformation as an exercise in execution, pushing rigid plans forward even when evidence suggests the strategy should evolve. This discourages learning, suppresses bottom-up insights, and creates resistance.

THE SOLUTION

Treat transformation as a learning journey. Instead of pursuing a static end state, aim for continuous learning and pivot as new information emerges. Manage a coordinated portfolio of pilots, capability-building initiatives, improvements to existing businesses, and carefully timed new ventures.

57

---

CHANGE MANAGEMENT

learning. Insights from the first wave of projects will then shape the priorities for the second wave, and so on, dynamically advancing the organization toward an evolving goal.

In the following pages we draw on our research and experience as consultants to more than 10 transformations at large companies in the United States, Europe, and Asia to offer a learning-oriented playbook for corporate systemic change. We’ll illustrate our recommendations by comparing GE’s more traditionally managed and ultimately unsuccessful transformation with that of DBS Bank, which unfolded as a cumulative learning journey that continues today. We’ll conclude by describing the key actions that leaders must take to keep their company on a coherent and well-informed trajectory.

Charting the Course

A clear sense of a desired destination is the starting point of every journey. That’s why all transformations need a vision. The vision must be ambitious enough to inspire employees to work on the transformation while remaining concrete enough to make employees and stakeholders believe it is attainable.

Setting a vision requires accepting that it will evolve as the organization learns and redirects its energies to reflect its learning. The only permanent characteristics of the vision are that it must inspire a collective endeavor focused on creating as well as capturing value and that it must unite employees from all parts of the company.

Let’s take a look at DBS Bank. In 2010, in the middle of the postcrisis downturn and just after its first-ever layoffs, DBS ranked last in customer satisfaction in Singapore—people joked that DBS stood for “Damn Bloody Slow.” The bank’s new CEO at the time, Piyush Gupta, saw the crisis as a chance to leapfrog global peers. His vision, to position DBS as “the new bank for the new Asia,” acknowledged the emergence of Asia as a global economic and political powerhouse. The first wave of its transformation journey (2010–2013) centered on “Asian service,” which the bank defined with three attributes: respectful, easy to deal with, and dependable (RED).

As the bank successfully streamlined processes to deliver RED, the vision pivoted to “Make banking joyful.” That vision drove initiatives to center the company around the customer experience and digitalization (2014–2018) and fueled the

ambition to become the “best bank in the world.” In the third wave (2019–2024), which focused on data, AI, and sustainability, the vision expanded again; DBS would become the “best bank for a better world.” While continuing to focus on customer experience, the bank is now working on “making banking invisible” through the use of AI and other emergent technologies. In addition, a focus on sustainable and responsible banking and social impact have emerged as new priorities.

Now let’s turn to GE, which pivoted fewer times than DBS did. Around 2011, then-CEO Jeffrey Immelt recognized the need to move beyond commodity hardware to higher-margin software, a market it predicted would grow quickly. So GE began its transformation effort with a software center in California. (The center developed the industrial internet platform Predix and became a separate subsidiary, GE Digital, in 2015.) As Predix evolved and the importance of software capabilities was confirmed, Immelt’s ambition for the business’s transformation grew: GE would be a top-10 software company. He promoted that goal heavily inside and outside the organization.

Although that shift seemed to be a natural progression and a plausible outcome, it strengthened a simmering “us versus them” culture clash. The (relatively) small software teams didn’t fit easily into GE’s historically dominant industrial culture. And because GE’s customers (many of whom were digital beginners) did not fully understand the value of selling analytics services, the company struggled to mobilize stakeholders around the actions it would need to take to become a top-10 software company.

Most problematic, perhaps, was the fact that GE’s pivots seemingly were not informed by what it had learned. Instead GE appeared to follow a preset road map: First deploy the Predix platform internally, then with existing customers, then with all industrial firms. The substantive vision of using the platform to sell diagnostic and data services was not materially adjusted in light of ongoing findings, nor were the internal tensions of introducing new services reconciled with the needs of the existing businesses. Ultimately, GE’s effort was more top-down and less learning-oriented than DBS’s approach. Rather than an exploration, it was a voyage whose destination was set from the start.

Next let’s look at how an evolving vision gets turned into reality.

58

---

The yardstick for a project's success should be how much the company will learn from it: Will it reduce critical strategic unknowns and close gaps in key capabilities?

The Transformation Journey

Each transformation requires the company's leadership to determine which projects and initiatives to launch in pursuit of the vision and how best to orchestrate them. Traditionally, initiatives are treated as steps toward the goal, and success is measured in terms of revenues and profits. However, the yardstick for success should be how much the company will learn from a project: Will it reduce critical strategic unknowns and close gaps in key capabilities? Of course, leadership must also consider the probable impact a project will have on momentum. Will it improve performance in established businesses and tangibly help scale up new opportunities?

As companies travel along on their transformation journey, they can pursue four types of projects. The first two serve as engines of learning, either as controlled experiments designed to test specific hypotheses or as investments in new capabilities. The second two integrate the ideas and practices that arise from earlier projects to drive value in established and new businesses.

1 Pilots. These projects test critical assumptions about the business opportunities and threats embedded in the transformation's vision. They are essentially structured experiments—ranging from new products and processes to different sales channels and monetization models—that should generate essential learning even when they don't yield immediate business results. For example, DBS's 2016 beta launch of its entirely digital retail-banking offer (Digibank) in India was a pivotal pilot in the company's transformation journey. The goal of the pilot was to learn how to attract and serve retail customers without having physical touchpoints. The primary success metric adopted was the number of new customers it acquired, not the revenue it generated.

And in less than two years, the number of customers Digibank had in India went from zero to 2 million. The venture struggled to make a profit, but the bank learned key lessons that led to material improvements in its operations and profitability in other markets. When launching Digibank in India, for example, DBS developed a methodology of scouting, onboarding, and developing ecosystem partners to strengthen customer acquisition and extend its value

offering. It later rolled out this methodology across the wider bank to strengthen its positioning in the home markets (that step was an options project, as we'll see below).

At GE, however, things unfolded differently. When its first pilot tests of the technology platform Predix suggested that many customers were not digital-ready and did not understand the new services, the company didn't take that lesson to heart. Although GE Digital did introduce customer clinics, it plowed ahead with scaling Predix. Rather than learning from negative information, it offered more training and doubled down on its plan.

2 Options. This type of project focuses on building the capabilities—in both people and technology—needed to scale the future business and the operating models that the pilots are testing. These investments should close critical capability gaps, but they typically do not yield direct business benefits. Their success should be measured by how widely and effectively the new capabilities are adopted or deployed. Investments in a training program for employees or the deployment of a new technology are examples. We call these kinds of projects options because they enable the company to carry out tasks that it would not otherwise be able to.

DBS launched numerous options throughout its transformation journey. The bank needed to build employee competencies such as customer journey thinking, disciplined experimentation, data-driven decision-making, and agile ways of working. Each competency required behavioral scaffolding—step-by-step methodologies developed using global best practices and adapted for DBS's specific context, tools, and coaching support. Employees were given time away from their main responsibilities to apply the practices they were learning to real problems, free from top-down, results-driven pressures.

On the technology side, the bank's decision to develop a set of APIs in 2014 was a clear option initiative. Although DBS didn't initially define a business case for them, APIs later became foundational to Digibank's go-to-market strategy in India. For example, API-enabled integration let customers open accounts through a branchless process, supported by biometric authentication at partner locations. Later, APIs also played a critical role in the creation of new

59

---

Within three years DBS rose from last to first in customer satisfaction among Singaporean banks, providing tangible proof to stakeholders that the transformation was working.

60 Harvard Business Review September–October 2026

---

ecosystem-based business models in Singapore, such as DBS car and home marketplaces, by connecting buyers and sellers with the broader ecosystem of partners.

At GE one key option initiative was the FastWorks innovation platform, inspired in part by lean-startup ideas. In order to convert the workforce to the concept, the company put 60,000 employees through a training program. However, it quickly became apparent that the lean-startup “fail fast” approach clashed with the existing “never fail” culture, which had resulted from GE’s Six Sigma approach to quality management. When first deployments did not immediately deliver results (and moreover, GE did not mandate the use of FastWorks in all units), many workers simply reverted to business as usual after the training finished.

To maximize learning, option projects must be tightly coordinated with the pilots, which often reveal the capability gaps that targeted options can address. Conversely, and even though they are not explicitly designed to deliver business results, options can spark ideas that shape future pilots. At DBS the Digibank pilot revealed the need to treat data as a strategic asset, which led to the company’s Data-First initiative—a set of option projects aimed at building data-driven decision-making capabilities across the bank. Similarly, talent hackathons launched in another option initiative not only helped middle managers adopt startup ways of working but also generated prototypes that were later accelerated into pilots and brought to market.

At GE, by contrast, coordination among transformation projects was lacking. The FastWorks program was never reconciled with Predix, and their tool kits (agile light versus agile at scale) were not fully compatible. As a result, the two approaches created friction, not synergy. The requirement for Predix to win more customers was at odds with the business units’ need to keep satisfying their current customers with the products they had. The portfolio of initiatives never became more than the sum of its parts; it was merely a collection of actions that ultimately hindered one another.

3

Established business improvements (EBIs).

Projects in this category apply new strategic insights from pilots and capabilities from options to drive top- and bottom-line growth in core operations. While most organizations do not consider EBIs to be part of the transformation journey, we strongly recommend including these projects in the transformation portfolio. Their success can be measured using traditional ROI metrics, and they are a powerful way to demonstrate early wins and sustain momentum.

EBIs can take place at any point during the transformation journey since they build on existing business models and integrate new capabilities quickly. DBS activated EBI

CHANGE MANAGEMENT

projects both early and late in its transformation journey. As part of its RED service ramp-up, for example, the bank made incremental improvements to its operating model in established markets to deliver faster, more reliable service. Within three years DBS rose from last to first in customer satisfaction among Singaporean banks, providing tangible proof to stakeholders that the transformation was working.

Ensuring that an EBI incorporates the lessons already learned in pilots and leverages capabilities developed in option projects is the key to success. GE again provides a cautionary tale. At the very start of its would-be transformation, it sought to improve its established businesses by applying Predix across internal operations. But unlike DBS, GE rolled out its EBI before Predix technology capabilities were sufficiently established and its business assumptions validated. The business units pushed back. Some quietly built their own tools instead, arguing that Predix lacked key features and that sales teams struggled to explain its value to customers. A series of technical problems and development delays further undermined the business units’ confidence that Predix could meaningfully improve their core business. As a result, GE business units came to see Predix as a distraction rather than a clear value generator, and trust between the units and GE Digital was damaged.

4

Ventures. The fourth type of project focuses on scaling validated business models by leveraging insights gleaned from pilots and capabilities

acquired from options. These ventures target new markets, customer segments, and business models to generate fresh revenue streams. Their success is measured by financial and growth metrics. Scaling up a new business depends on both validated assumptions and well-developed capabilities. Since each is necessary but insufficient on its own, moving into a venture project too early risks failure, as GE’s venture with GE Digital illustrates. The senior leadership team must exercise patience to ensure that its ventures are built on solid foundations.

Consider DBS again. Its first major venture—scaling up Digibank in India after the beta launch—did not occur until year six of the bank’s transformation journey. By then DBS had accumulated the knowledge and capabilities it needed to succeed. Later DBS’s portfolio expanded to include more

61

---

CHANGE MANAGEMENT

initiatives, such as Digibank Indonesia, signaling a confident shift toward scaling up new businesses.

GE, by contrast, moved rapidly. It deployed Predix internally in 2013 and started selling analytics services to existing customers in 2014 as a venture with a separate P&L. Then in 2016 it offered Predix as an “operating system” for all industrial firms worldwide. While the venture promised significant returns if GE secured a winner-take-all position, industrial customers struggled to see the value of GE’s services, and sales teams lacked the training and resources to explain and sell them. As a result, performance lagged, skepticism grew, and support from the board and shareholders weakened.

In addition to coordinating pilots and options, therefore, companies must balance where and when to apply new insights and capabilities to generate tangible business value, whether in EBI improvements or in new ventures. Activities should complement one another, avoid developing momentum in separate directions, and together deliver more than the sum of their parts. Sometimes new learning will lead a company to add new initiatives, and sometimes it will lead a firm to stop a successful initiative. Acting on the lessons each project teaches is critical to having a successful transformation journey.

The Challenge for Leadership

Managing transformation as a long-term journey is inherently difficult. The road map is bound to shift over time, which can cause employees to become disoriented or fatigued, and that can slow momentum. It is the responsibility of leadership to maintain stakeholder enthusiasm along the journey. That can be done in three ways:

Provide discipline. Transformations often begin with scattered initiatives across different teams, budgets, and goals. To prevent a sense of fragmentation or disillusionment, leaders must immediately and deliberately assemble a portfolio of projects and ensure that it corresponds to the vision in place. Each project must have clear goals (not always financial) for its contribution to the overall journey and must serve the broader context of the transformation. Regular communication helps employees understand why changes are necessary and can prevent the perception that some units or stakeholders are being shortchanged.

The executive team also needs to create a strong internal transformation team. At the start of its journey DBS created a transformation group, whose primary mandate was to identify critical gaps in skills and capabilities and to design and run transformation methodologies—such as customer journeys, business experiments, and agile practices—that helped employees across the bank embed new ways of working into their day-to-day activities. The group was situated within the technology and operations group rather than being a stand-alone unit.

This positioning had two benefits: The group was closely aligned with technology pilots and EBIs, and it avoided the perception of being a detached corporate function. The transformation group worked closely with the technology scouting and innovation team, which focused on building connections among internal and external stakeholders. Together, the groups operated as enablers and catalysts rather than as the primary engines of transformation. At GE, however, the main transformation thrust rested with GE Digital, which imposed change on the core business units.

At DBS the projects were widely communicated and explained so that even people who were not directly involved could understand what was happening and could adjust to the projects’ effects. For example, to help translate the vision of “Make banking joyful” into concrete actions, DBS created and actively promoted a model it dubbed

62

---

Transformation goals cannot be achieved through decisive action and resources alone. They must be explored, revised, and continually reconnected to an evolving vision.

“Acquire, Transact, Engage” across all business units. Each step was clearly explained, and training was made accessible to all employees. At GE, leaders delivered no unifying messaging, not because they did not know that communication was important (there was communication about successes and outcomes) but because the interdependent and complementary role of the main programs (Predix and FastWorks) was never identified as a key enabler of the transformation.

Welcome bottom-up ideas. At DBS the first pivot to customer journey-driven design thinking arose from learnings revealed in operational activities. The building blocks for portfolio choices came from knowledge dispersed throughout the company; top management prioritized and funded the most promising. Many bottom-up ideas gave rise to important projects in the DBS transformation portfolio. For example, the emphasis on customer journeys grew from a realization by a team working on process improvement that not all customer problems could be fixed by streamlining internal processes. The team developed a methodology and gave a demo to management, which was impressed and took it on board. Over time, customer journeys became a key element of the transformation, giving rise to the vision of “Live more, bank less.”

To be able to recognize the most-promising ideas, DBS’s top executives regularly participated in transformation activities. The CEO, for instance, met weekly with the Digibank team during the pilot development and launch. The Digibank pilot was an inflection point in the transformation journey, producing many bottom-up insights, such as the ecosystem business models and data-driven decision-making, that later were elevated into option initiatives.

At GE, though, the transformation was driven largely by top-down ideas, limiting pivots from employee learning and reducing buy-in from the wider organization. We have found no evidence that bottom-up ideas managed to trigger a significant change in the course of the programs that were being pursued. To the contrary, there is some evidence suggesting that any doubts about the chosen course or pace of advancement were brushed aside by senior management.

Incorporate pivots into measurement systems. Learning must be reflected in evolving scorecards that chart a clear—if dynamic—road map. At DBS early tracking focused on process improvement, measured as the time customers spend interacting with the bank. Over time, measures

shifted: first to customer “joyfulness” and then to revenue from digital touchpoints. These evolving measures signaled priorities and reinforced transformation logic. At GE the absence of such evolving metrics contributed to perceptions of unfairness, organizational tensions, and ultimately the breakdown of support.

DBS deliberately used the corporate scorecard to support its transformation effort. In 2016 transformation KPIs were introduced into the scorecard under the label “Make banking joyful,” and over the years their weight grew from 15% to 20%. The scorecard evolved to explicitly reinforce key elements of the transformation portfolio. Initially it focused on critical options and EBIs, but in the eighth year some of the more strategic venture projects appeared under the strategic priorities section (40% of the total).

In other words, the scorecard served as a vehicle to signal to the entire organization what elements of the transformation portfolio had the greatest salience at a given stage. It reinforced behaviors that supported those priorities and, over time, made the journey continuous and comprehensible. The scorecard helped leadership communicate, “This is where we have come from, and this is where we are headed.”

CEOS ARE EXPECTED to articulate and then achieve goals. And it’s natural for them to treat all challenges that way. But large transformations involve deep architectural changes that unfold over long time periods and produce outcomes that are at least partly unforeseeable. Transformation goals cannot be achieved through decisive action and resources alone. Instead they must be guessed at, explored, revised, and continually reconnected to an evolving vision of what the organization should be. That’s why we contend that leaders must become intrepid explorers, ready to adjust course as their goals and environment evolve. Their fundamental challenge, perhaps, is to balance the humility that exploration requires with the confidence needed to inspire the crew. ▼

HBR Reprint R2605C

EVGENY KAGANER is a professor at IESE Business School in Barcelona and the academic director of the IESE-MIT Global CEO Program. CHRISTOPH LOCH is a professor at IESE Business School in Barcelona and the former dean of Cambridge Judge Business School.

63

---

Collaborate on the Core.

Compete on the Edges.

64 Harvard Business Review September–October 2026

---

PHOTOGRAPHER CHRIS AULD

65

---

ABOUT THE ART

Chris Auld turned to cycling photography after 20 years of working as a commercial photographer, capturing numerous races, including the Tour de France.

IDEA IN BRIEF

THE PROBLEM

Most companies struggle with how much to collaborate. Go it alone, and you slow yourself down. But collaborate too much, and you risk handing competitors your advantage.

THE INSIGHT

The smartest companies draw a line. They collaborate on the “core”—things like infrastructure and standards—to help the whole market grow. Then they compete hard on the “edges,” where they can differentiate themselves.

THE SOLUTION

Find the best course by weighing five factors: market dynamics, technology life-cycle stage, your position in the technology stack, the level of competition, and social acceptance and regulation. Collaborate when doing so will grow the market or reduce risk, compete where you want to stand out, and keep adjusting as things change.

66

---

The lesson for business isn't just "collaborate more." It's "collaborate intelligently." Draft when it makes you go faster. Break away to win.

STRATEGY

G

VERY SUMMER THE TOUR DE FRANCE turns endurance into theater. For three weeks the world watches cyclists traverse mountains, wind-swept flats, and chaotic city finishes. If you want to see strategy in motion, watch the peloton. Riders from dif-

different teams—direct competitors with different sponsors, leaders, and ambitions—form a dense, fast-moving pack. Inside this temporary, informal alliance, everyone benefits from reduced wind resistance. The front rider does the most work as the riders behind conserve energy. Positions rotate. Over time the group moves faster than any rider could alone.

And then, suddenly, the truce dissolves. A climb steepens. Speeds drop. Aerodynamic advantages fade. The contenders must choose the optimal moment to break away. In the Tour, cyclists are constantly oscillating between two modes: collaboration when it makes the whole group stronger, and competition when differentiation decides the outcome.

In today's markets, speed, interoperability, and trust determine whether new technologies gain traction. Most companies can't afford to build a technological ecosystem alone. This is partly why we see so many circular deals in the AI sector. AI systems depend on shared infrastructure that no single company fully controls. Training models require chips, cloud capacity, data pipelines, software tools, and distribution channels, which are owned or supplied by different firms. Circular deals, in which companies invest in one another or sign mutually beneficial partnerships, help align incentives. Firms share key components, lock in supplies, and signal their commitment to a common ecosystem. Such commitments have led fierce competitors in this

space to band together to create shared-standards groups (for example, the Agentic AI Foundation) that help ensure that in the future, ecosystems remain interoperable. In effect, this sharing reduces risk. Each participant knows the others have skin in the game, which makes it easier for the whole technology ecosystem to scale up quickly.

But if companies collaborate too broadly, they risk losing hard-won competitive advantages. That means the real strategic question is not whether to collaborate but where to do it. To answer that, it's necessary to separate a company's activities into two layers. The first, the core, includes the infrastructure, interfaces, standards, and safety practices that allow its ecosystem to function and grow. The second layer, the edge, is where companies create distinct value for customers by using proprietary workflows, integrated experiences, brand loyalty, data, and business models. The challenge is that the boundary between the two keeps moving. What differentiates you today may become commoditized tomorrow, and collaboration increasingly happens through informal ecosystems rather than traditional alliances. For that reason, leaders need a practical framework to decide, case by case, when collaboration strengthens their strategy and when it quietly erodes it.

What follows is the framework I use to help leaders make that call. It is based on my academic research, interviews with dozens of industry practitioners, and my work as a consultant. It's built around five factors and grounded in identifiable patterns across both technology-heavy and low-tech industries. If the Tour has a lesson for business, it isn't just "collaborate more." It's "collaborate intelligently." Draft when it makes you faster. Break away when it's time to win. Businesses must learn to do the same.

THE HIDDEN COST OF INDEPENDENCE

Most leaders don't struggle because they dislike collaboration. They struggle because there's a high cost to failure, and failures happen frequently. If you undercollaborate, you can lose years to duplicated work, fragmented standards, and ecosystem distrust. If you overcollaborate, you can accidentally train your competitors, leak strategic know-how, or commoditize what you once owned. In companies' practices, I see three predictable mistakes.

67

---

68 Harvard Business Review September–October 2026

---

STRATEGY

First, executives often talk about cooperation as if it's binary: "We partner" or "we don't." But collaboration comes in many forms, ranging from lightweight coordination (shared standards or shared threat intelligence) to deeper codevelopment (formal R&D partnerships). The idea that collaboration must mean a full alliance with shared teams, road maps, and governance creates unnecessary fears about control, dependency, and value capture. The better model is to treat collaboration as a spectrum of options, each with different risk levels and different strategic payoffs.

Classic co-opetition guidance emphasizes how much of the challenge is about scope and control. How far does cooperation extend? Who governs the collaboration? And how can it be unwound if it stops making sense? Once leaders see collaboration as a set of tools, not a single big bet, they can choose the lowest-risk mechanism that still achieves their goal.

The second mistake is believing that collaborating on a shared core is mostly an act of goodwill. That common belief becomes self-fulfilling: Leaders starve collaborative efforts because they can't see the ROI, and then those efforts fail because they're underresourced. But many of the most impactful examples of collaboration are best understood as competitive moves designed to reshape the terrain of competition.

Apple and Google's joint work on exposure notification during Covid-19 is an excellent example. Two companies that compete intensely at the platform level built an interoperability layer that allowed public-health authorities to broadly deploy Bluetooth-based exposure notifications. Apple framed the collaboration as an effort to help health agencies reduce the spread of the disease (a social good for everyone) while focusing on the enhanced privacy and security of the iPhone (a point of differentiation between iOS and Android often underappreciated by consumers). Even if the circumstances were unusual, the strategic logic is familiar: When customers' adoption of a product depends on cross-platform interoperability and trust, the core is too important to remain fragmented. But collaboration on the core can offer an opportunity to shine a light on key aspects of differentiation.

The third mistake is blindly following the traditional cooperative strategy playbook. Classic cooperative strategy tends to emphasize formal alliances, joint ventures, and contractual partnerships. Those still matter. But in

69

---

the digital age many of the most important cooperative dynamics occur outside a formal framework. As technologies become more modular, layered, and ecosystem-dependent, firms increasingly collaborate through informal or community-driven approaches. That creates a new challenge: Leaders can't rely only on the familiar joint-venture checklist. They need a way to evaluate newer forms of collaboration.

THE FIVE-FACTOR FRAMEWORK

The economic logic behind the framework is that collaboration increases value for everyone. Collaborating on the core can either raise customers' willingness to pay (through increased trust, interoperability, quality, and adoption rates) or lower costs (through shared R&D and reduced duplication). Competing on the edges is a value-capture play. It's where you build the differentiated experiences, workflows, distribution, and brand promise that allow you to command premium pricing, lock in customers organically, and sustain or improve your competitive advantage. But where do you draw the line between collaboration and competition?

The good news is this line isn't mystical. Across industries, it tends to be shaped by five predictable factors. Once leaders explicitly incorporate those factors into their deliberations, they stop arguing about vague philosophies ("open versus closed" or "partner versus build") and start making clean, revisable decisions.

Now let's look at how each factor shapes the boundary between the core and the edges and how real companies have used the framework.

FACTOR 1

Winner-take-all dynamics. Some markets reward interoperability and allow many winners. Others tend to tip toward one dominant platform, standard, or ecosystem. In winner-take-all (or winner-take-most) markets, collaboration can be unusually risky or unusually powerful, depending on how it affects the tipping point. If the market is likely to tip toward a single-owner-controlled platform, you must be careful not to fund your future capter. If the market is likely to tip toward an open or shared standard, collaborating early can help prevent fragmentation and ensure you're helping control the standard

70

---

while still competing on the edges that matter to customers. If the market is unlikely to become winner-take-all, then more collaboration can be helpful.

Consider the technical body EMVCo, which is overseen by major payment networks including American Express, Discover, JCB, Mastercard, UnionPay, and Visa. It develops global specifications so that card-based payments work securely and seamlessly. Payments aren't a winner-take-all business. Most people have multiple credit cards and use more than one payment processor for purchases. The major payment networks collaborate on the core rails because they all benefit from interoperability. But they compete intensely on fraud tooling, issuing relationships, merchant services, pricing, and rewards. That's a classic "open standard core, differentiated edges" move. Standardize what must work everywhere; compete on devices, ecosystems, and user experience.

In winner-take-all markets, the goal is neither to collaborate nor not to collaborate. The goal is to collaborate in ways that prevent the market from tipping toward competitors and to ensure the competitive battle happens where you can win.

FACTOR 2

Technology life-cycle stage. Are you growing the category or fighting for share? Technologies tend to follow an S curve: uncertain early days, rapid growth, maturity, and eventual decline. The balance between collaboration and competition shifts along that curve. In early stages, a company's biggest problem is not usually beating rivals. It's getting adoption: proving viability, establishing standards, reducing uncertainty, and building complements. Collaboration can be a way to accelerate the growth of the category itself. As a technology enters the growth and maturity stages, the strategic path moves toward differentiation: feature advantage, customer experience, cost position, and product stability. Collaboration still happens, but often narrowly and defensively (around shared risks like security, for example). Finally, in the decline stage, as a technology is being phased out or replaced, competitors often collaborate to extend its life, especially if they are not the ones building the replacement technology.

Let's take a look at the electric vehicle (EV) industry. Early in the EV transition, the category faced a classic adoption barrier: Consumers, suppliers, and infrastructure providers

71

---

needed confidence that EVs would be a viable market. Tesla decided to open access to certain battery and charging patents to encourage broader industry adoption and ecosystem investment. The company collaborated with other EV manufacturers on parts of the enabling environment while still competing on product, brand, and execution. For instance, Tesla supplied battery packs for Toyota's RAV4 EV. Although many onlookers questioned this move at the time, it helped grow the EV market and had the added benefit of giving Tesla control of the standard for a growing industry.

Such collaboration early in the technology life cycle happens across a variety of industries. For example, in the pharmaceutical domain, "precompetitive consortiums" allow competitors to collaborate on fundamental research that will help them create new drugs and therapeutics.

The consortium led by Regeneron to sequence the genomes of 500,000 participants in the UK Biobank is one example. Regeneron and its competitors AbbVie, Alnylam, AstraZeneca, Biogen, and Pfizer each contributed $10 million to support the effort. Their collaboration sped up the sequencing timeline by three years. The companies that participated had access to this extremely important data and could use it to create new drugs.

For an example of a lack of collaboration in the growth stage, look at the streaming services industry in the late 2010s. As competition intensified, major players such as

Netflix, Amazon Prime Video, and Hulu sought to differentiate themselves by offering unique user experiences and exclusive content and using proprietary algorithms. Because they were focused on capturing market share and establishing brand dominance, they competed with one another instead of collaborating.

As noted earlier, once a market reaches the decline stage, its players often revert to collaboration. For example, consider compact discs (CDs). As digital streaming rose to prominence and CD sales slowed, CD manufacturers and music labels collaborated on marketing efforts and special releases to try to extend the life of the technology in the face of disruption.

Life-cycle thinking prevents executives from making a common error: assuming that because a collaborative move made sense early, it remains the right move forever. The core often stabilizes over time, and the edge battle intensifies.

FACTOR 3

Technology stack position. Technologies live in stacks. Lower layers (or infrastructure) benefit most from interoperability, while higher layers (applications) benefit most from differentiation.

We see this dynamic regularly within big tech. For example, when Google developed the cloud technology

72 Harvard Business Review September–October 2026

---

Kubernetes (which automates the management of applications, among other things), the company understood that because it was at such a low level in the tech stack, it wasn't going to drive competitive advantage. Therefore, Google shared it freely with the world and then donated it to the nonprofit Cloud Native Computing Foundation. Within a few years, Google's primary competitors in the cloud computing space (Microsoft's Azure and Amazon's Web Services) adopted Kubernetes and started relying on it heavily. They also contributed to its development. Although Google had given away an important piece of technology to its competitors, doing so created value by helping to speed adoption of cloud computing. In addition, Google got a multiyear head start over competitors on deploying and integrating this now-crucial technology.

Conversely, consider Uber and Lyft, which compete at higher levels of the technology stack. Although they rely on similar underlying technologies for their mobile apps (GPS navigation and payment processing, for example), the two companies engage in fierce competition to differentiate their service offerings, user experience, and brand position using technology in the middle (pricing and routing algorithms) and top (user interface) of the stack.

What does this mean for executives? If you're working on infrastructure layers, such as protocols, interfaces, safety practices, and enabling components, you should default to

asking, "What must be interoperable to unlock adoption?" Those core areas are candidates for collaboration.

If you're working closer to the end customer—for instance, on workflow integration, proprietary data, branding, or distribution—your default question should be, "What must remain unique for us to win?" The answers are your candidates for competition.

Stack position alone won't make the decision for you, but understanding it dramatically improves the odds that you'll draw the line separating the core from the edge in the right place.

FACTOR 4

Level of competition. When rivalries are intense, shared risk increases. It's intuitive that low competition can make collaboration easier: Firms feel less threatened, and market-building benefits are clearer. Although the incentive to collaborate decreases as competition approaches moderate levels, what's less intuitive, but extremely important, is that high competition can also increase collaboration in specific domains, especially around shared risk. Cybersecurity is one example.

Security is adversarial; attackers exploit the weakest link. If one firm learns about a new vulnerability and sits on it, an attacker can exploit it elsewhere, which ultimately hurts customers and erodes trust in the entire category. As a result,

73

---

STRATEGY

competitors often share intelligence on threats even while competing hard for customers. In 2014 Fortinet, Symantec, McAfee, and Palo Alto Networks created an informal agreement to share cybersecurity intelligence. The effort was so successful that the firms later formalized it by creating the Cyber Threat Alliance, which is still active today and has dozens of member companies, all of whom are competitors. From a strategy perspective, this is “collaborate on the core risk layer, compete on the edges of product and service.” You collaborate in areas where the marginal value of shared learning is enormous and where failure harms everyone’s legitimacy. You compete on what you can charge customers for (outcomes, integration, performance, and trust).

High competition also raises the bar for collaboration governance. If you’re collaborating in one area while fighting intensely elsewhere, you need sharper boundaries, clearer scope, and explicit paths for unwinding joint efforts.

It’s important to recognize that these principles apply to product market competition. With labor market competition (that is, when companies are competing for the same employees), the relationship is different. When labor competition is moderate, companies can offer collaborative opportunities (on R&D, for example, or on open source software) as a perk to attract better talent. However, when labor competition is low, companies don’t need to offer this benefit, and when it is high, they cannot afford to.

FACTOR 5

Social acceptance and regulation. When legitimacy is at stake, collaboration can help stabilize a market. Some technologies face skepticism, backlash, or regulatory uncertainty. In those cases, collaboration often becomes a way to build legitimacy, establish responsible standards, and avoid the chaos of conflicting rules.

Apple and Google’s cooperation on exposure notification is again instructive. The strategic subtext was simple: If public trust collapses, adoption collapses. Collaboration on a privacy-preserving core enabled broader trust and increased usability while leaving the companies free to compete vigorously in other areas.

Now zoom out beyond “technology” in the narrowest sense. Technology is broader than software and the digital

economy. It includes products, processes, and even category-level practices. The five-factor framework still applies. Let’s look at the soft drink industry. As public concern about sugar and obesity rose, major beverage companies came under scrutiny. One response was collective action—industry commitments and formal and informal partnerships. A 2014 collaborative effort between fierce competitors Coca-Cola, Dr Pepper Snapple Group, and PepsiCo aimed to reduce beverage calories consumed per person nationally. In one TV campaign, delivery people from each company happily walked down the street together to let consumers know they were all working collaboratively to address the public’s concern.

Similar cross-industry efforts are commonly seen in times of regulatory uncertainty, when competitors will work together to help shape the laws that govern their industry. From a strategy perspective, this follows the same pattern: Collaborate on the core legitimacy layer (shared commitments, transparency norms, public messaging, and measurable goals), and compete fiercely on products, branding, pricing, and distribution, which are the edges where customers choose who wins.

When social acceptance and regulation are central, treating legitimacy as a PR problem is the wrong move. It’s an ecosystem design problem. Collaboration can be the fastest way to set credible norms, especially when regulators are watching and the alternative is a patchwork of incompatible or punitive rules.

ANTHROPIC AND OPENAI’S COLLABORATION

To see all five factors in one contemporary case, look at Anthropic’s decision to share the Model Context Protocol (MCP), a standard that allows AI models like LLMs and agents to talk to external datasets and tools. Anthropic donated the MCP to the newly formed Agentic AI Foundation (AAIF), which it has joined with its direct competitor OpenAI (as well as other competitors in the AI agent space, including Google and Microsoft).

Anthropic competes in a market in which ecosystems form around developer tools, integration, and compatibility. Using multiple platforms simultaneously is possible but costly for enterprises. Establishing standards can reduce market fragmentation, and whichever company shapes the

74 Harvard Business Review September–October 2026

---

Be explicit about what information stays off-limits to your collaborators, how decisions will be made, and how a collaborative arrangement could change if conditions do.

standard influences the direction of the market. Anthropic's announcement says it wants to keep MCP "open, neutral, and community-driven" as it becomes the established, enabling infrastructure. If we assess that move using the framework, all five factors suggest this was the right decision.

Winner-take-all: If the agent ecosystem tips, it will likely tip toward shared interfaces and developer defaults. By sharing MCP with competitors, Anthropic put itself in a strong position to help set the direction of the standard.

The tech life-cycle stage: Agentic AI systems are in an early phase, which means market growth depends on reducing friction and uncertainty.

Stack position: MCP sits in the enabling layer, an interface and protocol that allows tools and models to work together; competitive applications can be built on top of it.

Level of competition: Rivalry is intense. No firm wants to bet the future on a competitor's proprietary interface. Neutral governance by the AAIF lowers that risk.

Social acceptance and regulation: As AI scrutiny grows, credible governance and openness can improve trust and accelerate responsible adoption.

Anthropic isn't collaborating to be nice. It recognizes that collaborating on the core can be the fastest way to grow a market and prevent it from moving in a direction Anthropic doesn't want. It has plenty of room to compete on the edges with things like models, products, safety systems, enterprise workflows, pricing, distribution, and brand.

HOW TO GET STARTED

A good place to start is to map your technology stack. Identify the layers that you operate in—from the infrastructure and interfaces that enable your technology to the user-facing experience. Consider what customers value at each layer. That will help clarify which parts function as shared foundations and which ones shape the customer experience.

Next, list potential core candidates. These are components for which interoperability, trust, or shared risk reduction could increase adoption or lower costs. At this stage it helps to cast a wide net. You can narrow the list later.

Then evaluate each candidate using the four other factors of the framework. Ask whether the market is likely to tip toward a dominant standard and where the technology

is in its life cycle. Consider how intense competition is and whether rivals face shared risks like security threats. Next, think about the status of social acceptance and the regulatory environment.

Once you've done that, choose the lightest collaboration mechanism that will achieve the goal. Often you don't need a formal alliance. A shared technical standard, a neutral-governed project, or shared threat intelligence may be enough. The key is defining the scope of collaboration and the control mechanisms clearly from the beginning and ensuring that you are complying with antitrust regulations.

You must identify where to draw the boundary between collaboration and competition. Be explicit about what information remains part of your competitive edge and stays off-limits to your collaborators, how decisions will be made, and how a collaborative arrangement could change if conditions do.

It's important to design your edge strategy in parallel. A common mistake companies make is building a strong shared core but neglecting the differentiated capabilities that win customers. Decide early where you will be meaningfully different from competitors and invest there.

Finally, revisit the boundary regularly. As markets mature, yesterday's edge often becomes tomorrow's core. Reexamining the line every six to 12 months helps ensure that your strategy evolves with the market.

THE TOUR DE FRANCE is not a lesson simply in teamwork or in individual heroics alone. It's a lesson in switching modes at the right moment. And it's one that companies should take to heart. Build everything yourself and you'll arrive late; collaborate without boundaries and you'll help rivals catch up or surpass you.

The strategic skill is drawing the line: Collaborate on the core to grow the market, reduce risk, and accelerate adoption, then compete on the edges to deliver unique value and capture it. In other words, ride in the peloton when it makes you faster, then choose your moment and break away. ▼

HBR Reprint R2605D

FRANK NAGLE is the chief AI economist at Microsoft and a research scientist at the Initiative on the Digital Economy at the Massachusetts Institute of Technology.

75

---

ORGANIZATIONAL DECISION-MAKING

ILLUSTRATOR RODRIGO VALENZUELA

AUTHORS

Kris Johnson Ferreira

Associate professor, Harvard Business School

Jordan Tong

Professor, Wisconsin School of Business

A new generation of tools can connect functions and reduce the friction that slows companies down.

76

---

ABOUT THE ART

Rodrigo Valenzuela's New Land paintings explore the history of westward expansion, using a labor-intensive printing process to represent the bureaucratic burdens endured by immigrants.

MAGINE A NEW TARIFF IS ANNOUNCED ON imported goods that are crucial to a company's business and will take effect in 60 days. The CEO of the company recommends "pulling forward" inventory—buying it sooner than originally

intended and importing it before the tariff hits—to protect margins and market share. The logic is clear. Yet weeks later the move hasn't been carried out as the CEO wanted. After rounds of revisions and compromises among different functions—including procurement, logistics, finance, legal, marketing, sales, and merchandising—the company implements a patchwork of suboptimal actions that fail to secure a large portion of the crucial goods before the tariff goes into effect.

Over the past decade, companies have invested heavily in smarter tools, including AI assistants that draft documents and summarize reports and machine-learning systems that forecast demand. When employed well, they improve the performance of clearly defined, narrowly scoped tasks within functions. But most companies have yet to address a deeper structural challenge: how to employ AI to make and implement enterprisewide decisions. This is the next frontier: AI-human systems that coordinate decisions and knowledge across functions.

These are what we call agentic AI orchestration systems, and in our field research on human-AI collaboration, we've interviewed executives at Walmart, Amazon, Ericsson, Ramp, Medtronic, and other companies that are building them. They're making real progress, but the initiatives are still in the early stages: Even the most advanced organizations have far to go to achieve full enterprisewide orchestration. Still, it's possible to see how this technology will evolve, and our additional ongoing research on the most effective ways for AI and humans to work together reveals how these systems should be set up. We consistently find that people create the most value not by competing with AI but by contributing information it does not have—such as tacit knowledge and updates about changing constraints and contextual shifts. The challenge, then, is not deciding whether humans or AI should make decisions. It is designing systems that combine the strengths of both.

Realizing orchestration systems' potential will require more than installing new technology. Companies will have to redesign how higher-level decisions flow from lower-level ones, how human-AI interactions are structured, and how workflows are integrated. The payoff will be substantial: higher-quality cross-functional decisions that are executed faster without surrendering human judgment or control. In fact, these systems will preserve—and in some cases strengthen—managerial authority by embedding it in the ways decisions are structured, informed, and governed. Organizations that build AI orchestration capability into their infrastructure—rather than relying on managers to stitch things together—will gain an advantage over their rivals.

THE HIDDEN BOTTLENECK

The decisions that matter most in modern organizations typically span multiple functions. They involve interacting

IDEA IN BRIEF

THE PROBLEM

Companies have invested heavily in AI tools that speed up work, yet cross-functional decisions still get bogged down.

THE CAUSE

Information about constraints and trade-offs remains fragmented in silos.

THE SOLUTION

Create master AI orchestration systems that coordinate analyses, route information, and surface trade-offs across the organization, while humans contribute contextual knowledge, set guardrails, and retain final decision authority at the enterprise level.

78

---

79

---

constraints, data from multiple sources, and human knowledge and judgment shaped by experience rather than spreadsheets—which often must be applied in dynamic situations where outcomes and downstream effects are inherently uncertain.

Tariff responses are just one example. The same challenges often emerge when companies experience a sudden, unexpected event that significantly increases or decreases demand for their offerings, suffer supply chain disruptions, or redesign their product portfolios. These kinds of situations require enterprisewide solutions, whose quality depends on how well the organization integrates partial, distributed knowledge into a coherent course of action.

Acting on the decision to pull forward inventory before a tariff takes effect, for example, requires addressing a web of interdependent questions: How will customers respond to

price changes? To what extent are suppliers willing to accelerate production? Do our warehouses have the capacity to hold additional inventory? Can our transportation networks accommodate a surge of deliveries? Are there product-lifecycle or obsolescence risks that make front-loading inventory unattractive for some SKUs? How will working capital, risk exposure, and earnings volatility change under different scenarios?

Most organizations have the teams, tools, data, and expertise needed to address such questions. But many aren't skilled at gathering information from disparate parts of the organization and synthesizing it quickly. That task typically falls to a senior leader, who must decide which questions to ask, anticipate constraints that haven't yet surfaced, reconcile conflicting objectives, and manage iterations as new information emerges. Much of that information exists

80 Harvard Business Review September–October 2026

---

We found that performance improves when humans are trained to intervene only when they hold important information the AI lacks.

only in people's heads. It surfaces informally, unevenly, and often too late. A logistics manager mentions that warehouse capacity and labor are already committed for a seasonal peak. A procurement leader points out that a key supplier is willing to speed up deliveries only if volumes are guaranteed across multiple product lines. Each late discovery forces the process to reset. Scenarios are reworked. Assumptions are revised. Meetings multiply.

It's tempting to blame such breakdowns on culture—on bureaucracy, risk aversion, or misaligned incentives. But the fundamental causes are these: No single individual's brain can hold all the relevant information, constraints, and scenarios. Information is asymmetric; each function knows things that others do not, and often people cannot articulate what will matter until it matters. Issues and risks don't show up all at once; they emerge one by one. So delays in carrying out strategies that entail a host of cross-functional decisions aren't signs of managerial failure. They're signals that the organization has exceeded its capacity to orchestrate decisions.

WHY BETTER AI TOOLS HAVEN'T SOLVED THE PROBLEM

It's natural to assume that artificial intelligence should provide relief. After all, many of the tasks that slow cross-functional decisions—forecasting demand, estimating risk, generating documents, extracting information from contracts—are precisely the domains where AI has made impressive strides.

Agentic AI can outperform humans when tasks are narrow and well-defined, inputs are complete, and historical data is representative of the current situation. But orchestrating cross-functional decisions means coping with situations in which those conditions often aren't present.

The solution is to combine human judgment and machine intelligence—both for individual tasks and across the broader decision process. At the task level, this means structuring human-AI collaboration so that people contribute information the system lacks and use it to shape outputs. At the orchestration level, it means connecting those outputs—along with the human knowledge embedded within them—so that they're validated, routed, and integrated across tasks and functions.

THE TASK LEVEL

In most organizations today, employees using task-based agentic AI are expected to improve its performance by monitoring and overriding it when they think they can do better. Yet in practice they often struggle to determine when and how to intervene. In our experimental research, we found that humans tend to develop a general level of trust in AI that they then apply to all situations, rather than adjusting their confidence in AI when its performance is likely to be strong or weak. That sometimes causes people to override AI recommendations when they shouldn't or miss instances when they should, leading to significantly worse outcomes in both cases.

Underlying this pattern is a flawed "human versus AI" mental model. The correct model is "human + AI." The right approach is to ask, "Do I know something important—context, constraints, or tacit knowledge—that the AI can't access or infer on its own?" Only when the answer is yes is human intervention valuable.

In the tariff example, this kind of human information includes tacit knowledge of supplier concessions or constraints and insights into how demand might change if store shelves are flooded with extra inventory. Such information is often sparse, subjective, and unevenly distributed across the organization. Most task-based agentic AI tools aren't designed to surface or integrate it systematically.

Our research points to a more effective approach. In a series of controlled experiments, we found that performance improves when humans are trained to intervene only when they hold important information the AI lacks—either supplying it as an input before the system makes its recommendation or using it to make an adjustment afterward—rather than asked to form independent judgments that compete with the AI's.

Practically, the goal is to have people address questions like these: Does the AI have all the critical input data? Can I still trust the historical relationships between the inputs and the outcome? Is this situation represented in the data the model learned from? If the answer to all those questions is yes, people need not intervene. If the answer to any of them is no, people should adjust the AI tool's inputs or output to incorporate the human-only information. This can be

81

---

The most consequential decisions aren't made within a single task; they emerge from linking together many narrowly scoped tasks.

accomplished by creating a simple checklist that employees can review when using task-based AI agents or by building an AI agent that prompts employees to ask these questions and suggests how to integrate the answers into the tool.

In the hypothetical tariff example, here's how human-AI collaboration could work:

→ A procurement large language model (LLM) scans supplier contracts and identifies excess capacity that could be tapped to speed up the delivery of orders before the tariff goes into effect. But a purchasing manager has had a recent conversation with a key supplier that has revealed critical information the system lacks: Several of the supplier's other customers are pursuing the same strategy, and the supplier will dedicate additional capacity to the company only if it expands its purchases to other items it doesn't buy from the supplier and sell in its stores. In response, the procurement manager revises the LLM's inputs or output to reflect that what appears feasible in the contracts is actually contingent on awarding the supplier additional purchase orders for items the company doesn't currently sell.

→ An AI tool has used data on early-season sales of the current product assortment to predict demand for it later in the season. A merchandiser, however, recognizes that early-season sales of the subsequent assortment would be adversely affected by overstocking current products to mitigate the impact of a higher tariff and therefore adjusts the AI's demand forecast to take that into account.

→ When reviewing AI forecasts for warehouse capacity utilization, a logistics manager recognizes that the relationship between inputs and outputs has shifted since the model was trained: The head of the seasonal-products division has decided to double the amount of in-store inventory for the upcoming Christmas season to support a new product launch. In response, the manager revises the AI's baseline forecast upward to reflect the expected increase and sees that there wouldn't be enough capacity to house extra inventory pulled forward in anticipation of the tariff.

This is a fundamentally different workflow. The responsibility of humans is not to form beliefs about what the best decision should be. Their job is to identify whether they possess information the task-based AI agent doesn't and, if they do, to revise the tool's inputs or outputs to incorporate that knowledge.

THE ORCHESTRATION LEVEL

The most consequential decisions aren't made within a single task; they emerge from linking together many narrowly scoped tasks. That explains why, despite significant investment in AI and even with better human-AI collaboration, organizations still struggle to move faster on their most important decisions. What's been missing is a system that ensures that human-only information and the outputs of individual tasks are validated and shared with other parts of the organization.

At the foundation of this system is a rule-based layer of agents we call connectors. They take outputs from task-based AI agents, verify that those outputs stay within management-specified guardrails, trigger the next step in the workflow, and pass relevant information on to other task-based agents as inputs or to a higher-level orchestrator. The role of connectors isn't to make decisions; it's to ensure that multitask or cross-functional workflows are executed effectively.

The connectors' guardrails let managers set business rules and decide how much freedom the system has when making decisions. These can be simple—like checking that something is feasible or making sure results fall within an expected range. They can also be more advanced, using other AI tools to run additional checks. For example, a connector might double-check that a forecast isn't unusually high or low according to limits set by management. If something doesn't meet the rules, the connector can stop the process, redirect it, or ask a conversational AI agent to get more input from an employee. In sum, connectors help information move through the system in a structured, consistent, and reliable way.

Overseeing this layer is the master AI orchestrator: an AI agent that employees engage with to share information only they possess, describe external issues that may affect the business (like a new tariff), surface potential opportunities (like pulling forward inventory), and explore what-if scenarios. The master orchestrator doesn't make high-level enterprise decisions; it structures and coordinates the analyses and information needed for humans to make them. In conversational exchanges, it gets employees to share human-only knowledge and articulate objectives, constraints, and decision boundaries. It helps identify and coordinate which tasks, workflows, and employees

82

---

need to be involved in a decision, routes questions to the appropriate tools and people, and manages the overall decision process.

The master AI orchestrator creates a plan, and the connectors execute it by triggering and advancing workflows, validating outputs, and routing information across tasks in a disciplined and repeatable way. Together, the orchestrator and the connectors create a continuous flow of information in which task outputs and critical human-only knowledge become structured inputs for other tasks, vastly improving coordination and drastically shrinking the time needed to make and implement decisions.

Let's return to the tariff example. In a traditional organization, the decision to pull forward inventory would trigger a chain reaction of emails, meetings, requests for analyses, and discovery of constraints. Even with task-based AI agents, the burden of stitching together information would still fall on

a human orchestrator, and critical information would often arrive late—or never travel beyond the silo where it resides.

With agentic AI orchestration, the process unfolds differently. The CEO identifies the tariff issue and the potential opportunity to pull forward inventory and enters this information into the master AI orchestrator, which asks the CEO to clarify objectives and constraints. It then engages the relevant workflows and functional managers—in demand planning, procurement, logistics, merchandising, and finance—and initiates parallel analyses using task-based AI agents. As these analyses run, connectors link them together and prompt the master AI orchestrator to elicit and share relevant human-only knowledge.

Let's look at what happens with the procurement workflow. The master AI orchestrator alerts the procurement manager of the CEO's desire to explore pulling forward inventory and triggers the procurement contract-scanning

83

---

LLM to identify suppliers with possible excess capacity. This task-based tool generates a list of relevant suppliers and contract terms, and a connector provides that output to the master AI orchestrator, which prompts the procurement manager for critical human-only information. After holding discussions with the potential suppliers, the procurement manager might respond that one supplier's excess capacity is available only if the company commits to buying additional items from that supplier. The moment this information is shared with the master AI orchestrator, it's immediately disseminated to all relevant task-based AI agents and employees. Merchandising is alerted to the implications for the breadth of the company's product assortment and shelf space, and connectors trigger its task-based forecasting agents to take the new information into account and update demand scenarios. Finance is also alerted about the conditional supplier capacity, and connectors automatically trigger task-based finance agents to recalculate the potential impact on working capital and earnings volatility. The final decision on whether to pursue the pull-forward strategy—and on what scale—is made by the senior leaders, such as the heads of supply chain, merchandising, and finance, who now have a fully integrated view of the trade-offs.

The effect is twofold. Speed increases because workflows proceed in parallel and iteration is no longer driven by late surprises. Quality improves because each task incorporates validated inputs from elsewhere in the system.

Notably, the orchestration system prevents task-based agents from working at cross-purposes. Last June data science leaders at Walmart, which is in the process of building such an orchestration system, told us that specialized agents can generate high-quality recommendations and human-in-the-loop actions tailored to their respective domains. However, when those agents operate independently, their outputs may conflict. The orchestration layer helps reconcile those tensions, ensuring that agents' outputs remain aligned with enterprise priorities and contribute to coherent business outcomes.

Agentic AI orchestration thus functions as a shared organizational infrastructure—one that's activated by questions and opportunities raised by leadership. Authority and accountability remain firmly with humans. Enterprise decisions aren't automated; what's automated is the orchestration of intelligence, including the routing of questions, the

integration of insights, and the continuous alignment of local analyses with enterprise objectives.

BUILDING TOWARD AGENTIC ORCHESTRATION

Organizations cannot use AI to break coordination bottlenecks if their underlying decisions are opaque, entangled, and idiosyncratic. So the first thing leaders must do is make sure decision processes are defined and structured well. That effort should proceed from the bottom up, and it won't succeed unless the entire organization commits to it. Only the top leadership can make that happen.

Step 1: Break decisions down into modules. Two companies we studied—a global e-commerce giant and Xsight, an online platform that helps manufacturers sell into overseas markets—found that orchestration improved significantly when decisions were divided into precise, well-bounded tasks. This entails specifying each task's inputs, outputs, constraints,

84

---

Authority and accountability remain firmly with humans. Enterprise decisions aren't automated; what's automated is the orchestration of intelligence.

and objectives. The scope of each task should be narrow enough that it can be easily understood, evaluated, and linked to other tasks. Information and assumptions should be standardized; for example, procurement, logistics, and finance should all work from the same definition of supplier capacity, whether it reflects theoretical production capacity, contractually committed capacity, or realistically achievable capacity.

Step 2: Strengthen task-based agentic AI. Begin applying agentic AI to the tasks that are largely driven by data that AI can access and where the expected returns are the greatest. In every case the collaboration between people and the AI should be structured. The AI should be designed to make its inputs, assumptions, and reasoning visible to the people involved in the task and to actively elicit the relevant knowledge that only they hold. For example, the system might ask the procurement manager whether he had recent supplier conversations that might affect what the model treats as available capacity. On the human side, people must understand what the AI can and cannot see so that they can recognize when circumstances have changed, when assumptions no longer hold, and when they possess knowledge the system has no way of inferring—and provide that information to the AI or, when needed, override the AI. To this end, people must be trained to ask, “Is the AI system missing something I know?” rather than “Do I agree with the AI system?”

Making this work requires a cultural shift. Everyone needs to understand that AI acts as a structured partner in decision-making. Leaders—from the top to the functional-team level—have to reinforce that mindset.

Step 3: Add the master orchestration. Once task-based AI agents and their human collaborators achieve a satisfactory level of performance, your company can take the next step to orchestrate their collective work. Start by coordinating closely related tasks within individual small groups. Build simple connectors linking one task to the next, and develop an initial master AI orchestrator to interface with the employees involved in those tasks. This will support an agile approach to developing the orchestration software—one that emphasizes iterative progress, rapid feedback, collaboration, and adaptability. When the orchestrator matures

(becoming more developed, capable, stable, and reliable) and works effectively for several groups of closely related tasks, it should be extended across more modules.

In multiple industries companies are beginning to build early versions of such infrastructure. At the fintech platform Ramp and at the healthcare technology company Medtronic, core decision processes have been organized into distinct tasks, each governed by a separate orchestrator. Ramp is now developing a single enterprise master AI orchestrator that will route employee queries across the modular systems and engage the appropriate agents as needed.

Step 4: Manage the system’s evolution. Agentic AI orchestration is not a system you build and leave alone. Leaders must continually ask, “Are modules defined narrowly enough for decisions to be orchestrated effectively? Is human-only information being captured in a way that makes it accessible across the system? Are incentives aligned to encourage early disclosure of constraints rather than late-stage escalation? Do employees trust and understand AI well enough to collaborate with it rather than compete against it?” The answers will change as the technology and the business’s capabilities, constraints, and objectives change—and leaders must update roles, governance, and training accordingly. They should see themselves as the architects of a living decision infrastructure—one that improves as the organization and its AI capabilities mature together.

AGENTIC AI ORCHESTRATION should not be viewed as a means to totally automate decision-making. The goal should be to automate the orchestration of intelligence—the ability to route questions, integrate insights, and align local analyses with enterprise objectives in real time. Organizations that build this capability will compound the value of every piece of human knowledge across the enterprise while their rivals are still discovering critical constraints in the third round of alignment meetings. ▼ HBR Reprint R2605E

KRIS JOHNSON FERREIRA is the Edgerley Family Associate Professor of Business Administration at Harvard Business School. JORDAN TONG is the Robert M. Steiner Chair of Business and the chair of the Operations and Information Management Department at the Wisconsin School of Business.

85

---

Don't Underestimate the Power of Customer Referrals

86

ILLUSTRATOR ANTONIO SORTINO

---

New research reveals they're a critical source of low-cost growth.

AUTHORS

Fred Reichheld Bain Fellow, Bain & Company

Jamie Cleghorn Partner, Bain & Company

Wojtek Kokoszka CEO, Mention Me

SALES & MARKETING

87

---

SALES & MARKETING

Just eight years after its launch,

ButcherBox, a subscription-based retailer of premium meat and seafood, rocketed past sales of $600 million. Initially the firm grew by marketing through paid influencers, but over time it increased its dependence on buying new customer leads through digital advertising. Then customer acquisition costs began spiraling as paid influencers and social media and search websites raised their prices. Meanwhile, the quality of the customers produced by the paid channels declined: They made smaller purchases, filled their baskets with a less attractive product mix, and defected at higher rates.

Searching for a new strategy to revitalize the company, ButcherBox's CEO, Mike Salguero, landed on the concept of earned growth, which prioritizes high-quality revenue growth from repeat purchases and referrals. (See "Net Promoter 3.0," HBR, November–December 2021.) Since each customer could consume only so much protein each month, referrals seemed like the bigger growth opportunity. Treating his referred customers as a separate channel, Salguero carefully examined their economics and discovered that they were a gold mine—much more profitable than customers acquired through paid ads. So his marketing team launched an incentivized referral program, creating a new stream of customers whose repeat purchases and word of mouth drove a new wave of growth. With this new focus,

ButcherBox has not only boosted referrals by 57% over the past year but cut its customer acquisition costs in half.

Companies spend an enormous amount of marketing money and management time on social media, digital advertising, sponsored search, promotional schemes, paid influencers, search engine optimization, and curated online reviews. Yet for most companies, the best customers still come through referrals—as Bain has confirmed in an analysis of data on more than 10 million individuals who participated in referral programs implemented by Mention Me, a customer advocacy platform. (One of the authors of this article, Wojtek, is its CEO.) Across a range of industries from fashion to food, health and beauty, financial services, travel, and telecommunications, we found that on average, while only 20% of new customers were referred, they generated more than 70% of all new-customer profits. (See the exhibit "Where Your New-Customer Profits Are Really Coming From.")

The outsize profit impact of referrals results from several advantages: Referred customers cost less to acquire, stay longer, and recommend your company to their own friends and family more often. And they buy bigger baskets of purchases and a more profitable mix of products.

Referrals also have a disproportionate impact on revenue growth. We already know from dozens of Net Promoter consulting projects that on average, customers who are promoters spend 1.6 times more than customers who are passives. (Promoters rate a company as a 9 or 10 on the survey question "On a scale of 0 to 10, how likely would you be to recommend us to a friend?" Passives rate it a 7 or 8.) Over the years, however, survey score inflation has swollen the promoter category to half of all respondents. That has muddled the differences between genuinely passionate advocates and people who are promoters in name only—customers who are

IDEA IN BRIEF

THE IDEA

Referrals are the most reliable source of profitable growth, yet most companies fail to measure or manage them. Customers acquired through recommendations cost less, stay longer, spend more, and generate additional high-value customers.

WHY IT MATTERS

While only about one-fifth of new customers arrive via referrals, they account for the vast majority of profits. Traditional marketing metrics obscure this effect by giving too much credit to paid channels and ignoring word of mouth.

WHAT TO DO

Track why every new customer came aboard, quantify referral economics, and identify, understand, and delight more true promoters. Invest in experiences that earn recommendations and campaigns that activate true promoters. Align incentives around referral generation to unlock compounding, lower-cost growth.

88

---

Referred customers cost less to acquire, stay longer, and recommend your company to their own friends and family more often. And they buy bigger baskets of purchases.

merely satisfied but not enthusiastic enough to actively recommend a brand. Our study of Mention Me's data (covering both incentivized and organic referrals) revealed that only 15% of all customers are true promoters, meaning they generate at least one new customer by making a referral. These true promoters generated almost three times more lifetime revenues (combining their own purchases with purchases by their direct and downstream referrals) than passives did. And the subset of true promoters who referred multiple new customers (we call them superpromoters) generated five times more revenues than passives did.

If companies examine only the impact of direct referrals, they will substantially undervalue the compounding growth produced by true promoters—and especially superpromoters. The cascade of customers referring friends, who in turn refer their friends and so on, explains why modest changes in referral rates drive exponential increases in growth.

If referrals are so powerful, why have so many executives overlooked them for so long? Part of the problem is visibility. Most companies don't track referrals in a systematic way, so referred customers get lumped together with customers acquired through other means. As a result, accounting systems and marketing attribution models give credit to campaigns that merely captured customers who had already decided to buy because of a friend's recommendation. That distortion makes paid acquisition appear more effective than it really is while masking the true economics of referral-driven growth.

Many leaders also focus on what's easiest to measure and manage, such as advertising spending and its correlation to new-customer volumes, rather than which experiences evoke delight or the social dynamics that trigger recommendations. And because traditional metrics often don't distinguish between short-term (ephemeral) revenue boosts from new-customer acquisition and sustainable revenue growth from long-term advocacy, executives may not see how well referrals drive sustainable growth and profitability.

In this article we'll look more closely at the economics behind referred customers and explain why CEOs should direct their teams to begin measuring and managing referral behavior more deliberately. We'll demonstrate how referral rates are a predictor of financial success and thus should be scrutinized not just by marketing executives but by the entire C-suite and the board of directors. We'll also offer practical ways to identify true promoters and design systems that turn customer advocacy into a reliable engine of growth.

THE POSITIVE EFFECTS ON ECONOMICS

Referrals can dramatically improve a company's cash flow. We believe that they explain the apparently anomalous cash generation we've observed in Net Promoter superstars like Enterprise Rent-A-Car, Chick-fil-A, IKEA, and Apple. Enterprise has grown from a small family leasing business in St. Louis to become the largest car-rental firm on earth, and it has never had to issue equity to pay for its fleet of more than 2.4 million vehicles. Chick-fil-A grew its systemwide revenues to almost $23 billion while building more than 3,000 company-owned stores, also without tapping equity markets. IKEA likewise has become the world's largest furniture retailer without ever issuing equity or debt. And Apple, a pioneering practitioner of the Net Promoter System, has become such a cash-generating juggernaut that it has repurchased more than $800 billion worth of shares since 2013.

These companies focus on delivering remarkable products and services that inspire customer referrals, so they don't need to rely as heavily as their competitors do on expensive advertising or price promotions to acquire new customers. How large a cash flow advantage can this approach provide? To find out, we compared the NPS leader to a peer that was an NPS laggard in five sectors for which public data on both was available. Those leaders were Hermès, Costco, IKEA, Apple, and the restaurant chain Texas Roadhouse.

We looked at sales, general, and administrative (SG&A) expenses as a percentage of revenues for each pair and found that on average the leader's SG&A percentage was half that of its follower—11% versus 22% of revenues. That means the NPS leaders generated 11 cents more cash flow on every dollar of sales than the NPS laggards did.

89

---

NPS leaders have lower costs because they don't spend a lot of money acquiring customers with unclear long-term revenue potential and little referral generation. (If only 20% of customers at a typical firm come through referrals, then the other 80% should be viewed with a degree of skepticism.) Efficient growth depends on recommendations from existing happy customers.

Earning referrals takes hard work and innovation. Executives may be tempted to take shortcuts with clever customer acquisition schemes, but the benefits rarely justify the costs unless many of those new customers mature into true promoters. This logic is especially relevant given the rise of AI-powered shopping bots. They're designed to be immune to search engine optimization, digital marketing, advertising, and paid influencers. The best bots increasingly favor legitimate reviews and recommendations from customers. As more and more interactions flow through them, a focus on earning referrals will become even more essential to attracting high-quality prospects.

KICK-START YOUR REFERRAL ENGINE

Companies that take referrals seriously build systems to identify which customers are recommending their brand and why. Doing this well requires more than surveys or marketing intuition. It requires operational processes that track referral flows and hold teams accountable for turning satisfied customers into active promoters. In this section we'll detail some of the key actions your company must take to benefit from the referral effect.

Identify new referred customers. You need to find a way to systematically pinpoint these people. One company whose example you could follow is CertaPro, the largest

residential painting company in North America, which has been measuring both NPS and referral leads for more than a decade. It asks all new customer leads how they heard about the company and records their answers in its customer relationship management system. CertaPro's lead inquiry process won't advance until the prospective customer responds, because the company considers that information as vital as a customer's address and credit card number.

Mention Me developed an online solution that makes it easy for new customers arriving via a website to select one of the six most frequently cited reasons they chose the brand or to offer another reason. As part of the process, companies can find out who made referrals by asking customers for the contact information for whoever recommended them so that they can send those people a message of recognition or a gift. Such individuals are the true promoters who make up a company's strategic core, so it's important to identify them and learn more about them—specifically what triggered their referrals.

Identifying all referred customers on first contact allows you to determine your referral ratio (the portion of new customers who are referred), which you can then use as a baseline for improvement. As we've noted, the average referral ratio for Mention Me clients is 20%, but it ranges from single digits up to more than 50%. At one extreme is Costco, which spends very little on advertising and promotions to acquire new members; nearly all its new customers are earned through referrals.

Low-tech solutions like CertaPro's can help you measure referral flows at first. Once you have a system in place, you can keep experimenting and refining your processes to reduce friction and make your data even more reliable. Later you can integrate customer feedback and loyalty management systems with your central customer-records database.

90 Harvard Business Review September–October 2026

---

Quantify your referral economics. To understand how much you can afford to invest in earning referrals and in systems to measure and optimize them, you must be able to determine their value and impact on profits and growth. When ButcherBox carefully allocated advertising, promotion, and onboarding costs to its referred customers, it discovered that the acquisition costs for them were a small fraction of the costs for new customers generated through other channels. The finance team then quantified the post-acquisition-cost value of referred customers, incorporating purchasing patterns, average order size, product mix and margin, frequency, customer retention, and profits from the new customers they subsequently referred. These calculations revealed the full value of earning more referrals.

Another company that quantifies its referral economics is Intuitive Health, which partners with hospital networks to build and operate facilities that provide urgent care and emergency room services under one roof. Its facilities treated more than one million patients in 2025. The company's annual revenues have compounded by more than 30% annually while its profits have been so impressive that its recent recapitalization provided original shareholders a 900% return.

Intuitive tracks the effects of referrals rigorously by region. In the Dallas–Fort Worth area, it discovered, referred customers use more services, boosting revenues, and have profit margins four times as high as those of other customers. Intuitive has also found that its focus on referral generation has reduced customer acquisition costs companywide from $57 in 2015 to just over $20 today. At every clinic, referral volumes figure in the business plan and are treated as a key performance indicator for managers.

Learn how to delight core customers. Creating true promoters and activating their referrals results not from

clever marketing but from repeatedly delivering fresh experiences that are so remarkable that customers naturally recommend a product or service to their friends and family. To achieve this, companies need to develop a granular understanding of exactly what will spark a referral.

Intuitive Health uses its NPS feedback process to learn what delights customers—and what dismays them. Local clinic managers strive to close the loop and probe for root causes of any failure to delight a customer within one or two days. They call not only dissatisfied detractors but every customer who gives a score less than 9 out of 10. Its surveys don't just identify sources of delight; they seek to calibrate key performance thresholds. For instance, Intuitive found that one critical indicator is the time it takes for someone to check in, wait, get treated, and walk back out the door. Operational turnaround time—the share of patients who get in and out the door in less than an hour—then became an important target.

Incentivize referrals strategically. To manage and reap the full suite of economic advantages, most companies should implement an incentivized referral program. But they should be sure to avoid the mistakes firms commonly make with them. Many programs never get integrated with financial systems that track the economics of referred customers. They fail to identify the true promoters who made the referrals—and never probe for root causes. Indeed, referral incentives are treated as a cheap marketing tactic to acquire new customers instead of a sophisticated way to drive profitable growth.

Programs can fall short in several other ways. For instance, if the incentive to give referrals is too large, people may be motivated to act through self-interest rather than by the desire to make a friend happy. Sometimes the incentives are misdirected or rewards are given in a way that blunts their

91

---

impact. For instance, if a business travel program posts referral credits to an executive's account, the executive may never even notice because an assistant processes the invoices.

Companies should experiment to find the right mix of incentives, membership, rewards, and recognition that will motivate true promoters and then should quantify the cascading chain of referrals they initiate. The most effective rewards tend to be modest and transparent to both referrer and referred—large enough to nudge referrers to use the system but not so generous that they motivate referrals absent the genuine enthusiasm earned by a valuable experience. Congstar, Deutsche Telekom's value brand in mobile telephony, provides a good example of this. Its Freundeskreis (circle of friends) membership offers 10% discounts to the referring customer and to the newly referred customer for as long as both remain active customers.

ButcherBox has refined its referral program to focus on its most loyal customers. It emails them a code they can forward to a friend who can use it to get a free box of products valued at $170. The friend must pay $20 for shipping, but that helps focus the program on legitimate prospects. The program is reinforced by sending the original customer an additional code for a free box to share with another friend if the first free-box recipient becomes a paying subscriber. This motivates the best customers to keep introducing more friends to the brand—as long as their referrals turn into customers.

Referral tracking can uncover hidden customer value, as the experiences of Bloom & Wild, a Mention Me client, show. Founded in 2013, the company has grown into Europe's largest online flower and gift platform, with revenues exceeding $150 million. Its data reveals that even modest customers can generate outsize downstream impact through referrals.

Consider one customer, Sally. Over several years she purchased only $112 in products. Yet she directly referred several customers, who collectively spent $3,015. Those customers then referred others, creating additional waves of buyers. In total, Sally's referral chain generated $8,793 in revenues, about 78 times the value of her own purchases. And that figure is likely higher when organic referrals (not solicited with incentives) are considered. Rather than treating low-spending customers as unimportant, Bloom & Wild views them as potential referral catalysts, having learned that there are many Sallys in the customer base.

Where Your New-Customer Profits Are Really Coming From

Most companies assume referrals come from a broad base of happy customers. But the data suggests something much more focused and more powerful is happening. A small slice of customers, the "true promoters," generate a disproportionate share of new-customer profits. Even though true promoters make up just 15% of the customer base, their referrals generate 72% of new-customer profits over three years. And the superior retention rate of referred customers means they generate an even larger share of lifetime profits.

Note: We measured profits from each cohort of new customers over the three years subsequent to their initial purchase. This results in a very conservative estimate of lifetime value but unlocks the statistical benefits of a much larger sample size. Source: NPS Prism, Mention Me

The company also encourages gift recipients to join the program. Each bouquet includes a QR code that lets the recipient thank the sender and earn loyalty points. Today gift recipients' referrals generate 40% of new customers, and standard referrals add another 15%.

Bloom & Wild's system uncovers true promoters so that the company can learn what delights them enough to trigger a referral. Analysts at the company can perform an accurate cost-benefit analysis on initiatives to wow customers, unleashing further creativity and investment.

Develop more-sophisticated referral-tracking capabilities over time. Even when companies offer incentives for referrals, most referrals still happen organically. Among Mention Me clients, incentive programs account for only about 7% of referrals, on average, with a range from 2% to 21%. The more deeply companies understand what's driving organic referrals, the better equipped they'll be to increase them. One key to doing that: Don't stop analyzing referrals after the first purchase.

Although referrals are typically considered a mode of customer acquisition, they can also prompt existing

92

---

customers to purchase additional product lines. Understanding what triggers those sales can help companies increase them.

Companies are also realizing that a referral doesn't always spark an instantaneous purchase. To fully understand referral behavior, they need to get better at tracking the chain of events that leads from a referral to a sale, even if the sale takes some time. Some companies now track this at the product level. For example, Mention Me has developed a method of including a tracker on clients' website product pages. When a customer shares a link to a product with a friend through a text, WhatsApp, or an email, and the friend clicks on it, the system records that incoming click as a referral and captures who made it. If the friend makes a purchase, the tracker connects that person's shopping and referral activity across website visits. Early results suggest that these referrals are highly valuable. Clicks from referral links convert to purchases at more than three times the rate of paid ads and tend to generate more-profitable sales.

HOW THE REFERRAL EFFECT WORKS IN B2B

While one individual typically decides what to purchase in B2C transactions, the number of people on a B2B buying committee can reach a dozen or more. This group is often influenced by the laws of social psychology, as Mimi Turner and Jann Schwarz of LinkedIn have shown. Their 2025 "buyability driver" research found that when B2B buyers are choosing among a short list of vendors that meet the basic requirements for quality, the leading factors are personal experience with a vendor, references from similar customers, and recommendations from colleagues who have direct experience with the vendor. Personal, trusted recommendations carry far more weight than even price, with 80% of buyers preferring the recommended offering versus the cheaper offering.

Despite the vital role played by customer recommendations, few B2B companies carefully track which individuals provided winning references or probe what motivated them to do so. Meanwhile, customer success teams tend to sit deep in the organization and focus on client retention, with little visibility on which referrals are bringing in new customers.

B2B companies should rigorously track which references and referrals resulted in winning bids, but their game plan for optimizing referral flows will be different. Instead of rewarding a customer for a referral, they should recognize and reward the internal team or individual who earned it. It's also optimal to integrate the reward program with reliable measurement systems used to manage internal processes, such as the compensation system. If bonuses depend on referral and reference data, the data will receive scrutiny that ensures its accuracy.

SALES & MARKETING

In complex sales processes involving multiple references and referrals, it makes sense for an executive to interview client decision-makers to understand how to allocate credit appropriately. Rewarding the asset builders will encourage the building of more assets.

Many executives personally scrutinize failed bake-offs (situations in which their company lost a sale to another). More should examine account wins to understand the role played by promoters who make referrals. Consider how much time and effort the finance staff now spends quantifying balance sheet items such as depreciation and goodwill, while ignoring the far more precious asset of true promoters. This is a huge, untapped opportunity.

IN THEIR QUEST for growth, too many leaders look outward, searching for clever ways to attract new customers. They should instead be looking inward to revitalize their referral engine. True promoters constitute the strategic core of a brand, providing a self-fueling growth engine that attracts the right new customers, who produce higher-quality revenues. They give recommendations because they want to make a friend's life better. They also know their friends' tastes, preferences, and priorities. That's why the customers acquired through referrals usually fit better, buy more, stay longer, and become true promoters themselves, further accelerating growth.

Companies like ButcherBox, Intuitive Health, and Costco understand the vital role played by referrals and work hard to understand how they're earned—and how they spread. They have reallocated advertising and marketing dollars toward delighting core customers to turn them into true promoters. The most advanced firms set clear targets for referral volume. They identify their strongest advocates, experiment with programs that encourage referrals, and invest in systems that reveal how their referrals multiply. As a result, they grow more efficiently than their competitors do and can reinvest the savings to deliver more-remarkable customer experiences. ▼

HBR Reprint R2605F

FRED REICHHELD is a Bain Fellow and an advisory partner at Bain & Company. JAMIE CLEGHORN is a partner at Bain & Company and the head of its customer practice. WOJTEK KOKOSZKA is the CEO of Mention Me.

93

---

It evolves over time. Here's how to navigate each phase successfully.

AUTHORS

Claudius A. Hildebrand Consultant, Spencer Stuart

Douglas L. Peterson Former CEO, S&P Global

94 Harvard Business Review September–October 2026

ILLUSTRATOR CARL GODFREY

---

95

---

LEADERSHIP

It's

A CENTRAL QUANDARY FOR CEOs: How can you deliver transformational leadership when your ability to execute depends on a group of people you don't hire, can't fire, and have to influence without

direct authority? That is, your board of directors.

Navigating the CEO-board dynamic is one of the most critical and underappreciated challenges of executive leadership. Given the high stakes, it's surprising how little attention has been paid to the way CEOs can manage it over time—as the CEO grows in the role, as the composition of the board of directors evolves, and as new business challenges arise.

S&P 500 CEOs serve for an average of nine years, a period during which they face sweeping changes in both their businesses and their relationships with directors. Yet most new CEOs are caught off guard by how much time and attention the board relationship demands. To be successful at it CEOs have to recognize that working with the board will involve a shifting approach during the course of their tenure: gaining trust in the early days, when they're under the microscope; developing the board into a strategic ally as the relationship matures; maintaining a posture toward the board that encourages debate and engagement; and ultimately shaping the board's future leadership.

There's a significant price to pay for getting it wrong. Allow the board to become too involved in short-term performance and you'll spend your tenure fending off counterproductive strategy pivots; fail to engage the board sufficiently and you'll lack the oversight needed to guard against complacency and excessive risk.

For years we've been thinking about how CEO-board dynamics play out. One of us (Doug) served as the CEO of S&P Global for 11 years, and the other (Claudius) is a veteran leadership adviser who has conducted research analyzing the performance of more than 2,000 S&P 500 CEOs over two decades. In this article, drawing on our extensive knowledge of CEO and board performance, we'll lay out a four-stage framework designed to help CEOs build a productive partnership with the board that creates the healthy tension essential for effective governance and strong company performance.

Each stage requires pivotal shifts in a CEO's thinking and behaviors.

1 Inheriting the Board

New CEOs often are unsure about what to prioritize. Feeling pressure to make a mark right away, many think they don't have time for the "soft" imperative of building a relationship with the board. That's a mistake, because it means they're forgoing an early opportunity to develop and accumulate social capital, which they'll need to rely on later.

When assuming the new role, CEOs tend to know little about the board's internal dynamics or individual members' views. CEOs hired from outside the company may barely know the directors beyond their interactions during the recruiting process. Even those promoted from within the

IDEA IN BRIEF

THE PROBLEM

CEOs must lead transformations while relying on boards they don't control. Many underestimate how much their relationship with the board must evolve and how mismanaging it can yield poor strategy, weak oversight, and even a premature exit.

THE SOLUTION

Adopt a four-stage approach to CEO-board collaboration: (1) Build trust early by understanding board dynamics. (2) Shape the partnership by influencing composition and engagement. (3) Fight complacency as trust deepens. (4) Focus on succession and long-term governance.

THE PAYOFF

CEOs who adopt this approach to their board relationships build productive tension, stronger governance, and better long-term performance. The result is a board that's a strategic asset rather than a constraint.

96

Harvard Business Review September–October 2025

---

97

---

Navigating the CEO-board dynamic is one of the most critical and underappreciated challenges of executive leadership, so it's surprising how little attention has been paid to it.

company (as the majority of CEOs are) typically have dealt with the board only in carefully orchestrated environments. In short, new CEOs often don't know how decisions get made on the board and where influence lies.

Adding to the challenge for incoming CEOs, today it's increasingly common for their predecessors to remain on the board of directors as executive chairs. In more than a third of the U.S. appointments made since 2020, the outgoing CEO became the executive chair—up from only about a quarter in 2015. That can create an awkward power dynamic for the newcomer. "Whenever I took a stance in the boardroom in the early days," recalled one CEO we interviewed, "heads would inadvertently turn to observe my predecessor's reaction."

In their first couple of years CEOs can take several actions to address directors' skepticism and build trust:

Map the board's hidden dynamics. Chief executives can establish authentic relationships early on by replacing Zoom calls with individual face-to-face meetings on directors' home turf. These get-togethers—at a favorite restaurant, say, or some other comfortable location—often yield insights about directors' personal styles, priorities, and perspectives. Having one-on-one meetings with directors is not new advice, of course, but we've seen many new CEOs either treat these meetings as check-the-box exercises or take shortcuts with them—by arranging meetings with two directors at a time, for example. CEOs who engage in such practices prevent themselves from building strong individual relationships with directors. They should instead recognize that early one-on-one meetings facilitate intensive "discovery work." Almost every question is fair game and yields valuable intelligence that can later be used to gauge and react to directors' comments in more-formal settings. As one CEO told us, "They all have a reason they joined the board, and it's good to know their motivations. This is the time to understand what each board member brings to the dialogue."

Honor the past while asserting independence. A new CEO can set the tone for a productive working relationship by balancing respect for the former CEO's legacy and existing relationships while also asserting independent leadership. In his initial years as CEO, Doug developed a partnership with the lead director that helped him navigate complex board dynamics and, critically, clarify any

ambiguity in the roles of the nonexecutive chair, the lead director, and the CEO. He also made the most of a preexisting relationship with one respected director, relying on that person as both a private sounding board and an advocate in the boardroom. In the best situations the departing CEO can also be an ally. One outgoing CEO Claudius worked with, for example, made a presentation to the board in which he summed up his weaknesses and pinpointed the areas in which he felt he had failed. His candor gave his successor some room to maneuver early on in developing his stature and building trust.

Define the rules of engagement. The start of a new CEO's tenure is the most natural time to define how the board and CEO will work together. The failure to set operating principles can be costly. In one cautionary tale, a CEO we're familiar with butted heads with the board over its requests for more information about corporate strategy and M&A opportunities, a conflict that eroded trust between the board and the CEO and contributed to the CEO's eventual departure. The decline in the relationship occurred even though the company had outperformed the S&P 500 under the CEO's leadership—a reminder that even high performance may not insulate a CEO from a fraught relationship with the board.

The goal should be to establish processes that encourage transparency and foster trust but also acknowledge the board's oversight responsibility and the CEO's need for support. One critical norm to establish right away concerns how much detail the board needs versus wants. Eager to demonstrate command of the business, many new CEOs overprepare, burying directors in operational minutiae when in fact they need strategic insight. One CEO we know learned that lesson in his first board meeting when he proudly presented a 100-page deck, only to have directors ask for higher-level strategic framing instead. He was able to recover because he made that misstep early, when both sides were still calibrating expectations.

Points of discussion and alignment should include what the organization's vision and mission are, what the CEO will be expected to deliver, the board's role in strategy and governance (not operations), and how everyone will communicate and interact—including the level of detail appropriate for board materials.

98

---

LEADERSHIP

2 Shaping the Partnership

After weathering several earnings cycles, navigating investor scrutiny, and building credibility, the CEO has typically earned enough trust to begin shaping the board's composition and engagement style. This period, usually starting in the second or third year, is the crucible where leadership is forged. The outgoing CEO, if still present, has by now stepped back. Executive chair arrangements typically run one to two years, and Claudius's research shows that in the vast majority of transitions the formal overlap period has concluded by year three. With that dynamic resolved, the new CEO's relationships with directors will have matured, which means the CEO can exert a growing influence on board discussions. This is also a time when the CEO may seek to pursue the chair role or redesign board committees to better support the company's priorities. During this stage CEOs must listen with humility, build consensus for change, and develop the courage to challenge the status quo. It's a delicate moment: Advocating for too much change too soon can be destabilizing, but letting board membership stagnate can hinder the cultivation of new ideas.

To successfully manage this stage CEOs should build lasting alliances by doing three key things:

Refresh the board strategically. Boards don't overhaul themselves often enough. In 2025, for example, the incoming class of S&P 500 directors represented only 7% of all board members, and almost half of S&P 500 boards had no change to their composition at all. The result is predictable: In a recent survey less than a third of CEOs told us that their board had the expertise needed to address today's business challenges.

Although nominating and governance committees drive membership changes and succession planning, during this stage CEOs can—and should—ask to provide more input into the process. Nobody can better communicate to the board how the business is shifting and what kinds of experience and skills the board will require to adapt.

Consider how Doug began remaking S&P Global's board in his second year as CEO. By that point the company was quickly transitioning away from its media and print

publishing roots and transforming into a modern analytics business that served financial markets, and he knew that the boardroom needed voices with expertise in technology, capital markets, operations, and risk management. There were several high-quality, long-serving directors when he became CEO, but they didn't have the background in the core areas for the future he envisioned. So when Doug broached the topic of refreshing the board, he did so in a way that drew parallels to the company's natural evolution as it exited its legacy businesses and invested in new growth markets.

The second stage is also a time when some CEOs feel they've earned the right to become the board chair. Our research has found that CEOs are most likely to be appointed chair before their fourth year and very unlikely to thereafter. But how CEOs request that role matters, as being too aggressive can backfire. One CEO we know learned that the hard way: He pushed intensely for the chair title sooner than the board was comfortable with, damaging his relationships with directors, who then held off making the change for another two years.

Manage the full spectrum of directors. During this stage CEOs need to develop techniques for facilitating robust dialogue in the boardroom. Learning how to manage people with different temperaments—from passive members to those who overstep their governance-and-oversight role—is essential. Making sure every voice is heard can be a tough balancing act, though. How does a CEO engage directors who volunteer less time and perspective while also moderating others who are too involved or provide too much advice? (For more on this topic, see "Managing Difficult Directors," HBR, May–June 2026.)

A strong lead director or board chair is key to success in this area. Many high-performing CEOs schedule a follow-up session with the lead director soon after board meetings and executive sessions in order to take the board's collective temperature—understanding which topics resonated, where directors had unspoken concerns, and which discussions required follow-through. To foster an open dialogue with the lead director, CEOs should signal that they value transparency about directors' questions and concerns.

CEOs will also develop their own particular approaches. During S&P Global's yearly strategic board meetings, for example, Doug liked to have directors and management

99

---

LEADERSHIP

break into working groups to explore different themes. These smaller groups created more opportunities for people to talk and share their viewpoints. Doug also found it helpful to hold what he called a “lightning round” at the end of strategic board meetings, during which he went around the table asking for concise input; everyone had a chance to participate, and diverse voices could be heard. Both approaches helped engage directors.

Orchestrate influence. There’s a point when new CEOs realize how much board activity happens during committee meetings in which the chief executives don’t take part. (Executive sessions where directors meet without the CEO present can feel like black boxes, especially early on.) This forces CEOs to recognize that they need a mechanism for staying informed and entrusting their executive team to interact with board committees.

We’ve observed that many CEOs, particularly early in their tenures, want to exert too much control over interactions with the board. They overscript their teams, and in their own relations with directors they focus on showcasing operational command rather than facilitating strategic dialogue. The CEO who presented the 100-page deck in his first board meeting learned to take a better approach: He started exploring how best to engage the board and eventually pivoted to providing a four-page memo before every meeting, highlighting his top concerns and the strategic questions for which he needed directors’ input. This shift resulted in robust discussions and much more productive meetings. The lesson: Board engagement isn’t about demonstrating how much you know. It’s about focusing the attention of directors on where their insight can add the most value.

3 Navigating the Trust-Oversight Paradox

As the board’s trust in the CEO deepens, hidden risks can emerge, including overconfidence on the CEO’s part and risk aversion on the board’s part. Call it the trust-oversight paradox: Although CEOs typically welcome the shift from intense oversight that comes during this stage, it can signal a weakening of constructive pushback from the board. Additionally, by this point new directors have gradually joined the board and power has transferred to the CEO,

100

---

CEOs are most likely to be appointed chair before their fourth year and unlikely to thereafter. But how CEOs request that role matters, as being too aggressive can backfire.

101

---

A board's reluctance to raise the topic of transition may leave the CEO in the driver's seat on succession—a situation not always in the best interest of the company.

who now has a harder time hearing dissenting voices. Several CEOs shared with us the feeling of being trapped in an echo chamber. The result is often complacency on all sides, with an unfortunate bias toward the status quo.

Such dynamics can pose real risks to the business. Two-thirds of the CEOs that we studied in our CEO-life-cycle research, for example, performed worse in their second five years than in their first five—but their boards were nevertheless much less likely to act on poor performance during the second five years. Essentially, boards seem to be unresponsive to underperformance after the sixth year of a CEO's tenure. Claudius observed that dynamic when working on succession planning with a board that could not imagine that anyone but the long-tenured CEO could lead the company, even though the CEO had presided over six years of bottom-quartile performance.

Here are key actions that CEOs need to take at this stage:

Combat complacency. CEOs should shake things up if social cohesion comes at the expense of rigorous governance, shorter board meetings produce less debate, and the number of questions that challenge management decreases. That's what one Fortune 100 CEO did when he recognized the need to reinvigorate board oversight in his sixth year. After noticing that board meetings had become increasingly perfunctory, he introduced "challenge sessions," during which he rotated committee assignments to bring in fresh perspectives, created formal roles for skeptical questioning in major decisions, and had external experts critique strategy.

Keep the board current. At this stage CEOs have generally established themselves as leaders of the board, with or without the chair title. Even so, they're not always able to change the board's composition as fast as they would like to. They can help build its skill set in other ways, however, including by prioritizing ongoing education—on governance issues, the competitive landscape, the regulatory environment, and changes to the business and its markets. They can also encourage field visits. Doug did both. With an eye to AI's long-term impact, he introduced AI teach-in sessions for the board and started including AI-related discussions at every strategic board meeting. He also made sure his directors had opportunities to visit frontline offices, speak with employees in the field, and get hands-on product demos.

Engineer ways to hear hard truths. As their faith in the CEO grows and as the CEO exerts more control, directors may be less likely to publicly question the CEO on plans or underlying assumptions, so major issues could go unraised in board meetings. Establishing channels for board feedback that don't demand open dissent—such as executive sessions held without the CEO—can help ensure that those issues are raised and can give directors a voice without requiring them to challenge the CEO in front of their peers. Paradoxically, the CEO's success in building trust now entails deliberately making space for directors to test that trust.

4 Paying It Forward

The final years of a CEO's tenure bring a dual challenge: maintaining strategic momentum while setting up the future leadership of the board and the company. CEOs at this point may be experiencing a complex emotional tug-of-war between driving today's business performance and looking to the future—both theirs and the organization's. It's a time when CEOs often develop a heightened awareness of their professional mortality. They begin to wonder about the right time to step down, how they should see their role once they exit (Stay on as executive chair? Make a clean break?), and whether the board and the organization are ready for the next leader. They may grow increasingly concerned about the degree of board support for their slate of succession candidates and about whether the board has the right people around the table for the future. This stage requires a mental shift from personal achievement to institutional legacy—and with that shift comes a focus on paying things forward and some deeper reflection on long-term impact beyond financial metrics.

CEOs can successfully navigate this stage by relying on three main approaches:

Break the succession silence. In our research we found that boards are unlikely to broach the topic of a transition with a long-serving, high-performing CEO. In fact, boards often want the CEO to stay as long as possible, and their reluctance to raise the topic of transition may leave the CEO in the driver's seat on succession—a situation not always in the best interest of the company. In one situation

102 Harvard Business Review September–October 2026

---

that Claudius observed, the board stood by as the CEO made himself indispensable by torpedoing the succession-planning process. Unwilling to engage in a dialogue with the CEO about a transition, the company's directors made a move only after an activist investor intervened by demanding an external hire, which scuttled the pathway for a promising internal successor.

More often the CEO can serve as a catalyst for discussions of succession planning with the board, including by helping to define what's needed in a future CEO, overseeing the development of internal candidates, and sharing personal plans and milestones. During Doug's tenure at S&P Global, he used the completion of a major merger as an inflection point for considering succession. The directors spent months defining the qualities needed in the next CEO and delineating the experiences and skills they desired. Those conversations were particularly effective because they didn't come out of the blue; they built on a long-running dialogue about succession planning that included making sure the board got to know the entire executive team.

Shape the board for your successor. As chief executives approach the end of their terms, their influence on board composition is at its peak, and how it's wielded can set the organization's trajectory for years to come. An outgoing CEO has a unique opportunity to use credibility and relationships to champion thoughtful renewal: The CEO can advocate for the addition of directors whose skills and perspectives align with the company's evolving strategy, encourage a healthy mix of tenures, and support robust succession planning for board leadership roles. By prioritizing a diversity of experiences, strategic expertise, and a culture of constructive challenge, the CEO can help create the conditions for ongoing success, ensuring that the board remains a true partner to the next leader and not just a legacy artifact. In this way the departing CEO's final act is about more than just continuity. It's about future-proofing governance and empowering the board to guide the organization through its next era.

One outgoing CEO Claudius worked with began thinking about those things two years before her planned exit. As her transition timeline became more concrete, she realized she had the chance to support the continued performance of the company—her legacy—through the makeup of the board. So she looked at it through a lens that extended beyond her tenure, and she learned that the lead director was planning to leave shortly after she was. Together they designed a longer buffer between the two departures in order to maintain stability and provide support for the new CEO.

Let go of control. For many CEOs one of the most difficult parts of the transition is ceding control of the process,

LEADERSHIP

especially when they've diligently developed internal succession candidates and worked closely with the board on a succession strategy. Once a changeover is imminent, it's the board's job to select a successor, and the CEO must step back and allow the board to do that work.

The transition between CEOs can be a challenging time for departing CEOs, who can feel a loss of power and prestige as the board and the organization pivot their attention to the new leader. As their tenures wind down, exiting CEOs have to confront the question of their post-CEO role. We've seen outgoing CEOs lean in to this unique period by visibly supporting their successors, helping raise the new leaders' profiles with the board and other stakeholders. That can include setting up a process for joint meetings with clients to hand over relationships, connecting with employees, and scheduling introductions to shareholders, regulators, and policymakers. That kind of approach increases the board's confidence in the incoming CEO and helps rally the organization around the new chief executive by making the hand-off visible.

In cases where the outgoing CEO remains with the company as an adviser, a board member, or an executive chair, it's important to define the timeline (shorter is better) and to remain in the background as much as possible. Ideally, the departing CEO provides counsel and support privately, staying out of the spotlight to avoid the risk of undermining the new CEO's authority.

THE DIFFERENCE BETWEEN CEOs who thrive and those who merely survive often comes down to one factor: how deliberately they develop a robust relationship with the board. Successful CEOs understand that the board is not a static oversight body; it's a dynamic force that can either propel or impede the company's progress. The question isn't whether your relationship with the board will evolve, because it will. The question is whether you'll shape that evolution intentionally or simply let it happen. ▼

HBR Reprint R2605G

CLAUDIUS A. HILDEBRAND is a consultant at Spencer Stuart and the coauthor of The Life Cycle of a CEO (PublicAffairs, 2024). DOUGLAS L. PETERSON was the CEO of S&P Global from 2013 to 2024.

103

---

104 Harvard Business Review September–October 2026

---

FINANCE & INVESTING

Most firms get capital allocation wrong. Here's how to do it right.

AUTHORS

Paul Blase Principal, PwC

Paul Leinwand Principal, PwC

ARTISTS RYAN KOOPMANS & ALICE WEXELL

105

---

FINANCE & INVESTING

ABOUT THE ART

Photographer Ryan Koopmans and digital artist Alice Wexell integrate digital foliage into images of abandoned buildings, blending the concepts of structure and growth.

Big businesses have an investment problem. While the U.S. GDP grew by an average of 3% annually from 2014 to 2023, largely driven by technological innovation, the annual investment rate of medium to large American corporations declined by a median 16% in that time period. And this pattern is not unique to the United States.

A 2025 OECD paper drawing on both national accounts and firm-level data across 17 advanced economies found that real business investment is roughly 23% below its pre-financial-crisis level on a weighted average basis.

This gap suggests that many companies are compromising their ability to grow over the long term. For some, it reflects pressure to return cash to shareholders. For others, it reflects a risk-averse investment philosophy, a vacuum of good ideas, or complex processes that prevent firms from properly allocating capital to a compelling portfolio of initiatives. No matter what the reason is, companies have to invest to grow—and they need to know how much to invest.

106 Harvard Business Review September–October 2026

---

We discovered that there's an "investment sweet spot"—a ballpark answer to the important question of how much companies should be investing in growth.

To help them with that calculation, we and our colleague Aaron Reeves analyzed the financial reports and performance of 2,900 U.S. public companies over 10 years (2014–2023). We discovered that there's an "investment sweet spot"—a ballpark answer to the important question of how much companies should be investing in growth. We also found that 75% of companies were underinvesting or overinvesting, resulting in valuation multiples that were anywhere from 20% to 70% lower than those of the firms that hit their sweet spots.

In the following pages we'll explain how we quantify the sweet spot and present findings that will help firms understand how much they should invest for growth. We'll then draw on our experience advising companies on capital allocation to offer executives a process they can use to identify and manage their most promising growth projects.

How Much Should You Be Investing?

The 2,900 companies we looked at were significant players in their industries. All had market capitalizations greater than $50 million, and together they spanned 150 business domains (as defined by industry codes). We began by examining their value-creation performance, calculating the ratio of their enterprise value to their revenue (EV/R) over 10 years. Then we compared the companies on two dimensions of performance that are a central focus of both leaders and investors: asset growth and return on assets.

Asset growth. To evaluate a company's year-on-year investment increase, we determined its 10-year median asset-growth rate (the closest metric to total investment change). We then percentile-ranked the companies within their assigned subindustries and grouped them into quartiles.

Companies that are investing heavily should, in general, be growing their assets at a high rate. Companies in mature, consolidating industries may have high asset growth, too, if they've acquired struggling competitors.

Return on assets. We also calculated each firm's 10-year median return on assets, percentile-ranked the companies within their subindustries, and grouped them into quartiles. Typically, companies that aren't investing for growth focus on short-term returns—which can inform the type of investments they pursue. For instance, they may be more likely to acquire a troubled competitor whose turnaround could generate immediate returns than to invest in new technologies whose returns may come much later.

We then grouped the companies' industries into three maturity stages—accelerating, steady growth, and mature—and saw that the sweet spot for investing in the future differs across them. (See the exhibit "The Investment Growth Map.")

Accelerating industries. Companies in accelerating industries—such as biotechnology, payment-processing services, healthcare equipment, and renewable electricity—have a median annual revenue growth rate of about 10% over a 10-year period. If you're in such an industry or considering entering one, you need to be prepared to significantly build up your assets ahead of the returns they generate. If your asset growth is below the investment sweet spot, your EV/R can be as much as 35% to 50% less than it would be if your investment rate were optimized.

Companies in accelerating industries maximize EV/R when they are in the first (highest) quartile of asset growth and fourth (lowest) quartile of return on assets. These companies (including both firms with less than $1 billion in revenue and large multibillion-dollar corporations) have asset growth rates around 8% and ROAs around 3.8%. Those

IDEA IN BRIEF

THE PROBLEM

Many companies either underinvest in growth or overspend on initiatives that fail to generate long-term value—hurting both competitiveness and valuations.

THE RESEARCH

An analysis of 2,900 U.S. public companies over a decade shows that firms achieve the strongest valuation multiples when they operate within an "investment sweet spot" calibrated to their industry's maturity and growth dynamics. Companies investing outside that range saw valuation penalties of 20% to 70%.

THE SOLUTION

Leaders should align investment levels with their industry context, manage growth initiatives as an integrated portfolio rather than isolated projects, and tie capital allocation directly to strategic priorities. Companies also need governance systems and cultures that support disciplined experimentation, clear growth metrics, and smart risk-taking so that they can reinvest confidently without overextending.

107

---

Companies in mature industries face a real risk of overinvesting and chasing growth that turns out to be a mirage. This carries a penalty.

numbers signal that companies are plowing profits back into the business, driving up assets while driving down returns.

For companies in accelerating industries, the challenge is to figure out what asset base they'll need in order to have the capacity to meet the market's aggressive growth expectations. For example, companies building marketplaces or digital platforms trying to generate network effects to rapidly accumulate customers have to make outsize early investments to get big quickly. That carries a lot of risk, especially for startups. Smart companies find the right balance of organic and inorganic investments and ensure they have enough of both to feed the growth machine.

Steady growth industries. If your company is in a steady growth industry, such as healthcare facilities, construction materials, building products, data processing, or asset management, the challenge is striking the right balance between asset growth, which supports sustained revenue momentum, and return on assets, which increases the value of the asset base.

Steady growth companies in the sweet spot maximize EV/R when they are in the second quartile of asset growth (at about 5%) and in the third quartile of return on assets (also about 5%). For the approximately 73% of companies in this category that are outside the sweet spot, we see EV/R multiples that are 37% to 50% lower than those of firms in it, presenting a great opportunity for improvement.

Steady growth companies that are investing in the optimal range have typically already built a base of assets that allows them to increase their revenue. They generally invest in R&D to add new solutions to their product portfolio that diversify revenue streams or in smaller, tuck-in acquisitions.

Mature industries. If your company is in a mature industry, where additional investments may have diminishing returns, you need to be particularly purposeful about your investment. If you're planning to increase it, you'll need to send shareholders a clear message about exactly how the additional assets improve your growth prospects.

Companies in mature industries face a real risk of overinvesting and chasing growth that turns out to be a mirage. This carries a penalty: Our study found that the 48% of companies in this category that accelerate asset growth beyond their investment sweet spot can see EV/R multiples fall by 20% to 70%. For companies in mature industries, the sweet

spot is in the fourth quartile of asset growth (about 1.6%) and in the second quartile of return on assets (about 6%). Those that strike this balance are limiting their investments, making as efficient use of their assets as they can, and reaping profits while returning cash to shareholders.

A challenge for firms in mature industries is that disruptive companies may use emerging technologies to reinvent their competitive landscape. For example, during the late 1990s the department-store business was a mature industry, but in the early 2000s companies in it were forced to pivot as retail was reinvented by e-commerce. In the face of disruption, companies may need to rework their strategies and invest in emerging technologies to remain competitive or may have to enter higher-growth industries. Both moves will typically require them to increase their asset growth dramatically and decrease their ROA—in effect to become startups again, at least when it comes to investment. If investors like what they see, they'll reward companies with higher multiples. But that rarely happens: Only about 5% of companies in this situation are able to significantly move to the first or second quartile of asset growth and still achieve top-quartile EV/R multiples.

Given that the majority of companies we studied are investing outside their optimal ranges, it's clear that many need to rethink how they manage the investment process—from sourcing compelling ideas for growth initiatives to tracking and integrating investments in the business. Let's look at each of these in turn.

Identify and Rationalize Your Growth Investments

With all the aging assets that need refreshing and technology improvements that just keep up with requirements (or what your competitors are doing), the true growth opportunities you should finance may not be entirely obvious. You must not only uncover them but dedicate capital specifically to them—something companies often neglect to do. Consider the holding company of one global multiline business, which set a goal of generating 20% of its $20 billion in revenue from new technology-driven solutions, including AI, within the next five years. The company didn't explicitly account for the investment needed to deliver on that aspiration—and didn't understand that it would take

108

---

The Investment Growth Map

Many companies struggle to decide how much capital to dedicate to growth efforts, undermining their prospects by underinvesting or overspending ineffectively. Research on 2,900 U.S. firms over 10 years revealed that businesses in accelerating, steadily growing, and mature industries have different investment sweet spots, which are determined by their asset growth rates and return on assets. Ideally, companies in accelerating industries should be in the top quartile of businesses for asset growth and in the bottom quartile for ROA. Companies in steady growth industries should be in the second quartile for asset growth and the third quartile for ROA. Companies in mature industries should be in the fourth quartile for asset growth and the third quartile for return on assets. When companies land in their sweet spots, they maximize their market value.

Source: PwC

approximately $5 billion. Nor did it ask the capital planning group to evaluate fundraising options.

In parallel, the investment committee, in its annual budgeting process, allocated $800 million in capital to strategic initiative requests from multiple business units. They included several projects that could have been used to jump-start the new growth strategy—for instance, a partnership to develop a predictive maintenance solution to support a software-as-a-service business model. It wasn't categorized as a growth investment but went into the same partnerships-and-alliances line item as vendors providing IT help-desk services. Meanwhile, to keep up with the competition, the company increased the chief technology officer's budget for emerging technologies with cross-business-unit synergies, such as AI. The CTO prioritized investment in a pilot to test potential AI applications for service optimization, but it wasn't recognized as a growth investment,

FINANCE & INVESTING

even though the goal was to develop a solution that would increase service revenue.

Your challenge, then, is to figure out how much you're actually investing in growth and manage your investments more effectively. To do that, take the following steps:

Calculate your growth investment rate (GIR). We recommend using this metric, which simplifies the sweet spot analysis to asset growth rate minus return on assets, to see how much you're investing relative to your peers. That number can vary across companies from -5% to 50% or sometimes even higher. If it's positive, it means you're investing in assets at a rate that is ahead of your proven ability to generate a return on them for investors. Whether that suggests you're overinvesting or not depends on your context. How aggressively your competitors are investing, your investment philosophy, your confidence in your growth prospects, your certainty about your strategy, your potential for differentiation, your risk appetite, and the maturity of your industry all inform if and how your investment pattern needs to change.

For example, our analysis indicates that in the late 1990s mature retail companies such as Blockbuster, Borders, and Tower Records each had a negative GIR owing to their perceived lack of growth prospects and investor pressure to maximize returns. Then these companies encountered new players such as Netflix, Amazon, and Apple's iTunes. As it became clear that the upstarts with their new digital business models were permanently disrupting the market, the incumbents' leaders should have seen that their approach needed to change. If they'd been benchmarking their GIR against the new competitors, they might have realized they had to invest in building digital assets while maintaining a retail footprint. Our analysis, done retrospectively, shows they needed to increase their GIR by a multiple of three or four to keep up.

Ensure that your investments are managed comprehensively. The next step is to look at how you manage your portfolio of capital investments. In many companies each growth project is evaluated independently of the others. To attract funding, managers in the unit sponsoring a specific project typically put together detailed multiyear financial models that set early, overly ambitious revenue targets based on what often turns out to be a perfunctory assessment of market potential and customer demand. The evaluation methodology can vary considerably: How you evaluate a new

109

---

Companies have to do more than just invest more (or less) in growth. They have to target the opportunities and resources that most align with their growth strategies.

110 Harvard Business Review September–October 2026

---

FINANCE & INVESTING

factory or customer solution will differ from how you evaluate a new market entry. Worse, investments will be managed on different timelines, and different categories of investments will be overseen with different governance mechanisms.

To avoid confusion, we advise leaders to follow a practice that venture capitalists and private equity firms use. They manage the mix of smaller and larger investments in their portfolios by explicitly categorizing them according to their growth and return goals and time horizons. Some investments will have near-term goals; others will lay the foundation for long-term returns. That approach will instill discipline and allow your senior team to understand how various investments add up to deliver on overall company targets.

Drive your investments with a clear purpose and logic. Now look at what specifically you're investing in and how you track performance. Many management teams evaluate the progress of projects only with financial metrics like return on investment or internal rate of return. But these metrics don't provide the insights about long-term growth potential that you need in order to make timely and informed capital reallocation decisions about bigger bets. In fact, we found that approximately 62% of the companies we analyzed can't explain how investments in R&D and assets will convert to revenue, which increases the risk that they won't generate the future margins firms have targeted.

Ultimately, investors want to put their money in businesses that have an expected return, a focus on the right growth markets, an articulated differentiation, and a way to win. We addressed this in our 2024 HBR article, "Create a System to Grow Consistently," in which we explained how the ability to link investments to a clearly defined strategy and value-creation model greatly improves valuation multiples. What this means is that companies have to do more than just invest more (or less) in growth. They have to focus on the opportunities and resources that most align with their growth strategies rather than pursue a disparate mix of projects identified by individual teams or executives.

Companies that get this right make it a clear leadership objective to establish the right system of metrics to track their investments' progress. Such metrics allow you to trace over time how investments in new products, solutions, services, and capabilities convert into sales to target customers—and then into growth and financial impact.

111

---

FINANCE & INVESTING

A good example of this comes from Dexcom, a healthcare company. Its strategy is to be the market leader in direct-to-patient diabetes treatment solutions. Its managers track how investments in R&D and Dexcom’s partner ecosystem increase the value of its continuous-glucose-monitoring platform. They measure the rate of customer adoption and revenue from specific new digital features, connected health-monitoring devices, and consumables to evaluate if the asset investments are delivering the anticipated growth and whether to increase, maintain, or decrease the investment rate over time.

Connect the investment portfolio to your strategy process. The management of your growth portfolio needs to be part of your strategy calendar. Companies should conduct regular project reviews comparing actual and targeted investment performance at defined stage gates, with specified capital reallocation rules informed not just by financial indicators but by the market’s reaction and customers’ responses.

To achieve this, many companies establish formal cross-functional investment committees and incubators with explicit mandates, scorecards, and escalation paths to speed up capital reallocation decisions. The challenge is designing them in a way that keeps the unit sufficiently integrated with the broader corporation. All too often we’ve found that these units become innovation silos that are ignored by the mainstream business units, with the result that companies miss or even kill many good ideas.

One company that breaks this cycle is the benefits insurer Unum Group. It recently set up a strategic innovation function, reporting to the chief strategy officer, that’s responsible for allocating capital for innovation. This function manages a stage-gated process spanning customer discovery, incubation, and minimum-viable-solution development. With each project, further capital allocation is subject to market acceptance and performance on key indicators like customer adoption. The process is seamlessly integrated into Unum’s annual strategic-planning process, which manages stop/start/continue decisions for the next year, adjusts growth portfolio targets, and refines investment needs.

Create a Culture of Bold Investment

Culture isn’t easy to change—especially when it comes to attitudes about appropriate risk and return. As long ago as

1966, Syracuse University professor Ralph Swalm published an article in HBR (“Utility Theory—Insights into Risk Taking”) presenting research that showed that professional managers were, in general, very reluctant to sponsor risky projects. More than half a century later, Dan Lovallo and his coauthors concluded in their 2020 HBR article, “Your Company Is Too Risk-Averse,” that little had changed. Company cultures reward playing it safe and frown upon failure.

In our experience the best way to counter this barrier to growth is to anchor investments in your strategy, systematically creating multiple and diverse projects that are closely monitored and reassessed as new data comes in. Progressive Insurance offers a good example of how to do this. The company was an early mover in integrating telemetry data on driving behavior into auto insurance offerings. Its leaders didn’t know the exact solution that would win or when it would generate revenue, but they knew telemetry was an incredibly important investment theme with fundamental implications for the company’s direct-to-consumer, risk-based-pricing business model.

Progressive experimented with numerous technologies and consumer adoption techniques over a five- to 10-year time horizon. It introduced formal after-action reviews for the executives and teams involved in the experimentation—giving them a forum to discuss setbacks. If they could explain why the setbacks occurred and what could be done differently in the future to accomplish the long-term goal of the investment, the funding continued. The process created incentives to surface failures quickly and avoid perpetuating bad investments while rewarding smart risk-taking, learning, and perseverance. The result: Progressive became an early leader in monetizing telemetry in insurance.

Companies can also harness the creativity of their organizations using forums like innovation challenges and ideation events to surface bold ideas. But these need to be regular occurrences, not one and done. And leaders must visibly recognize and reward teams of employees (versus individuals) for surfacing and executing bold ideas. Organizations that create financial rewards for those teams and back them up by providing measured capital to pursue their initiatives make the organization psychologically safer for experimentation.

Last, leaders need to reinforce the acceptance of diverse opinions. They must encourage employees to speak up

112 Harvard Business Review September–October 2026

---

Culture isn't easy to change—especially when it comes to attitudes about appropriate risk and return. Company cultures reward playing it safe and frown upon failure.

without any fear of reprisal when the status quo is stifling innovation. That can be difficult in the consensus cultures that many firms favor, where the rigorous discussions required to shape ideas worthy of big investments often fail to take place. The good news, though, is that in most companies you can find employees who will provide the important challenger mindset that your organization needs.

MANY LEADERS STRUGGLE with the fundamental decision of how much to invest in growing a business. Making the right call starts with accurately evaluating which investments will actually lead to growth and understanding whether and why you're allocating too much or too little to them. For many companies this will require a complete

rethink. The objective should be not to eliminate risk but rather to embrace it in a rational fashion. The ambition should be a corporate investment program that—for companies in any industry and in any situation—finds the investment sweet spot. ▼

HBR Reprint R2605H

PAUL BLASE is a principal at PwC U.S. and leads the firm's growth platform. PAUL LEINWAND is a principal at PwC U.S., the leader of the PwC Leadership Center, and an adjunct professor at Northwestern's Kellogg School. He is a coauthor, with Mahadeva Matt Mani, of Beyond Digital: How Great Leaders Transform Their Organizations and Shape the Future (Harvard Business Review Press, 2022).

113

---

Evidence. Insight. Action.

Get expert-backed intelligence—drawn from the HBR Advisory Council and global business leaders—and explore hundreds of white papers and research reports from HBR Analytic Services.

RECENTLY PUBLISHED

Building Strong Brands Sponsored by Frontify

Driving Business Value with Agentic AI Sponsored by Wipro

Leveraging Alternative Payment Models to Ensure Broad Access to Personalized Medicine Sponsored by BU Questrom

Procurement at a Crossroads: Evolve or Be Left Behind Sponsored by Zip

Accelerating Growth Through Revenue Enablement Sponsored by Seismic

View all the reports at hbr.org/hbr-analytic-services

Register for the HBR Advisory Council at hbr.org/advisory-council.

---

Advice & Inspiration

Experience

MANAGING YOURSELF

Five Ways to Wield Influence When You Lack Authority

by Pamela Meyer

MOST LEADERS EVENTUALLY hit the same wall. They may have a position, a title, and a winning track record, but at some point they will need to mobilize people who aren't their direct reports, won't be able to provide input into key decisions, or won't have control over their budgets.

How do you get the results you want from employees, peers, and bosses when you have no formal authority over them? How do you make your voice heard when

Illustrations by JULIE GUILLEM

115

---

experience

you're not involved in every discussion? This is known as the indirect influence problem. It's more common than many leaders realize—or are willing to admit—and most have no idea what to do about it.

Some push harder to exert what authority they do have, but this often triggers what the psychologist Jack Brehm dubbed reactance—the instinctive tendency to push back when you feel your autonomy is threatened. Decades of research across the fields of marketing, public health, and organizational psychology confirm the same finding: When someone tells people what to do, they often do precisely the opposite.

Meanwhile, the conditions that once made workplace authority effective are changing. Employee engagement in the U.S. has fallen to 31%, according to Gallup. A recent Edelman survey found that trust in employers declined for the first time in 26 years. Work has become more distributed, specialized, and dependent on informal networks, such as Slack channels and alumni networks. As McKinsey reported, "New data tell us that there is no situation in which leaders need to follow the old maxim 'Do it because I said so.''

Today's leaders need to find different ways to motivate and drive change so that team members choose to act instead of being forced. Drawing on two decades of research into what builds or destroys trust, including thousands of interactions with C-suite executives and management teams, I've identified five paths that leaders can follow to wield indirect influence in high-stakes settings.

PATH 1

INFLUENCE SPOTTER

A few years ago I was brought into a Fortune 100 consumer-goods company to work with a newly appointed chief innovation officer tasked with modernizing operations. She had a budget and the CEO's backing, but every initiative she launched seemed to garner polite agreement in meetings, followed by inaction. During meetings with the division presidents, she noticed that many looked to David, a quiet EVP who ran one of the legacy business units, for a read on how to respond to her. When she spoke with David one-on-one, she discovered that he talked with the company's recently

retired COO weekly. This person no longer had a seat at any formal decision-making table, but his opinion still governed how the old guard thought about risk. So she tracked the COO down to ask what he'd learned from a transformation effort that had failed a decade earlier. When she brought up her conversation with the COO, and the concerns he'd raised, with the team, David's posture changed, and within a week her plans began to progress.

As this story shows, power brokers aren't always obvious. A C-suite or otherwise senior title is only one indicator. Consider Jeff Dean, the lead of Google AI: He is less widely known than Alphabet CEO Sundar Pichai but nonetheless

116

---

The indirect influence problem is more common than many leaders realize—or are willing to admit—and most have no idea what to do about it.

carries a lot of weight within the broader company. Ask yourself: Whom does everyone look to for an opinion before making decisions? What matters to those leaders, and who shapes their thinking?

To recognize hidden sway:

Track decision processes, not job titles. Behind-the-scenes influencers often guide the group without formal authority.

Watch where people look. Pay close attention to gaze patterns. The way heads turn or a group’s eyeline converges can reveal whose opinions matter.

Notice who speaks first and most often. Early contributors on crucial issues typically emerge as perceived leaders.

Spot calculated rule breaking. Failure to adhere to a dress code or social norm in a high-status setting signals confidence if done deliberately.

This path is particularly useful when you suspect that what appears on the surface isn’t the full story. It is invaluable if you are new, lack clout, or face entrenched norms or informal coalitions and cliques. When visible doors won’t open, look for the confidantes, advisers, drinking buddies, golf partners, or other trusted voices.

PATH 2

BACK CHANNELER

At one pharmaceutical company I worked with, two division heads had stopped speaking, blaming each other for a botched product. They were communicating exclusively through assistants, and their conflict was becoming

common knowledge across the organization. The CEO was tempted to order the two to resolve the issue, but he worried that mediation from the top would backfire. Instead he asked me, as an outside consultant, to create a back channel. (An impartial colleague or HR leader respected by both executives also would have been a good choice.)

I met with each division head separately and asked both of them to name three legitimate grievances they had with their counterpart and three things the other had done well during the launch. I shared those lists without commentary, then engineered two reasons for them to function professionally alongside each other without needing to address the conflict: a vendor briefing, followed by a budget preview. By the second meeting, they were able to make eye contact. A week later they had a private dinner to work out a reconciliation. The back-channel approach was effective because it gave each person an off-ramp that didn’t require a public act of contrition—just enough private acknowledgment to make the next step possible.

Back channels allow leaders to lower the stakes, protect egos, and explore options in psychologically safe conditions. They reduce reactance and open up real reflection.

To create a private connection:

Notice when a back channel is needed. Watch for public agreement that does not match private behavior, or repeated yes answers with no follow-through.

Open the door without pressuring. Keep the outreach brief, optional, and easy to decline.

Ensure that the channel is useful.

Show up with new information, a clear trade-off, or one question that will affect a decision.

Start tentatively, then become firmer. Use soft language to open the conversation, then shift to clear commitments when it’s time to make a decision.

This path works only if the back channeler is a trusted party; otherwise, it can feel like a power grab—or worse, a threat. Use this approach when formal discussions are frozen, issues are charged, or relationships are tense. It should be a temporary measure, however, employed just long enough to open new possibilities before returning to transparent collaboration.

PATH 3

PRECISION MOVER

A newly appointed COO at a regional hospital network was asked to revive a staff-restructuring proposal that had been dead on arrival 18 months earlier. She had a new plan and momentum, but when I was hired to advise her, I told her to wait. At the time, a looming union negotiation had everyone in a defensive crouch, and I was worried that fear and uncertainty about it would taint even the best proposal.

Shortly after management and workers reached a deal, the CFO announced strong quarterly results, and two senior skeptics of the first restructuring attempt left the leadership team. The COO pitched her plan then, and it was quickly approved.

Precision movers know not to act when people are experiencing the kind

117

---

In your own battle for influence, directness still has its place, but the real skill lies in recognizing when a subtler approach would be more effective.

of anxiety, defensiveness, or anger that exacerbates reactance. Timing, phrasing, and intensity matter. Good points don't land if they're delivered at the wrong moment. It's better to play the long game, reading moods and stress levels before springing into action.

When Apple's engineers wanted to show the notoriously critical CEO Steve Jobs a prototype of the touchscreen they were developing, they first brought it to Apple's design chief, Jony Ive, knowing that he would pick the right time to share it with the boss. Ive obviously did choose a moment when Jobs was more receptive to the idea—even though it wasn't his—and Apple eventually produced and sold more than 3 billion iPhones.

To master strategic timing:

Control the flow of information.

Hold back sensitive details until you've assessed trust levels.

Leverage the power of silence.

Rather than rushing to respond, create conversational pauses to prompt deeper reflection.

Time your ask to the leader's bandwidth. A person buried in a crisis has no room to evaluate any new proposals. Think in terms of windows and time horizons, not moments. What failed in February may pass in April.

Notice when the discussion is overheating. If people are talking faster or debating in circles, narrow the decision to what can be settled now.

Precision movers are the most powerful when the environment is in flux and key players are feeling uncertain or confused. Ask yourself: Am I patient? Am I emotionally attuned enough to detect mood shifts? Can I withhold

information until the time is right? If so, this path can deliver outsize influence.

PATH 4

CONNECTOR

By age 32, Marcus had turned networking into a system: attend a conference, offer a handshake, follow up with a note, repeat. When his Series A startup hit a wall in 2019, he needed a warm introduction to a procurement lead who could potentially save the company, so he reached out to 11 of his contacts, but none felt any particular obligation to help him. The company folded within six months.

Five years later, Marcus was a VP of product at a midsize software-as-a-service company, where he'd become known as someone who showed up for colleagues—he'd defended a coworker in a difficult executive review, taken on a time-consuming project no one else wanted, and written a genuinely useful recommendation for a junior engineer.

When his company needed to land a partnership with a vendor known for long sales cycles and no patience for cold outreach, Marcus didn't create a deck. Instead, he called the colleague he'd defended in that review years ago, who was now a director at the target company. Three weeks later the deal was signed.

There are two kinds of networkers: those who collect people like business cards—a human Rolodex built for extraction, activated only when they need something—and those who spend years making deposits they don't expect to see returned. The first kind can fill a room with people; the second can move those people.

Connectors change outcomes by widening the field of options. Instead of pushing people toward solutions, they link the right people at the right time so that problems can be solved with better information and less defensiveness.

To harness networks effectively:

Manage information boundaries.

Decide when to connect people and when to keep them separated.

Leverage knowledge from diverse backgrounds. By connecting different groups, you can access varied perspectives that may be invisible to others.

Identify the gap you need to bridge. Look for stalled work, missing coordination, or absent expertise.

Translate before you connect.

When groups speak different professional languages, do the interpretation for them.

The connector path is especially useful when you're dealing with siloed teams, running a project that cuts across functions, or trying to realign a fractured group without triggering egos or resistance. It requires a leader with strong people skills.

PATH 5

PULSE READER

A senior partner at a global consulting firm kept noticing a pattern on tech-heavy engagements: Asian junior analysts were being placed into coding and data architecture roles—sometimes ones they hadn't been trained for—at unusually high rates. She convened a closed-door session with these staffers and opened with a direct question: "Where have you felt overestimated or underestimated in how you were staffed?" After

118

---

experience

some hesitation, a few people offered candid responses, such as “I think we get placed into technical roles based on how we’re stereotyped, not on what we’ve actually done” and “I’ve been staffed into coding work I wasn’t ready for, and it’s hard to push back.” They agreed to let her take this feedback to leadership without divulging names. When she did, she also asked her peers a direct question: “Where might we be confusing perceived strengths with actual capability?” Within weeks, staffing leads began verifying technical readiness more explicitly, and junior analysts had clearer ways to decline roles that didn’t fit.

Pulse readers notice not just what is said aloud but also where energy drops

or attention converges, when silence carries weight, and how opinions are forming. They track shifts in alignment, hesitation, discomfort, and momentum. They consider the whole system. And they’re able to both diagnose and shift emotional currents to achieve strategic change.

To get a better read on your team:

Create psychological safety first.

Charismatic leaders can unintentionally silence others through the “awestruck effect,” in which people suppress emotions when in their presence. Restraint works better when you need to read between the lines.

Connect emotionally before logically. By tapping into feelings such as

hope and belonging and by showing genuine curiosity, you can make your messages more compelling.

Carefully choose your words. When you actively frame a challenge in a neutral or positive light, you’re better able to evoke certain emotions and shape which solutions people consider viable.

Distill complex problems to their essence. When issues feel overwhelming, use structured techniques to sharpen focus. Try the 40-20-10-5 method, in which you progressively distill a problem from 40 words to just five.

This path demands emotional intelligence. It is most successful when the vibe is off, group identity and cohesion are slipping, and anxiety or fear is warping decision-making. If people look to you for emotional cues, this is often the best approach.

IN MOST ORGANIZATIONS expertise is distributed, coalitions form outside the org chart, and decisions take shape long before they become official. People guard their autonomy and react strongly to feeling managed. In your own battle for influence, directness still has its place, but the real skill lies in recognizing when a subtler approach would be more effective. You can choose from any of these five paths or use them in combination depending on the type of reactance you’re facing and the kind of influence you want to exert. ▼

HBR Reprint R2605J

PAMELA MEYER is the founder and CEO of Calibrate and the author of How to Read the Room: The Art and Science of Social Observation (St. Martin’s Press, 2026).

119

---

How We Did It

The CEO of HCLTech on Pivoting for the AI Era

by C. Vijayakumar

WHEN GENERATIVE AI became widely available, in late 2022, we at HCLTech quickly recognized that it would have enormous implications for our organization and our industry. For decades IT services have operated via linear models, with staffing and revenue growing in tandem. Every major technology wave, from the internet to digitalization to the cloud, created demand for more engineers, more consultants, and larger teams. Advances in AI have changed that equation: Vast amounts of knowledge work can now be completed dramatically faster with much less human effort. The technology is not just improving productivity around the edges; it's altering the economics of our business.

At HCLTech that understanding has prompted us to consider difficult questions. How do we prepare 227,000 employees for a future in which some current roles may no longer exist and many of the most important ones have yet to emerge? How do we persuade managers that success should no longer be measured primarily by team size? And how do we help clients adopt AI when we know it might disrupt our own revenue streams?

We have opted to embrace AI wholeheartedly, and it has already changed how we operate internally, create value for clients, and measure success. Our

120 Harvard Business Review September–October 2026

Photograph by MACKENZIE STROH

---

The bigger message was that we all need to prepare for a world in which some roles would no longer exist, emphasizing the urgency of upskilling.

AI strategy rests on several important objectives: proactively transforming our services; building differentiated intellectual property (IP) to accelerate AI adoption in enterprises; creating new AI-led services like Physical AI and AI Factory, new AI ecosystem partnerships, and new tools spanning the entire tech stack; and most important, turning our people into AI builders, AI superusers, and human-in-the-loop decision-makers.

Today, thanks to this organization-wide strategy, careful investments in technology and training, and deep partnerships with clients, we are well on our way. And we hope our story will resonate with any organization hoping to adapt its business for the AI era.

A HISTORY OF REINVENTION

HCLTech has a long history of advancing on new frontiers. It was founded in 1976 as HCL by Shiv Nadar and a group of colleagues with a radical vision of building world-class technology from India at a time when few believed an Indian startup could compete meaningfully with global companies. We have continually evolved over the decades, shifting from IT hardware, to engineering and R&D, to services and software. Each transition required us to move aggressively to where technology and client demands were heading rather than protecting the business as it currently existed.

I joined HCLTech in 1994 as a young engineer, worked my way up through the ranks, and served in several leadership roles before being appointed CEO, in 2016. With so many technology

changes on the horizon, I understood that leading the company would require an entrepreneurial spirit: the willingness to try new things, learn fast, and adapt to changing dynamics.

In recent years we have been working with some of the world's top enterprises to deliver advanced engineering and R&D services, which in many respects paved the way for the AI systems reinventing industries today. About a decade ago our big bet was to expand more meaningfully into enterprise software and platforms that helped clients innovate and grow. That work spanned the technologies, platforms, and research initiatives that underpin the next generation of AI solutions. While the pace of progress has been extraordinary, we are still in the early stages of an AI-led transformation that will require navigating new technological, operational, and societal challenges with the same discipline and ingenuity that brought us to where we are now. We believe this is an inflection point and a once-in-a-lifetime opportunity to use technology to amplify both human potential and enterprise value.

SHIFTING THE MINDSET

Executing such a large transformation starts with facilitating a mindset shift. Nearly everyone in our business came of age in a world where larger teams meant greater impact and higher revenues. But now teams should adopt AI to increase their output and do more with fewer people.

Speed is also of paramount importance. No one should hesitate because of a desire to preserve the old ways or

protect the current cash flow. We need to be comfortable transforming and even cannibalizing parts of our existing business, confident that we are putting ourselves in a stronger position for the future.

As I've shared publicly, over the next few years we are targeting a fundamentally different operating model, one in which growth is no longer tied to an increase in head count but is instead driven by enabling our team members to become AI builders and AI superusers, amplifying their potential through AI-augmented, platform-led delivery.

We first announced our intention in February 2025, aiming not to create anxiety but to communicate transparently, bring our people along on the journey, and foster the mindset shift required to succeed. Our objective is to align our teams around a shared vision and, most important, ensure that they are energized and excited to help build that future together.

How did we bring everyone into this new way of thinking? With conviction and communication.

When I realized the power of AI, I chose not to beat around the bush with our teams. That's why I publicly announced an audacious goal: to double our revenue with half the number of people. The bigger strategic message was that we all needed to prepare for a world in which some current roles would no longer exist, emphasizing the urgency of upskilling. Translating that goal into baby steps, our near-term ambition is to deliver at least 5% revenue growth without having to add more people by enhancing our existing workforce with AI. Of course,

121

---

experience

incremental revenue growth after that 5% will likely require additional talent. But the message is clear: “We are no longer dealing with business as usual. AI changes everything. And your company is encouraging you to stay ahead of the curve.”

Direct, consistent communication has been key to ensuring that the challenge is well-received. I began by calling a conference of our top 150 services-delivery and corporate-function leaders to make sure they were aligned and willing to help percolate the new thinking through their teams. I then traveled to our global locations to meet all our sales teams (all 1,200 members) to explain the strategy and assure them that the changes would ultimately prove positive for anyone willing to work hard, learn, and adapt. With our board, and externally to clients, the media, and shareholders, my message has been exactly the same. We are aggressively adopting AI across every role, as well as helping customers do so. Some jobs and revenue streams will disappear. Others will be created. AI will operate autonomously in very few areas; most often we will have a human in the loop. The future is employees and AI agents that work together. While we expect certain revenue streams to slow down in the near term, we are comfortable making disciplined investments today to strengthen our long-term growth trajectory and future value creation.

INVESTMENT IN TECH AND TRAINING

Transformation efforts succeed only when companies devote ample resources to them. We are currently reinvesting

part of our profits (see the exhibit “HCLTech”) back into AI technology, the creation of differentiated IP, research and development, and training. We have built a larger team around the chief technology officer and asked him to phase in and scale up initiatives across the organization, starting with broad awareness of the new tools and initiatives and moving into deeper technical and leadership development.

One key platform, AI Force, applies AI to every step of the software-development and data life cycles as well as to both IT and business operations. Many companies currently use AI only to help write code, but we thought it important to build a solution to increase productivity throughout the entire process. Our ability to deploy this service for clients has been a key differentiator in two of our largest wins this fiscal year, one for Guardian Life Insurance and one for a major fashion retailer. Our focus is not only on transforming our services with AI but also on helping our clients successfully transform their business models through AI.

We have also created an office of responsible AI governance and brought in a senior executive to lead it. That group will make sure every platform we build incorporates guardrails so that all AI-enabled decisions are checked for biases and can be traced, reviewed, and reversed if necessary.

On the training front, our expectation is that 5% to 10% of team members will be AI builders who will create new solutions for others to use. We are upskilling our top technologists to fit this category as well as hiring new AI specialist talent. We’re asking the other

FACTS & FINANCIALS

HCLTech

Headquarters: Noida, Uttar Pradesh, India No. of employees: 227,000+

Source: HCLTech

90% to 95% of our people to become AI superusers in their respective areas by availing themselves of the tools we’ve built or bought, prompt libraries we’ve created, and programs we offer. Our hope is that all employees can boost their productivity by a factor of three to four.

We recognize that it will take time for all 227,000 of our team members to fully leverage a technology that is still evolving at breakneck speed. But curious, driven employees have been quick to experiment and become adept, and I can confidently say that most of our people now have the motivation and skills to do their jobs more easily and effectively with AI.

122

---

We work in partnership with tech firms, but the roles are distinct and complementary: They build frontier models while we make those models work inside real organizations.

I try to lead by example. I use Claude, Anthropic's AI assistant, for research that I previously asked my teams to deliver, and I treat the AI chatbot as a consultant with whom I can discuss and debate ideas, fine-tune my strategic thinking, or engage in problem-solving. I even created my own revenue-forecasting app. And I keep sharing these examples with my team and the broader organization so that they see that I'm embracing our AI transformation right alongside them.

PARTNERING WITH CLIENTS

Of course, HCLTech's long-term success rests on helping our clients embrace AI technology in the same way we are. So far, adoption has been gradual, for several reasons. I sometimes joke that God was able to create the world in seven days only because he didn't have an installed base and didn't have to worry about whether his new systems were compatible with the legacy ones. Companies are rightfully approaching the latest AI tools in a cautious and calibrated manner.

The first mile in our work with them is obtaining all the security, legal, and privacy clearances we need and completing the platform evaluation. Over time, as our own knowledge has accumulated, we've learned how to complete this process much more quickly. If it took three months to get approvals for the first client, we're now able to accomplish it in two weeks. From there we progress to running pilots, most of which have been extremely successful. The last mile is scaling up the new technology across client companies:

training their teams, measuring output, and showcasing real velocity increases or cost savings. Again, over time, we have created recipes for different functions and types of organizations.

Some companies have chosen to simply work directly with AI providers such as Anthropic or OpenAI to drive transformation at scale. We believe that view misses how enterprise change actually happens. We work in partnership with technology firms, but the roles are distinct and complementary: They build frontier models while we make those models work inside real organizations, using our deep knowledge of industries, legacy environments, operations, and governance to help embed AI into existing technology platforms securely, scalably, and in a way that is tied to measurable business outcomes.

Some recent client wins have included helping a healthcare system create an AI adviser for clinicians that has increased patient satisfaction and saved 10 minutes of administrative work per interaction; shifting a bank's trade surveillance from a sampling system to 100% AI; and using AI to optimize cargo packing to increase revenues at an aircraft manufacturer.

We also help some of the world's leading technology giants speed up AI-led innovation, including advancing robotics capabilities such as full-body robot dexterity and helping semiconductor clients reduce development cycles and accelerate time to market for next-generation products. That traction is reflected in our high-tech services business, which is now among the fastest-growing in the industry.

In our view we will grow our market share by proactively helping our clients embrace AI and further drive innovation.

MEASURING SUCCESS

Internally, we are evaluating the progress of our AI transformation by tracking adoption, fluency, productivity gains, cost savings, and innovative solutions. While our existing book of business will face a near-term reduction of 2% to 3%, we expect to gain a larger share of wallet with our clients and increase our broader market share. We also see our new AI-led revenue streams gaining momentum. Our advanced AI revenue boasts a $620 million annualized run rate (as of March 2026), and we expect it to grow 30% a year. With all of that, our revenue per employee continues to show an upward trend.

Externally, our key metrics are the number of clients for which we've deployed AI and the extent of deployment in each. If we aren't making progress with a client, or if we think we could be moving faster, we consider which interventions might get it up to speed. Most important, we are exploring new opportunities with every engagement, helping our clients to capitalize on what is arguably the most transformative technological shift of our generation. While we are encouraged by our strong start, we believe HCLTech is still at the beginning of the journey, with significant opportunities ahead. We are excited about the future and the value we can create for our people, our clients, and all stakeholders. ●

HBR Reprint R2605K

123

---

HBR Executive

“ The clarity and relevance of the content to real executive decisions made the frameworks and insights useful. They helped sharpen thinking, structure conversations, and drove more-focused outcomes. ”

HBR Executive Subscriber

Lead with insight.

Act with impact.

HBR Executive empowers teams of any size to navigate uncertainty, stay ahead of emerging trends, and plan for a stable future.

Leadership teams succeed together.

Use HBR Executive for:

• AI activation
• Upskilling
• Peer networking
• Strategic alignment

HBR Executive is an annual subscription that includes content and conversation from the experts at Harvard Business Review.

Subscribe to HBR Executive today. hbr.org/executive-teams

---

HBR's fictionalized case studies present problems faced by leaders in real companies and offer solutions from experts. This one is based on the Ivey Publishing case study "Porsche Drive (A): Vehicle Subscription Strategy" (case no. W39684), by Vaughan Griffiths, Varun Gupta, Darren Manion, Sarah Sargent, and Xiao Zhang, which is available at HBR.org.

Is Our Subscription Program Really Working?

by Varun Gupta, Xiao (Shawn) Zhang, Vaughan Griffiths, and Jing Li

JON WEBER, SENIOR VICE PRESIDENT of customer strategy for Hohenbruck Motoren North America, was driving to work in his HX7, the luxury carmaker's SUV model. Cutting through a residential neighborhood to avoid highway traffic, he eased to a stop at a traffic light, glancing at the time on the dashboard.

Then he saw the other HX7. The same color as his, it was being delivered to a brick house where an HX3 hatchback was parked. An employee from the dealership greeted the customer, who emerged from the house in running shorts and exchanged keys with him.

Then the employee got into the hatchback and drove away, leaving the customer grinning at the gleaming SUV—all before Jon's light turned green.

This was Hohenbruck Access, the carmaker's subscription service, in action. It had launched in 2021 as an experiment designed to test simple but radical questions: Would affluent customers pay to use a Hohenbruck without purchasing one or committing to a traditional lease? Would subscribers become buyers? And would the service change how customers view Hohenbruck's luxury brand?

Illustrations by JORI BOLTON

125

---

The ease and speed of the exchange felt almost shocking, a clear disruption of Hohenbruck's ownership journey—and potentially of its brand mystique.

The first subscription offering had leaned hard into flexibility. For $4,000 per month, customers could switch between models in as little as 24 hours, using the HX3 hatchback for the workweek, the S90 sports car for the weekend, and the HX7 SUV when family visited. Recently Hohenbruck had expanded the program to 11 cities and introduced single-model subscriptions that started at $2,000 a month. All contracts included insurance, maintenance, roadside assistance, and delivery. The subscriptions were more expensive than traditional leases but were easier to start and stop.

Jon had been meaning to dig into the service's user insights, but gathering data was difficult because customer engagement happened through franchised dealerships. He knew the basics: Most subscribers were new to Hohenbruck. They were usually younger than its traditional buyers and lessees. Some eventually bought a Hohenbruck car.

Watching the exchange left Jon keen to investigate further. The ease and speed of it all felt almost shocking, a clear disruption of Hohenbruck's ownership journey—and potentially of its brand mystique. A Hohenbruck was supposed to be desired, waited for, aspired to. It was not supposed to simply arrive. Yet here was a man in running shorts with an HX7 he did not and might never own.

GIVING THE PEOPLE WHAT THEY WANT

A few weeks later Jon sat at a conference table as dealer representatives from Texas, Georgia, Illinois, and California

appeared on a wall monitor. "Thanks for joining me today," Jon said. "I'd like to hear about how customers have been reacting to the Access program."

The Texas dealer spoke first. "Access is bringing in people we weren't seeing before," he said, "and many of them eventually buy." He read a software executive's review: "A 20-minute test drive would never have told me whether an electric vehicle fit my life. A two-month subscription revealed everything I needed to know."

The Illinois dealer agreed that multicar subscriptions were a powerful sales tool. It took time to fully grasp the differences between, say, the S90 Touring, the S90 Sport, and the S90 GT. The multicar program helped customers figure out what they wanted, leading to higher satisfaction. "If the point is to turn curiosity into confidence," she said, "nothing we do works better."

The dealer in Georgia nodded. "My team has begun tracking customer comments from follow-up calls, and we've found that Access makes Hohenbruck feel less intimidating," he explained. "Here's how one subscriber put it: 'I love the brand, but I never felt like I belonged in the showroom. This was an easier way to get to know the cars.''

The California dealer was less effusive. "Some subscribers are serious prospects," she said, "but others just want the logo for a month. They get the single-car subscription, post a few videos of themselves driving a Hohenbruck on TikTok and Instagram, and then cancel. This is not the type of customer we want to be associated with."

Jon was still scribbling notes as the meeting ended. It was clear that Access

deepened the pool of potential customers—but was it producing results Hohenbruck didn't want? Was it diluting the sense of exclusivity that the brand had spent decades cultivating? He wrote down a final thought: "Access might be the most effective—and most expensive—test drive we've ever built."

MORE OPTIONS, MORE PROBLEMS

A few days later Jon met with the founder of GearVale Labs, Eli Jaimeson. GearVale had created Access's customer-facing booking and approval platform, as well as the back-end system that matched people with available cars, tracked mileage and utilization, and coordinated dealer inventory. The contrast between the two men was hard to miss. Jon wore his usual Italian-made suit and sipped an espresso. Eli wore skateboard shoes and nursed a giant bottle of Diet Mountain Dew.

They were reviewing the service's pricing model on a screen in front of them. The takeaway was clear: Hohenbruck Access was hovering around breakeven. In stronger markets it had a modest margin. In weaker ones there was no margin at all.

"Anyone can do the math," Eli said. "The subscription price has to exceed the true cost to serve. That includes depreciation, insurance, delivery, maintenance, cleaning, idle days, mileage, employee time, and remarketing."

He clicked to the next screen. "We had more ability to resell retired vehicles than we realized early on, because used-car values were so strong. We overestimated depreciation, which made the economics look better. That

126

---

cushion is disappearing.” Eli pulled up a dashboard. Across the 11 markets, fleet utilization was averaging 64%; it was robust in Chicago and Los Angeles but below 50% elsewhere. “If resale values or utilizations decrease, some markets could be underwater in a year,” he concluded.

The next screen showed the cost stack. Depreciation was the biggest variable. Insurance followed it. Delivery and pickup—once treated as part of the service’s magic—became real costs every time a subscriber switched cars. Maintenance was predictable until it wasn’t. A car retired from the fleet at 10,000 miles was one thing; a car retired above 20,000 miles was another.

“The issue isn’t whether people will subscribe,” Eli said. “It’s whether the

margin is good enough after we include the full cost of serving them.”

“So, what’s your recommendation?”

“Stop multicar,” Eli said. “The business has to stick to single-car subscriptions. The margins there are much more attractive because the logistics are so much simpler.”

Jon frowned. “The dealers think multicar is what converts people. They worry that single-car feels too much like a rental without the commitment.”

“Multicar converts some people. But if the business depends on letting customers treat the S90 lineup like a tasting menu, it won’t pencil. Every swap adds cost. Every extra car in the fleet adds idle time. Every promise of unlimited flexibility forces us to carry inventory we may not use.”

He took a sip of his soda and clicked on a comparison of customer types. Enthusiasts had a high willingness to pay when they could explore different models in anticipation of a purchase; one S90 customer had said he would have paid an even higher monthly fee because he learned so much. But subscribers who didn’t want to buy were different. They valued convenience, bundled insurance, and short-term usage but were more price-sensitive. A Houston doctor who’d needed an HX3 for three months loved the service but wanted a lower monthly fee.

“That’s why single-car is the scalable product,” Eli said. “Different customers than our traditional buyers and a lower willingness to pay, but good margins because of the reduced complexity—and much lower risk.”

“That would also mean accepting that many subscribers may never buy.”

“Some won’t. But they’ll still pay us.” Jon leaned back. It was clear that Eli no longer saw the service as a way to sell cars. He was making a different claim: that a subscription itself should be the core product. Eli was pushing to make Hohenbruck Access an entirely different business.

BUT IS IT COMPELLING ENOUGH?

The next week Jon was in his office when Klara Richter, senior vice president of international strategy, appeared on his screen from Germany. Behind her

127

---

was a pale-wood conference room, the walls empty except for a framed photograph of an early S90, the iconic car that made the brand global.

"I want to start by saying this clearly," Klara said. "The U.S. team built something real with Access.

We've been talking about innovation at Hohenbruck for years, and this was a fantastic example of putting that into practice."

Jon nodded but did not interrupt. "Still, I have read your materials twice, and I think we need to have more clarity about the strategy here," she said. "Hohenbruck Access can be a customer-acquisition tool. But if this is marketing, it is very expensive marketing. And you need dealers to support

a model that does not always reward them the way sales and leases do."

Klara then turned to the second scenario: Access as a recurring-revenue business. "If we price it properly, simplify operations, and shift to single-car subscriptions, it could be a new way to monetize the brand even if subscribers don't become buyers.

"But," she continued, "the big question remains: Can we build that business without cannibalizing our existing one?"

Should Hohenbruck Access continue and, if so, in what form?

THE EXPERTS RESPOND.

ROBERT PHILLIPS is the former head of marketplace data science for Uber and a former professor at Columbia Business School.

Jon should pursue Access as a real business—if Hohenbruck is willing to commit.

Amazon did not begin as a cloud-computing company, but it became extraordinarily good at managing large-scale computing infrastructure, and Amazon Web Services was born. Hohenbruck may be facing a smaller version of that scenario. While Access started as a way to introduce customers to the brand, it has uncovered a demand for a different

business model. Some people may not want a lease or ownership; they may instead want flexibility, service, and luxury driving with less commitment. Who knows how many of these potential customers might come forward once the program gains real traction?

But Jon needs data, not anecdotes, to make that argument. Hohenbruck has rolled out Access across different cities at different times, which creates an opportunity for comparison. At a minimum, Jon should evaluate Access as a customer-acquisition channel. What kinds of people does it bring in? Are they truly younger? Do they offer higher lifetime value? How many become buyers? What is the cost of acquiring them?

If Access is helping Hohenbruck win loyalty from a new demographic group, that could be extremely valuable. But the company must be honest about the trade-offs: the opportunity cost of renting vehicles instead of selling them, the expense of maintaining a fleet, the time that dealers spend administering the program, the risks that arise when customers drive cars they do not

own, the need to support customer accidents, and the potential brand damage of service failures.

As for stakeholder management, the dealers will be the trickiest. Some will reject Access outright, saying Hohenbruck is about aspiration and ownership. Others will be enthusiastic. A large middle group will wait to see whether headquarters is serious. Jon and his team must anticipate and manage that split. Success will require identifying and supporting dealers who are enthusiastic about the change and are able to make it work. They can serve as ambassadors for the program and mentors to other dealers.

If Jon's analysis shows that Access has the potential to become a sustainable, profitable business, that is what he should recommend. He must acknowledge all the areas affected by such a change and emphasize that nothing can be accomplished without the CEO's full backing. After showing Klara the data, the economics, and the dealer model, he should ask her to make the decision: Is Hohenbruck willing to become not only a company that sells and

128 Harvard Business Review September–October 2026

---

“Some subscribers are serious prospects, but others just want the logo for a month. They post a few videos of themselves driving a Hohenbruck on TikTok and Instagram, then cancel.”

There was a heavy pause. “If we make Access too attractive, more customers may never become owners. Maybe that is acceptable. Maybe it is even the future. But we should not pretend this is only a sales funnel if we are really building a rental business with better margins.”

Jon had no crisp answer. The marketing case was strategically appealing but financially hard to prove. The business-line case was financially imaginable but strategically unsettling.

Between them sat a possibility he wasn’t ready to say aloud: Despite its success and early growth, Hohenbruck Access might be interesting enough to keep experimenting with but not compelling enough to justify the capital, dealer complexity, and management attention required to make it much bigger.

“The way forward is up to you,” Klara told him. “At our next planning meeting, do not bring me another rollout map. Bring me a decision.” ▼

VARUN GUPTA is an associate professor of logistics and supply chain management and leads the logistics program at the University of North Georgia’s Mike Cottrell College of Business. XIAO (SHAWN) ZHANG is an assistant professor of operations and IT management at Saint Louis University’s Richard A. Chalfetz School of Business.

VAUGHAN GRIFFITHS is the former manager of mobility services at Porsche Financial Services. JING LI is a clinical assistant professor of supply chain and operations management at Purdue University’s Mitch Daniels School of Business.

leases exceptional cars but also one that sells flexible access to them? Great companies do not just protect the core. They recognize when the edge of the business is pointing toward the future.

ZABIH ARIA is the senior director of strategy and transformation at Lincoln, Ford Motor Company’s luxury brand.

Jon should reimagine Access before expanding it.

In most new initiatives we assume that growth will lead to more efficiency and better margins. But some business models have diseconomies of scale. Balancing Access’s variables—vehicle values, insurance, maintenance, logistics, fraud risk—could get harder as the program grows, and some factors, such as accident risk and weather exposure, could be trickier in some places than in others. Currently Hohenbruck has neither the expertise nor the technology to manage all of this.

The biggest inefficiency, though, is the dealer structure. If Hohenbruck allows one dealer to roll out the program, competing dealers might object. If it asks competing dealers to cooperate, it may have to create something like a joint venture. If it bypasses dealers and runs Access directly, it risks looking like a direct-to-consumer vehicle business, which raises franchise law and dealer relations issues. The result would almost certainly be lawsuits. That’s a very hard environment to build a national offer in.

The Access program has value, however, because it attracts

younger customers. Luxury car-makers have a wealthy but aging customer base, and if they don’t reach new buyers, they’ll go from profitability to obsolescence in a matter of years.

That’s why Jon should reimagine Access as a carless marketing program rather than a recurring-revenue business. A partnership could offer several options: Hohenbruck could work with an established rental company that has the scale, systems, insurance expertise, fraud controls, fleet management experience, and logistics infrastructure to run this kind of operation. Think “Hohenbruck Access, powered by Hertz.” Hohenbruck is on the front door, but the rental partner manages the fleet and operations. Or maybe Access becomes part of a premium loyalty program for the partner’s elite customers who are demographically attractive but aren’t yet Hohenbruck owners.

All that said, Access might be more viable as a stand-alone business in the future. If autonomous driving reduces the human effort required to move vehicles around, the logistics could

improve dramatically. If laws around franchise relationships change, the business model could become more attractive. Hohenbruck should remain ready for that moment by preserving the knowledge base, expertise, and capabilities created by the program and continuing to learn about pricing, customer segments, dealer friction, insurance risk, and the economics of extended trials.

Jon should go back to Klara with a clear message: Access is valuable but isn’t ready for a national rollout. He can tell dealers that Hohenbruck will not create a civil war by forcing a direct or unfairly allocated model across markets. And he can keep Eli focused on preparedness rather than on defending an expansion plan that may not pan out. This road map gives Hohenbruck the benefits of innovation without letting a promising experiment become a costly distraction. ▼

HBR Reprint R2605L

Reprint Case only R2605X

Reprint Commentary only R2605Z

129

---

Executive Summaries

September–October 2026

LEADERSHIP

STRATEGY

CHANGE MANAGEMENT

STRATEGY

“Losing Doesn’t Work for Me”

Adi Ignatius | page 38

Following a successful nine-year stint as the CEO of PayPal, Dan Schulman was ready to retire. But after the board of Verizon repeatedly urged him to step in, he became the CEO of the telecommunications giant. His task: to turn around a company that has seen steep declines in market share, share price, and customer satisfaction. In this wide-ranging conversation with HBR’s editor at large, Schulman discusses how he’s trying to shake up Verizon’s culture. “The company had a lot to be proud of, but it was resistant to change... When the pace of change external to a company is faster than the change of pace internally, you’re falling behind.” He also reflects on both the risk and promise of AI and what it takes to be an exceptional leader.

HBR Reprint R2605A

AI Is Revolutionizing Strategic Decision-Making

Felipe A. Csaszar page 44

For decades, the limits of time and brain capacity meant that teams could consider only so many strategic options before making decisions. Strategy tools—SWOT analyses, portfolio matrices—were simple because that’s what planning meetings needed. AI changes those dynamics. It can generate and evaluate thousands of strategic options and build rich, continually updated views of markets and competitors. And it can stress-test potential plans through structured debate that isn’t influenced by internal politics. Since the same AI tools are available to all companies, lasting advantage will go to those that pair them with proprietary data, integrated workflows, and faster execution. In practice, that means casting a wider net for strategic options before narrowing the field, replacing static models with real-time ones, and making AI-assisted challenges a routine part of major decisions.

HBR Reprint R2605B

Transformation Should Be a Learning Journey

Evgeny Kaganer and Christoph Loch page 54

Most corporate transformations fail because they’re approached as fixed, top-down programs with predefined goals, timelines, and financial targets. But as the environment changes—because of technological disruption, for example, or new customer expectations—those assumptions quickly become obsolete. That’s why companies should treat transformation as a learning journey: an evolving process in which strategy, priorities, and even the destination are continually refined. Drawing on research into large-scale transformations and contrasting the experiences of DBS Bank and GE, the authors identify four types of transformation initiatives—pilots, options, established business improvements, and ventures—that together generate critical learning and sustain momentum. They also recommend three leadership practices to keep transformation efforts coherent over time.

HBR Reprint R2605C

Collaborate on the Core. Compete on the Edges.

Frank Nagle | page 64

Companies don’t win by isolating themselves, but they don’t win by overcollaborating, either. The real advantage comes from knowing what to build together and what to keep for yourself. It makes sense to collaborate on the “core”—things like infrastructure, standards, and trust, which allow an ecosystem to function—and to compete on the “edges”—those activities that allow companies to differentiate themselves. This article introduces a practical framework leaders can use to determine the right collaboration strategy. It includes five factors: market dynamics, technology life-cycle stage, your position in the technology stack, the level of competition, and social acceptance and regulation. By weighing each factor, leaders can devise a collaboration strategy that grows the market without giving away their edge.

HBR Reprint R2605D

How AI Agents Orchestrate Work Across Silos

Kris Johnson Ferreira and Jordan Tong page 76

While many companies now use AI to improve individual tasks, most still struggle to apply it to the cross-functional work that makes or breaks enterprise performance. Drawing on field research at companies including Walmart, Amazon, Ericsson, Ramp, and Medtronic, the authors argue that the next frontier is agentic AI orchestration: systems that connect task-based AI tools and organizational workflows and solicit and integrate critical information that only humans have. In these systems, AI performs analyses, routes information, and surfaces trade-offs, while humans contribute context and tacit knowledge, set guardrails, and make the final calls. Organizations that redesign decisions around orchestrated human-AI collaboration can respond faster, leverage knowledge more effectively, and execute higher-quality enterprise decisions with greater speed and consistency.

HBR Reprint R2605E

130

---

SALES & MARKETING

LEADERSHIP

FINANCE & INVESTING

MANAGING YOURSELF

HOW WE DID IT

Don't Underestimate the Power of Customer Referrals

Fred Reichheld, Jamie Cleghorn, and Wojtek Kokoszka | page 86

Companies that systematically track referral behavior uncover a powerful, underutilized growth engine. An analysis of more than 10 million consumers reveals that while about 20% of new customers come through referrals, they account for 72% of the profits all new customers generate. Referred customers cost less to acquire, stay longer, buy more, and refer more high-value customers to your company. Yet most firms overlook them because attribution systems give too much credit to paid channels and fail to capture word of mouth. Leading companies, however, treat referrals as a growth metric, invest in identifying true promoters, and design experiences that inspire recommendations. By shifting the focus from buying customers to earning their advocacy, these companies improve margins and build more durable, cash-generating growth businesses.

HBR Reprint R2605F

The Four Stages of the CEO-Board Relationship

Claudius A. Hildebrand and Douglas L. Peterson page 94

CEOs succeed or fail in large part based on how deliberately they manage their evolving relationship with the board. That relationship follows a predictable progression: building trust and understanding early on, shaping the board into a strategic partner as credibility grows, guarding against complacency once trust is established, and ultimately focusing on succession and long-term governance. Each phase requires different behaviors, from investing in one-on-one relationships and setting clear expectations to actively refreshing board composition, encouraging rigorous debate, and planning leadership transitions. The central challenge is balancing trust with oversight—ensuring that the board is engaged enough to provide meaningful feedback but not so involved that it undermines effective execution.

HBR Reprint R2605G

How Much Should You Be Investing in Growth?

Paul Blase and Paul Leinwand | page 104

One of the biggest strategic challenges companies face is how much they should invest in growth. Drawing on an analysis of a decade's worth of performance data on 2,900 U.S. public companies, the authors have identified an "investment sweet spot," in which firms balance asset growth and return on assets in ways that maximize valuation multiples. Companies in accelerating, steady growth, and mature industries have different sweet spots, they explain, and firms operating outside their optimal ranges can suffer valuation penalties of 20% to 70%. But effective capital allocation is also about aligning investments with strategic priorities and long-term growth logic. To achieve this, leaders must manage investments as an integrated portfolio, establish clear growth metrics and governance systems, and reward disciplined experimentation and intelligent risk-taking.

HBR Reprint R2605H

Pamela Meyer | page 115

As formal authority becomes less effective in modern organizations, leaders increasingly face the challenge of influencing outcomes without direct control. So, they must learn to exert indirect influence in five ways: identifying hidden power brokers who shape decisions; using discreet back channels to defuse conflict and build alignment; carefully timing interventions to individual and organizational readiness; cultivating networks of connections who can help solve problems; and reading emotional and social cues to surface unspoken concerns and guide action. Across these approaches, the central idea is to create conditions in which others choose to act rather than feeling compelled to comply.

HBR Reprint R2605J

The CEO of HCLTech on Pivoting for the AI Era

C. Vijayakumar page 120

When gen AI emerged, HCLTech recognized that it would fundamentally alter the economics of IT services, breaking the traditional link between revenue growth and workforce expansion. CEO C. Vijayakumar responded by driving an enterprise-wide transformation centered on mindset change, aggressive investment in AI platforms and training, and transparent communication about the future of work. The company is now equipping employees to become AI builders and superusers while helping clients accelerate adoption through scalable, industry-specific solutions.

HBR Reprint R2605K

POSTMASTER

Send domestic address changes, orders, and inquiries to: Harvard Business Review, 20 Guest Street, 4th floor, Brighton, MA 02135. GST Registration No. 1247384345. Periodical postage paid at Boston, MA, and additional mailing offices.

Harvard Business Review (ISSN 0017-8012; USPS 0236-520), published every other month for professional managers, is an education program of Harvard Business School, Harvard University; Srikant Datar, dean.

Published by Harvard Business School Publishing Corporation, 60 Harvard Way, Boston, MA 02163. Printed in the U.S.A.

131

---

Life's Work

“From my family and my mentors, the advice has always been: ‘Be who you are. You cannot be anyone other than yourself.’ And I don’t want to be. I like me.”

Dionne Warwick

Born into a family of gospel singers, Warwick studied music in college and began her career as a backup vocalist. But it wasn’t long before she was recording solo hits. She went on to win six Grammys and in 2024 was inducted into the Rock & Roll Hall of Fame. Though her heyday was in the late 1960s and early 1970s, she has remained culturally relevant since then. In the 1980s she was an AIDS activist; today, at age 85, she’s a savvy social media presence with a new album of duets featuring contemporary stars.

Interview by Alison Beard

HBR: What gave you the courage to try a solo career?

WARWICK: Well, I was working with two of the most prolific songwriters of our day—Burt Bacharach and Hal David—doing demos and background work, and they wrote me a record, which started it all. Hal was not only a lyricist; he was an absolute poet. And of course, Burt was a musical genius. And I sang because that’s what I do. We made each other great. We each relied on each other’s talents. And the songs will live on forever.

Did you ever encounter racism or sexism?

Racism, yes, during my first tour of the Southern regions of our country: being told, “You can’t go in that store” or “You can’t go in that restaurant.” Coming from East Orange, New Jersey, I’d never been confronted with anything of that nature. I found it interesting. “What do you mean I can’t go in there? Of course I can.” That was my attitude at all times. So I never let racism enter my life. I never allowed it. As for sexism, no. There were no efforts to force me to do anything that I didn’t want to do. I was never denied anything. I was never looked at as less than what I am. I’ve always been appreciated for what I brought to the table. I am Dionne, and everyone who knows me knows I’m Dionne.

What lessons do you try to pass on to the next generation?

I don’t give advice. Nobody takes it anyway. I will encourage, and if I’m asked a question and have an answer, I will give that. But most of our youngsters know where they want to go and how they want to get there, and

they’re going to follow their own trail. I can’t tell them what I did because what I did is no longer available to them. The music industry has changed so much that I wouldn’t know how to approach it today.

But you’re a big hit on social media!

I think that’s because I tell the truth. Most people aren’t accustomed to truth, but when they see it and know it’s coming from someone who is completely authentic, it’s a wonderful joy. Sometimes when you tell the truth, it stings a bit, but I try to always leave people with a smile and smile myself. As my grandfather told me, “Frowning gives you wrinkles. Smiling doesn’t.” So I try to, as I say, show all my teeth to people.

Which of all your achievements has meant the most to you?

I refer to awards as rewards. They’re things you earn. They’re presented to you because you’ve done exactly what you set out to do. And each one has been important to me. That said, when I started out, my goal was just to make people happy and enjoy what I was doing musically. Who knew where it would lead?

Looking back, is there anything you would do differently?

Not one thing. Every up and down has something to do with who you are and how you come out of it. So do what you do, and enjoy the ride. When I look out in that audience and see nothing but smiles, I know I’m still doing something right, and it’s easy to keep being me. ♥

HBR Reprint R2605M

Derek Preston/Paul Popper/Popper/Debbie Gentry Images

132

For more from Dionne Warwick, go to HBR.org.

---

HARVARD BUSINESS REVIEW

HBR Innovation Summit 2026

Turning breakthrough ideas into real results

LIVE VIRTUAL EVENT | NOVEMBER 12–13, 2026

Innovation requires new thinking in the era of AI. Where should you invest? What capabilities should you build? How do you create lasting value?

The HBR Innovation Summit features conversations with the CEOs of FedEx, IDEO, HCLTech, and Lumen Technologies, as well as research-based masterclasses from leading faculty. Explore how to integrate AI and emerging technologies, rethink business models, and scale innovation across your organization.

Leave with practical frameworks and ideas to turn innovation into measurable business results.

HARVARD BUSINESS REVIEW HOSTS:

Adi Ignatius Editor at Large

Alison Beard Executive Editor

VIEW THE AGENDA AND REGISTER TODAY

hbr.org/innovationsummit

---

OVERSEAS DUAL TIME CARDINAL POINTS – SOUTH

Reference 7930V, titanium

Calibre 5110 DT/3

VACHERON CONSTANTIN

GENÈVE

ONE OF

NOT MANY.

▌ 本日全报思想与文化深思雷达 (Intellectual & Cultural Significance Radar)

在系统通盘梳理本期报纸所涉及的所有管理实践、技术部署与组织战略报道后，以下 3 个具体事件在制度结构、权力技术与认知局限层面最值得学者进一步深思：

1. AI 智能体的组织架构化与角色分配 [F11_22 🔍]（版面：第11版/专业栏目）

   • 事件简述：调查显示 23% 的组织已将 AI 智能体正式列入组织架构图并为其分配了正式角色。
   • 制度批判切入点：这标志着权力运作从“工具使用”转向“结构嵌入”。当非人类实体被赋予正式角色，传统的科层制问责机制（Accountability）面临重构，权力在算法执行与组织职能之间形成了新的交互界面。

2. AI 智能体经理（Agent Manager）的职能定义 [F2_3, F2_4]（版面：主报深度版面）

   • 事件简述：公司开始设立“智能体经理”角色，旨在为 AI 智能体制定入职计划并管理其在组织中的运作。
   • 制度批判切入点：这是一种新型的“翻译权力”。管理者的核心能力正在从直接的行政指令转移到对 AI 智能体行为逻辑的定义与引导上，实现了从“人员管理”到“逻辑编排”的权力转移。

3. 认知能力有限性与战略复杂性的矛盾 [F3_6 🔍]（版面：主报深度版面）

   • 事件简述：Felipe Csaszar 指出，人类在规划周期中能承载的可能性和评估的替代方案有限，而 AI 使战略在几何级数上变得更加复杂。
   • 制度批判切入点：这揭示了人类认知生物学局限与数字化战略扩张之间的结构性张力。当战略复杂性超过人类认知阈值，决策权在事实上被移交给能够处理海量可能性的算法系统。

---

▌ 精选核心专题深度思想论证 (In-Depth Dialectical Monograph)

专题一：算法科层制的崛起：AI 智能体嵌入与组织结构的数字化重构

本期报纸揭示了一个深刻的制度转型：AI 不再仅仅是提升效率的插件，而是正在成为组织架构的正式成员 [F11_22 🔍]。当 23% 的组织将智能体列入组织架构图时，我们面对的不再是简单的技术升级，而是一次官僚制逻辑的数字化重构。

从制度现实来看，这种嵌入引发了管理逻辑的根本转变。传统的科层制依赖于明确的等级和人类经理的判断，而现在，企业开始设立专门的“智能体经理” [F2_3 🔍] 来为 AI 制定入职计划 [F2_4 🔍]。这意味着 AI 智能体被赋予了某种“准员工”的制度身份。然而，这种身份的赋予与其实际的工具属性之间存在张力：组织试图通过正式角色来规范 AI 的行为，但 AI 的执行逻辑本质上是基于概率和数据的，而非基于组织文化或伦理共识。

援引韦伯的官僚制 (Weberian Bureaucracy) 理论，传统的科层制通过标准化的程序来消除不确定性。而 AI 智能体的嵌入，实际上是将这种“标准化”推向了极致。当战略决策因复杂性而超出人类认知能力 [F3_6 🔍] 时，组织倾向于依赖 AI 来评估替代方案。这种技术延伸导致了一种新的权力分布：权力不再仅仅集中在拥有职位的领导者手中，而是部分转移到了能够定义 AI 行为逻辑的“智能体经理”以及支撑 AI 运行的算法架构之中。

在这种治理范式下，真正的制度风险在于，当组织习惯于依赖 AI 简化决策流程 [F1_1 🔍] 时，人类领导者可能会在认知上产生依赖，导致对复杂制度环境的直觉感知退化，最终使组织陷入一种由算法定义的、缺乏灵活性的“数字铁笼”之中。

专题二：激励机制的认知偏差：从财务奖励到身份认同的权力操纵

本期报纸中关于现金奖励与创新 [F8_15, F8_16] 以及员工身份与响应 [F9_17, F9_18] 的研究，共同揭示了现代企业如何通过精巧的激励技术实现对个体行为的引导。

首先，研究通过分析中国上市公司的专利数据发现，现金奖励并不总是能产生更好的创意 [F8_15, F8_16]。这一发现挑战了传统的“经济人”假设，即简单的财务激励可以线性地提升创新产出。它揭示了激励机制在面对复杂创造性劳动时的失效，证明了单纯的物质驱动无法覆盖创新的深层心理需求。

其次，普鲁德霍姆（Prud’homme）指出，有效的创新管理需要考虑员工身份的差异 [F9_18 🔍]，因为员工倾向于产生符合其身份的受奖结果 [F9_17 🔍]。例如，专家型人才更倾向于强调创造力。这揭示了一种更深层的“身份规训”：企业不再通过统一的奖励标准来管理员工，而是通过识别并利用员工的“身份认同”来引导其产出。

从批判视角看，这是一种从“强制性激励”向“认同性引导”的权力转移。当企业意识到“奖励活跃度获得更多想法，奖励影响力获得更好想法” [F10_20 🔍] 时，管理层实际上是在通过操纵奖励的定义，来定义什么是“有价值的贡献”。这种权力运作比直接的财务控制更为隐蔽，因为它将个体的自我实现（身份认同）与组织的绩效目标（影响力/活跃度）进行了深度绑定。

综上所述，企业制度正在通过从“通用激励”转向“精准身份管理”，将权力运作隐匿在对个体心理需求的满足之中，从而实现了更高效率的结构性控制。

---

▌ 面向学者的制度研究开放性追问 (Open Research Horizons)

1. 算法角色的法理地位：当 AI 智能体被正式列入组织架构图 [F11_22 🔍] 时，其在法律意义上的“代理”关系如何界定？如果 AI 智能体在执行正式角色时造成损失，责任链条如何在“智能体经理” [F2_3 🔍] 与算法供应商之间分配？
2. 认知外包与领导力退化：面对几何级数增长的战略复杂性 [F3_6 🔍]，人类领导者将决策支持外包给 AI 后，组织内部的“战略直觉”是否会发生系统性萎缩？这种萎缩如何影响组织在极端非线性危机中的生存能力？
3. 身份激励的异化研究：当企业利用员工的身份认同 [F9_17 🔍] 来设计激励机制时，这种做法是否会导致员工为了获得特定身份的认可而进行“表演性创新”，从而掩盖真实的生产力问题？

▌ 本日全报生活世界痛感与情感政治深思雷达 (Lifeworld & Affective Significance Radar)

在系统通盘梳理本期报纸所涉及的所有报道后，以下 3 个具体议题在个体生命体验、认知局限与组织异化层面最值得学者进一步深思：

1. 【认知局限与战略复杂性的冲突】 [F3_6 🔍]（生活领域：组织决策/认知心理）

   • 事件简述：Felipe Csaszar 指出，战略决策的障碍不在于智力不足，而在于人类认知能力的有限性（疲劳、政治因素、时间压力），而 AI 的介入使战略复杂度呈几何级数增长。
   • 情感政治切入点：揭示了“认知过载”带来的存在性焦虑。当战略复杂度超越人类大脑的承载极限，决策者在面对 AI 驱动的几何级增长的替代方案时，可能会产生深层的无力感与对算法依赖的被迫认同。

2. 【AI 智能体的“非员工化”与身份剥离】 [F11_21 🔍], [F11_22 🔍]（生活领域：劳动与就业/组织架构）

   • 事件简述：尽管约三分之一的领导将 AI 称为“队友/员工”，且 23% 的组织将其列入架构图，但核心论点在于“不应将 AI 智能体当成员工”。
   • 情感政治切入点：探讨“角色定义”的权力斗争。将 AI 排除在“员工”定义之外，本质上是在维护人类劳动的独特性与情感特权，防止人类在组织中的主体地位被纯粹的功能性智能体所稀释。

3. 【激励机制的身份认同错位】 [F8_15 🔍], [F9_17 🔍], [F10_20 🔍]（生活领域：创造力/心理契约）

   • 事件简述：研究中国上市公司数据发现，现金奖励并不总是能产生更好的创意；专家型人才倾向于强调创造力，而奖励“活跃度”与奖励“影响力”产生的结果截然不同。
   • 情感政治切入点：分析“量化激励”对内在动机的侵蚀。当组织试图用统一的财务激励来驱动创新时，实际上忽略了员工基于身份（专家型 vs. 执行型）的差异化情感需求，导致个体在追求“奖金”与追求“创造力”之间产生心理撕裂。

---

▌ 精选核心专题理性与情感辩证论证 (The Dialectic of Logos and Pathos Monograph)

专题一：【认知极限与算法扩张的辩证：论战略决策中的“认知过载”】

系统理性模型解构（Deconstruction of Systemic Logos）： 在 [F3_6 🔍] 的论述中，战略决策被建模为一个在有限认知资源（Cognitive Resources）约束下的优化问题。系统理性认为，AI 的价值在于通过处理几何级数增长的复杂性，来弥补人类在疲劳、政治压力和时间限制下的评估缺陷。在这种逻辑下，AI 是一个“认知增强插件”，旨在将决策过程从“有限理性”推向“计算理性”。

生活世界真实痛感的现象学深描（Phenomenological Pathos）： 然而，对于决策者而言，这种“复杂性的几何级增长”并非简单的计算量增加，而是一种深层的心理压迫。当 AI 能够瞬间生成成千上万个替代方案时，人类决策者面对的不再是“选择”，而是一种“选择的瘫痪”。这种痛感源于主体性的丧失：当人类无法在脑中承载 AI 所呈现的可能性空间时，决策行为从“基于洞察的判断”异化为“对算法推荐的背书”。

Logos 与 Pathos 的辩证摩擦（The Dialectical Friction）： 此处发生了深刻的辩证摩擦：系统理性试图通过 AI 消除“认知局限”，但结果却创造了一种新的、更高维度的“认知囚笼”。AI 越强大，人类在战略航向设定中的“掌控感”就越弱。这种摩擦揭示了技术理性的悖论——旨在增强人类能力的工具，最终可能因为其效率的绝对优势而导致人类决策能力的萎缩。

话语温度差与社会心理韧性诊断（Affective Prognosis）： 对比报告中冷色调的“认知能力有限”与决策者在面对复杂环境时的焦虑感，可见巨大的话语温度差。这种状态预示着一种新的组织心理危机：领导者的自我认同将从“战略愿景的创造者”转向“算法结果的审核员”。社会心理韧性的重建需要承认人类认知的“有限性”并非缺陷，而是一种必要的过滤机制，用以在复杂性中锚定人类的价值判断。

专题二：【激励量化与身份认同的异化：论创新管理中的“影响力”陷阱】

系统理性模型解构（Deconstruction of Systemic Logos）： 在 [F8_15 🔍] 至 [F10_20 🔍] 的研究逻辑中，创新被量化为“专利数量”和“创新程度”。系统理性试图通过建立一套精准的激励函数（$\text{奖励} \rightarrow \text{产出}$）来驱动创意。特别是 [F10_20 🔍] 提出的区分：奖励“活跃度” $\rightarrow$ 获得更多想法；奖励“影响力” $\rightarrow$ 获得更好想法。这是一种典型的行为主义管理逻辑，将人类的创造冲动简化为对特定奖赏的条件反射。

生活世界真实痛感的现象学深描（Phenomenological Pathos）： 然而，对于“专家型人才”而言，创造力往往源于一种非功利性的、基于身份认同的内在驱动（Intrinsic Motivation）。当创造力被纳入“影响力”或“活跃度”的量化考核时，专家体验到的是一种“认同的异化”。他们必须在“追求纯粹的学术/技术突破”与“迎合组织定义的‘影响力’指标”之间进行权衡。这种痛感在于：我的专业价值被简化为了一个可被奖励的指标。

Logos 与 Pathos 的辩证摩擦（The Dialectical Friction）： 此处发生了激励逻辑与身份认同的辩证摩擦。系统理性认为“差异化奖励”是有效的管理工具 [F9_18 🔍]，但这种差异化实际上是在将员工进行“标签化”分类（如专家型 vs. 其他）。当员工意识到自己的行为被预测并被通过奖励引导时，原本自发的创造力可能转化为一种策略性的表演。这种摩擦揭示了管理理性在面对人类复杂情感（如自豪感、好奇心）时的粗糙性。

话语温度差与社会心理韧性诊断（Affective Prognosis）： 研究报告使用“追踪 2,376 项产出”等客观数据，掩盖了创新过程中个体经历的孤独、挫败与顿悟等热色调情感。这种量化管理导致组织内部产生一种“指标焦虑”。社会心理韧性的诊断结果显示：过度依赖量化激励会削弱组织的长远创新能力，因为真正的突破性创新往往发生在那些不被量化为“活跃度”或“影响力”的、看似低效的探索之中。

---

▌ 面向学者的伦理与社会心理开放性追问 (Open Research Horizons)

1. 【认知外包与主体性消亡】：当战略决策的复杂性完全由 AI 承载时，人类领导者的“判断力”是否会演变为一种纯粹的仪式性权力？我们如何定义 AI 时代下人类决策者的核心主体性？
2. 【智能体角色边界的伦理界定】：既然 AI 不应被视为“员工” [F11_21 🔍]，那么在组织架构中为其分配“正式角色” [F11_22 🔍] 之间存在怎样的逻辑矛盾？这种矛盾如何影响人类员工的心理安全感？
3. 【量化激励对内在动机的侵蚀机制】：在不同文化背景（如研究中提到的中国上市公司）下，财务激励如何通过改变个体的“身份认同”来影响创造力的质与量？是否存在一种非量化的、基于认同感的激励模型？