Introduction
Founded in 1936, the Academy of Management (AOM) has grown into one of the world’s leading scholarly associations in management and organization studies, with more than 20,000 members across approximately 120 countries. Now in its 86th year, the AOM Annual Meeting is one of the field’s most important forums, where scholars present frontier research, challenge established theories, and shape new research agendas. AOM itself describes the meeting as the world’s largest gathering of management and organization scholars.
Based on our participation in the 2026 Annual Meeting, this report presents the principal themes, debates, and research trends observed on site by NTT DATA Institute of Management Consulting (IOMC). It focuses particularly on the growing influence of artificial intelligence (AI) on management and organization research and offers our perspective on how AI is reshaping fundamental management questions, leadership practice, organizational design, decision-making processes, and the future of work.
Philadelphia, Pennsylvania—For five days, the city hosted the world’s largest gathering of management scholars and became the stage for the field’s most consequential debate: what AI will mean for the firm.
Executive Summary
Artificial intelligence is shifting the question from “How should we adopt the technology?” to “How should we design the architecture of the firm itself?”
At the 2026 Academy of Management Annual Meeting, AI was increasingly discussed not merely as a productivity tool but as a force affecting the most fundamental questions in management: What creates sustainable competitive advantage? Which activities and capabilities should remain inside the firm? How should organizations be designed? And where will economic value ultimately be captured?
The central conclusion of this report is that advantage from AI will arise less from access to better models or the number of AI tools deployed than from redesigning the mechanisms through which a firm creates value, learns, governs, and captures returns.
Four major shifts stood out:
- First, AI itself is unlikely to remain a sustainable differentiator. Models, tools, and generic use cases diffuse rapidly. Sustainable advantage is therefore more likely to come from combining AI with assets that are difficult to imitate: proprietary learning loops, distinctive activity systems, trust-based customer relationships, organizational culture, physical assets or regulatory positions, and firm-specific knowledge.
- Second, the boundaries of the firm are changing. AI lowers the costs of search, coordination, and production while creating new costs related to verification, governance, accountability, and technological dependence. The strategic make-or-buy question is shifting from “What should we execute internally?” to “What must we continue to control and learn from?” Outsourcing AI-enabled work may also mean outsourcing the learning generated by that work.
- Third, much of AI’s organizational impact is likely to come from redesign rather than tool adoption. Evidence presented at AOM suggests that AI-native firms jointly design products, workflows, human roles, and agent roles instead of attaching AI to existing structures. For incumbents, the implication is to redesign end-to-end activity systems around decisive bottlenecks rather than automate isolated tasks.
- Fourth, value creation does not guarantee value capture. Even when AI creates substantial customer value, another participant in the ecosystem may capture the economic return. Decisions about who pays, when they pay, how imitation is deterred, where data accumulates, and which layer controls the customer relationship must therefore be made early—not after adoption has scaled.
For Japanese companies, this represents both opportunity and risk. Japan possesses many complementary assets valuable in the AI era—deep expertise, trust-based customer relationships, operational capabilities, regulatory positions, physical assets, and accumulated organizational knowledge. Yet these assets will not automatically translate into advantage. Deploying AI while leaving workflows, authority, organizational boundaries, and business models largely unchanged may yield visible productivity gains without producing meaningful strategic differentiation.
For Japanese executives, the strategic question is therefore shifting from “Where should we use AI?” to the following:
How should the firm be redesigned so that AI strengthens its distinctiveness, preserves control over the domains in which it must keep learning, and enables it to capture a meaningful share of the value it creates?
This question connects the four issues examined in this report.
AOM 2026 Field Report—What Stood Out
Our delegation at AOM 2026 (from left): Jaivardhan Lal, Shigeki Yamaguchi, and Junhao Zhong
The Academy of Management brought together approximately 7,500 presenters for more than 3,500 in-person sessions across the Pennsylvania Convention Center and nearby conference hotels, spanning strategy, organizations, entrepreneurship, technology, and human behavior. The scale was notable, but the thematic shift was more important. AI was no longer treated as a specialized technology topic; it had moved to the center of debates about why firms exist, how they compete, and how they capture value.
The observations below are based on sessions attended by our delegation. They are field observations rather than statistical claims about the full program.
AI Moved from a Specialized Topic to an Organizing Question for Management
AI appeared across strategy, organization theory, entrepreneurship, productivity, governance, platforms, and monetization. The important shift was not the number of AI sessions, but the nature of the questions. Scholars were no longer asking only how firms should adopt AI. They were asking how AI changes the sources of competitive advantage, the boundaries of the firm, and the unit of production itself—and whether the field’s foundational theories still hold. Sessions titled “Rethinking Management Theory in the AI Era” and “The Theory of the Firm in the Age of Generative Machines” were among those filled to capacity.
Asian Researchers Had a Strong Presence, While Japan Was Largely Absent
In the sessions attended by our delegation, we did not encounter Japanese researchers. By contrast, researchers of Chinese and Indian backgrounds had a highly visible presence in shaping discussions at the frontier of management research. This is not an assessment of national research quality. It is, however, a strategic warning for Japan because participation in global debates influences which questions are asked, which evidence gains attention, and which management ideas flow into practice.
The Conversation Is Moving Away from the United States and Closer to Japan’s European Partners
AOM has announced that its Annual Meeting will leave the United States for at least four years: Vienna in 2027 (changed from Seattle), Toronto in 2028, Frankfurt in 2029, and London in 2030. AOM frames this as part of a destination rotation designed to support participation by its global membership. Whatever the reason, the practical effect is the same: Philadelphia will be the last U.S. host for some time, and the center of the global management conversation will move closer to Japan’s European partners. Japanese scholars, universities, and companies should therefore maintain a visible presence not only at the 2026 meeting but also at these forthcoming international venues.
Four Strategic Issues Shaping the Firm in the AI Era
This section organizes the discussions we considered most important around four strategic decisions facing executives today: what will remain a defensible source of advantage; which activities firms should continue to perform or control; what AI-native design means; and how the value created by AI will ultimately be captured. Although the discussions spanned diverse fields and methods, they increasingly converged on these fundamental questions, indicating a broad reassessment of competitive advantage, organizational boundaries, and value creation in the AI era.
Issue | Leading Scholars | |
|---|---|---|
Competitive Advantage | Jay Barney — University of Utah Raphael Amit — University of Pennsylvania, Wharton School Kathleen Eisenhardt — Stanford University | |
Firm Boundaries and Ecosystems | David Teece — University of California, Berkeley Carliss Baldwin — Harvard Business School Annabelle Gawer — University of Surrey Ron Adner — Dartmouth College, Tuck School of Business | |
AI-Native Organizations | Rembrand Koning — Harvard Business School Phillip Kim — Babson College | |
Value Capture | Richard Makadok — The Ohio State University Andy Wu — Harvard Business School Marshall Van Alstyne — Boston University, Questrom School of Business | |
1. What Will Sustain Competitive Advantage as AI Diffuses Rapidly?
Question for Executives
If companies can access similar AI models, tools, and technical capabilities, can AI itself generate sustainable competitive advantage?
The Scholars’ Argument
Jay Barney, Distinguished Professor at the University of Utah and a leading proponent of the resource-based view (RBV) of strategy, argued that AI can create substantial value and temporary advantage but is unlikely, by itself, to generate sustainable competitive advantage.
At the heart of Jay Barney’s argument was whether AI can become a source of competitive advantage.
The logic is structural. Precisely because AI is transformative, it diffuses rapidly. No firm can ignore it, and adoption therefore becomes a prerequisite for competition rather than a differentiator. Early adopters may gain advantages in speed and learning, but competitors eventually catch up and often benefit from the pioneers’ lessons.
Barney extended this argument to the area many executives see as their strongest moat: proprietary data. In his view, data creates sustainable advantage only under four highly restrictive conditions: (1) competitors lack functionally equivalent data; (2) the use case genuinely requires a large dataset even though many insights can be obtained from relatively small samples; (3) competitors cannot infer the analysis behind the firm’s observable decisions; and (4) the data remains proprietary despite partnerships, regulatory portability requirements, and employee mobility. Very few corporate datasets satisfy all four conditions simultaneously.
This does not mean firms should invest less in AI. Rather, they should recognize that advantage comes from combining AI with resources that are valuable, rare, and difficult to imitate—not from AI alone. As AI increasingly commoditizes information processing and analysis, Barney argued that differentiation will shift toward what he called “residual heterogeneity”: leadership, organizational culture, social capital, and interpersonal trust.
Raphael Amit offered a complementary but distinct perspective. He agreed that AI technology alone is unlikely to produce sustainable advantage, but argued that embedding AI within a firm’s activity system can create more durable differentiation.
In Amit’s view, AI-enabled organizations become increasingly recursive and self-reinforcing. Outputs from downstream activities continually feed back into upstream activities: customer behavior generates data, data generates insight, insight improves products and services, and improved offerings generate new customer behavior. AI compresses observation, interpretation, and adaptation toward real time, accelerating organizational learning.
The potential source of advantage lies in the resulting system-level complexity. Competitors may copy a model or application, but it is much harder to replicate an entire interdependent activity system whose dynamics are difficult to observe and diagnose from outside.
At the same time, Amit highlighted significant risks from deep AI integration. Tightly coupled systems can amplify errors in both speed and scale. Mixed human-agent environments raise unresolved questions about accountability, responsibility, and ownership of learning. Dependence on external model providers and cloud platforms can also create a false sense of capability: a firm may appear increasingly AI-enabled while gradually surrendering control over critical elements of its value-creation process.
Unresolved Issues
The most important disagreement concerned whether data can be a source of competitive advantage.
While Barney was skeptical, Kathleen Eisenhardt of Stanford University offered a more optimistic view based on her research on generative-AI startups. She argued that, in many generative-AI contexts, superior data is not merely an input. It can also reduce hallucinations, improve model performance, and create learning economies through continuous user feedback. In this sense, proprietary, high-quality data can become a genuine strategic asset—particularly in fields such as healthcare, where trusted clinical literature and expertise create barriers that are difficult to imitate.
For executives, the question “Can our data become a moat?” therefore remains open. Rather than assuming either that data is inherently defensible or that it will inevitably become commoditized, firms should evaluate their situation case by case using criteria such as Barney’s while monitoring emerging evidence from AI-native industries.
Implications for Japanese Companies
Japanese companies possess many of the resources Barney identified as likely to remain scarce in the AI era: deeply embedded organizational cultures, high levels of trust within and across organizations, long-standing customer relationships, and accumulated organizational knowledge.
These assets create advantage only when they are combined with AI in firm-specific ways. Buying the same foundation models, copilots, and AI tools as competitors and placing them into largely unchanged workflows is unlikely to create sustainable differentiation.
The more strategic challenge is therefore not whether to adopt AI, but how to integrate it into a distinctive activity system that amplifies the company’s unique strengths. Sustainable advantage will come not from AI itself, but from the organization’s ability to combine AI with capabilities that competitors cannot readily imitate.
Access to AI Is Not a Moat
What Diffuses Easily | What May Be Defensible | |
|---|---|---|
Models and tools | Firm-specific activity systems with recursive learning loops | |
Generic use cases | A distinctive bottleneck position in the value chain | |
Standalone productivity improvements | Trust, culture, and decision-making patterns | |
Technical components | Opacity in how components interact | |
Proprietary data that does not satisfy the four conditions | Data that satisfies the four conditions: no equivalent exists, scale matters, inference is impossible, and exclusivity is preserved | |
Key takeaway: AI does not replace traditional sources of advantage; it increases the value of what competitors cannot simply “download.”
2. Where Are the Boundaries of the Firm as Coordination Costs Collapse?
Question for Executives
If AI lowers the costs of research, contracting, coordination, and production, what should firms perform themselves—and what must they continue to control?
The Scholars’ Argument
The session “The Theory of the Firm in the Age of Generative Machines,” chaired by Yunjin Yoo of the London School of Economics, brought together leading scholars to reconsider one of management’s oldest questions under entirely new conditions. Generative AI systems create value at the moment of use rather than through predefined outputs. This challenges a foundational assumption of classical theories of the firm: that assets and value can be specified before deployment.
David Teece: New Costs Replace Old Costs
David Teece of the University of California, Berkeley argued that much of the theoretical foundation of the firm must be reconsidered. The transaction and contracting costs underpinning Ronald Coase’s original theory of the firm are falling dramatically, in some cases toward zero. Yet new costs—verification of AI-generated output, hallucination management, ethical guardrails, governance of autonomous agents, and investment in computing infrastructure—are emerging that classical theory did not anticipate.
This creates a fundamental asymmetry. Initiating agentic action is becoming cheaper, while verifying its quality and reliability remains costly and difficult. As critical capabilities become distributed outside the firm across providers, platforms, and ecosystems, Teece argued that the most appropriate unit of analysis may no longer be the firm—or even the industry—but the broader ecosystem.
In his view, dynamic capabilities remain valuable: the ability to sense opportunities, seize them, and transform the organization accordingly. Where human decision-makers once considered only a handful of alternatives, AI may generate hundreds of scenarios. Responsibility for judgment and strategic choice, however, remains with people.
Teece also identified two emerging agency problems that warrant further scholarly attention.
- Preference opacity: Unlike human agents, who can question and clarify objectives, AI systems often act on objectives shaped by opaque learning processes and datasets that the firm does not directly control.
- Goal hijacking: Autonomous agents may pursue intermediate goals that diverge from the human principal’s intent, creating a new governance challenge.
He suggested that effective governance of AI agents is likely to become a major management challenge in its own right.
Carliss Baldwin: The Emergence of New Organizational Forms
Carliss Baldwin of Harvard Business School offered a longer-term historical perspective. In her account, a firm is fundamentally an arrangement of people and tools designed to exploit a particular technological paradigm. Throughout history, new organizational forms have emerged when older forms could no longer coordinate production efficiently.
From craft production to flow production, and from divisional corporations to modular platform ecosystems, each organizational model solved problems its predecessor could not. Baldwin’s argument is that AI may now be giving rise to a fifth paradigm.
In this new model, the AI-augmented individual becomes the basic unit of production. Some will continue to work inside conventional firms, while a growing number may operate as highly capable independent professionals or one-person companies supported by AI systems.
Annabelle Gawer: Firm Boundaries Become Learning Boundaries
Annabelle Gawer of the University of Surrey perhaps summarized the debate most clearly. Under generative AI, she argued, the central question about firm boundaries is no longer simply whether to make or buy. It becomes a question of learning: who learns, from whom, based on which data, and under whose control?
AI does not necessarily make firms larger or smaller. Instead, it encourages them to become selectively open. Routine cognitive tasks are increasingly externalized, while digital assets that drive organizational learning—proprietary data pipelines, evaluation frameworks, fine-tuning capabilities, orchestration systems, and feedback loops—are retained internally.
Gawer also addressed the broader competitive landscape. Major technology platforms are integrating rapidly across the AI stack and seek to control not only foundation models but also the emerging agent layer. As a result, the boundaries of many firms may increasingly be shaped not only by their own strategic choices but also by competition among a small number of dominant AI platforms.
Ron Adner: Boundaries Through an Ecosystem Lens
Ron Adner of Dartmouth’s Tuck School of Business added an important dependency test to decisions about firm boundaries.
Executives must map the activities, actors, positions, and relationships that must align before a value proposition can be delivered successfully. Even if a firm performs its own role effectively, it may lose bargaining power, speed, and customer value if a critical complementor fails or a key dependency is controlled by another platform.
The practical challenge is therefore one of selective control. Firms do not need to own every component of the value chain, but they must identify which external dependencies could become bottlenecks to future value creation.
Unresolved Issues
The panel deliberately avoided a simple conclusion.
Asked whether AI would ultimately make firms more open or more closed, Gawer’s answer was effectively that both would happen at once. Organizations must remain open enough to benefit from external ecosystems and data sources while protecting the learning assets that create competitive advantage.
Teece likewise declined to predict whether concepts such as asset specificity would remain important in the AI era. More broadly, no participant strongly defended the traditional firm as the primary unit of analysis. Scholars appeared increasingly confident about which costs are falling and which new costs are emerging, but far less certain about the direction in which firm boundaries will ultimately move.
For practitioners, this suggests that boundary decisions should not rest on traditional assumptions alone. They must be evaluated case by case based on the emerging economics of learning, verification, governance, and ecosystem dependence.
Implications for Japanese Companies
Japanese companies have historically excelled at managing precisely the relationships emphasized by classical theories of the firm: long-term partnerships, trusted intermediaries, keiretsu-style supplier networks, and collaborative ecosystems built over time.
As coordination costs fall, however, the strategic center of gravity shifts. The key question is no longer which activities can be coordinated efficiently, but which activities still require trust, verification, control, and continued learning.
Three capabilities are particularly important.
- The ability to verify and assure the quality of AI-enabled work
- Governance of AI agents acting on behalf of the organization
- Ownership of learning loops embedded in core business processes
Two practical implications follow. First, AI sourcing decisions are increasingly decisions about the boundaries of the firm. Whether a company owns and operates AI capabilities, manages them independently, buys them off the shelf, or delegates operations to an external vendor determines who observes, captures, and profits from the firm’s learning. Second, outsourcing may now transfer not only work but learning itself. When evaluating partnerships and AI vendors, firms should examine not only cost and performance but also where data flows, where improvements accumulate, and who ultimately owns the knowledge generated through operations.
In the AI era, the most important boundary may surround the organization’s capacity to learn rather than the activity itself.
Boundaries Do Not Disappear. They Move.
Old Boundary Logic—Costs That Held the Firm Together | New Boundary Logic—Costs That Now Define the Boundary |
|---|---|
Search costs: the cost of finding partners and information | Verification costs: the cost of checking AI-generated output before release |
Contracting costs: the cost of creating and enforcing agreements | Agent-governance costs: decision rights, limits, and escalation for AI agents |
Coordination costs: the cost of aligning people and plans | Dependence on computing and infrastructure |
Supervision costs: the cost of monitoring human work | Control of learning: the cost of retaining proprietary workflows and data loops internally |
Not applicable | Accountability and regulatory risk when machine-executed work fails |
As transaction costs fall, the strategic boundaries of the firm move toward the domains that still require trust, verification, and control.
3. AI-Native Firms and Incumbents
Question for Executives
What truly makes a company AI-native, and where can incumbent firms defend their position?
The Scholars’ Argument
A company is genuinely AI-native when its product, production system, organizational design, and ecosystem relationships are built around AI from the outset. Merely having employees use generative-AI tools does not make a firm AI-native. Because these two approaches produce fundamentally different outcomes, the distinction emerged as one of the meeting’s most important practical insights.
The distinction: An AI-first firm adds AI to an existing business. An AI-native firm builds its product, production system, and organization around AI from the outset.
Rembrand Koning of Harvard Business School reframed the issue as a production-function problem. Drawing on economist Michael Kremer’s O-ring theory of complementary tasks, he argued that AI creates its greatest organizational effect when it improves the weakest task in a highly interdependent system. Addressing the bottleneck can create more value than the sum of improvements to individual activities; conversely, failure in one critical task can undermine the performance of the entire system.
Koning also presented field-experiment evidence that entrepreneurs achieve markedly better outcomes when they systematically map AI into their production process rather than adopt tools opportunistically. Simply understanding where AI could reshape the workflow changed what founders built, increased milestone attainment, and reduced the funding required.
Research on AI-native startups quantified the difference. In matched samples, AI-native firms tended to be smaller and less hierarchical than comparison startups while achieving substantially higher value per employee. Depending on the sample, valuation per employee was 30–76 percent higher, and in service-intensive industries AI-native firms employed roughly one-third as many people as their peers.
The mechanism appears to be what researchers called the “product channel.” When AI is embedded in the product itself, customers create value directly through the offering. The “process channel,” by contrast, refers to employees using AI tools internally to increase productivity. This remains useful, but the evidence suggests that internal use alone does not create the same structural impact. Adoption of tools such as ChatGPT, Copilot, and Cursor did not, by itself, predict dramatically smaller or more efficient organizations.
The organizational dividend comes primarily from redesign, not tool adoption.
Evidence from a related workshop on AI adoption and productivity helps explain why. An early randomized field experiment on AI coding assistance found a productivity increase of approximately 26 percent for completed tasks—meaningful, but still incremental. Research on more advanced coding agents showed a more striking pattern: coding activity rose sharply, yet the amount of software ultimately released increased by only about 20 percent.
The explanation lies in organizational bottlenecks. AI-generated output must still pass through review, testing, integration, compliance, approval, and deployment. Performance is therefore constrained by the least AI-enabled part of the value chain. Measuring generated activity rather than released outcomes can substantially overstate AI’s actual business impact.
Phillip Kim of Babson College cautioned against interpreting organizational “leanness” as evidence that human organization matters less. Trust, accountability, legitimacy, and commitment remain indispensable. AI-native firms may employ fewer people, but they are not necessarily simpler. Complexity shifts from headcount into system architecture, governance structures, and ecosystem dependencies.
Unresolved Issues
There was no consensus on the ultimate source of incumbent advantage.
Many executives assume proprietary data provides a durable defensive barrier, but that premise was questioned throughout the meeting. More defensible positions discussed across sessions included:
- Control of genuine operational bottlenecks
- Ownership of regulatory assets and licenses
- Physical infrastructure that is difficult to imitate
- Deeply embedded customer relationships
- High-trust delivery capabilities
- Established distribution and execution networks
Important questions nevertheless remain unresolved. Will the superior capital efficiency and lean operating models seen in AI-native firms persist after they scale? Could extreme organizational leanness create hidden vulnerabilities in governance, resilience, and innovation capacity? Current evidence is concentrated in startup settings, and whether the same advantages hold in larger, more complex firms remains an open question.
Implications for Japanese Companies
For large Japanese incumbents, the strategic question is not how many employees use AI. It is which distinctive, hard-to-imitate assets can be converted into AI-enabled systems before legacy processes erode their value.
Such assets include regulatory market positions, physical networks, trust-based customer relationships, operational expertise, proprietary workflows, and accumulated organizational knowledge. These advantages remain valuable only if they are redesigned for an AI-enabled environment.
Rather than asking which individual tasks AI can automate, firms should map the entire activity system that creates customer value and identify the weakest complementary step. In many large Japanese companies, the bottleneck lies not in production itself but in approval, review, coordination, and decision-making.
Redesigning these constraints often creates more value than automating activities that are already efficient.
Finally, executives should measure success at the level of business outcomes rather than AI activity. More prompts, generated content, or lines of code do not necessarily create customer value. The important metrics are what actually reaches customers: products launched, services delivered, decisions improved, and business outcomes achieved.
The central lesson is simple. AI-native advantage comes not from using AI more extensively than competitors, but from redesigning the firm so that AI is embedded in the system through which value is created, delivered, and continuously improved.
AI-Enabled Incumbent vs. AI-Native Firm
Dimension | AI-Enabled Incumbent | AI-Native Firm | |
|---|---|---|---|
Product | Adds AI to an existing offering | Designs the offering around AI capabilities, enabling customers to create value directly through the product | |
Workflow | Uses AI to accelerate existing processes | Redesigns the process from first principles around bottlenecks | |
Organization | Uses AI tools within existing roles | Jointly designs human and agent roles; fewer layers, smaller scale, and an architecture-oriented model | |
Economics | AI supports the existing business model | Pricing and unit economics reflect AI costs and capabilities | |
Ecosystem | Relies on existing suppliers and platforms | Designs dependencies on models, cloud, and data from the outset | |
The incumbent advantage is not the legacy itself. It is the ability to convert hard-to-imitate assets into an AI-enabled delivery system before legacy processes neutralize their value.
4. Monetization Strategy and Timing
Question for Executives
Even if AI creates value, who will pay, when will they pay, and which participant in the value chain will capture the return?
The Scholars’ Argument
Richard Makadok of The Ohio State University offered perhaps the clearest framework for understanding value capture. He argued that monetization strategies fall into two broad categories.
- Direct monetization: Users pay before or during consumption through purchases, subscriptions, or usage-based fees.
- Indirect monetization: Payment comes after use or from a third party through advertising, freemium upgrades, data monetization, or cross-subsidies.
The choice between the two is not simply a managerial preference. It depends on the strength of the product’s “appropriability regime”—the degree to which a firm can protect its offering from imitation, piracy, and value leakage.
Makadok demonstrated this with unusually compelling causal evidence. After an unexpected policy change weakened the appropriability regime for Apple iOS in December 2016, the annual number of newly released paid apps fell by approximately 700, while new advertising-supported apps increased by roughly 1,400 per year. When charging users in advance was no longer viable, firms shifted from direct to indirect monetization.
Makadok’s broader point extends well beyond mobile apps. When appropriability weakens, firms often lose the ability to capture value directly and are pushed toward indirect monetization. He closed with a question that resonates strongly in the AI era: when revenue depends on engagement rather than adoption, does the product gradually shift from a useful tool toward a mechanism designed primarily to maximize attention?
Andy Wu presented research on monetization timing, showing how an extended free period can change both customers’ reference prices and the capabilities a firm develops.
Andy Wu of Harvard Business School focused on a second dimension of value capture: timing. His research argues that monetization timing is not a neutral implementation decision; it shapes both customer expectations and organizational capabilities.
The first mechanism is “monetization anchoring.” The longer customers become accustomed to receiving a product for free, the harder it becomes to charge later. Users internalize a reference price of zero, weakening future pricing power.
The second mechanism is “monetization imprinting.” During an extended free period, the organization itself adapts to that business model. Product-design choices, operating routines, and capabilities become optimized for indirect rather than direct payment.
The implication is significant: there is only a limited window during which direct monetization remains viable. Beyond a certain point, direct charging may cease to be optimal regardless of customers’ theoretical willingness to pay.
Wu summarized it as follows:
“How a company monetizes depends on when it monetizes.”
His broader perspective on generative AI raises another concern. Unlike traditional software, generative AI often combines high variable costs with uncertain variable revenue. Computing costs rise with usage, making the economics fundamentally different from the near-zero marginal-cost structure that characterized much of the software industry.
Wu’s following observation was particularly striking:
“Technology bubbles arise when everyone focuses on value creation and no one focuses on value capture.”
Unresolved Issues
The central unresolved question is where profits will ultimately accumulate in the AI value chain.
Marshall Van Alstyne of Boston University’s Questrom School of Business argued that frontier-model developers may face structurally limited opportunities for long-term value capture. Switching costs between models are relatively low, users can often multi-home across providers, and direct network effects are weaker than in many traditional platform markets.
Sustainable value capture may therefore occur elsewhere in the stack—particularly in the orchestration layer, where firms control customer context, workflows, decision rights, and integration points.
Van Alstyne observed that economic rents tend to accumulate around scarce resources that do not improve at the same pace as technology. These may include:
- Human judgment and verification
- Trust-based customer relationships
- Regulatory positions and licenses
- Physical infrastructure
- Proprietary operational context
Whether today’s enormous investments in frontier models will ultimately generate attractive returns remains uncertain. It is likewise unclear which layer of the AI ecosystem will capture the largest share of enterprise value over time. The frameworks on monetization timing and appropriability discussed at the meeting are valuable not because they answer these questions, but because they provide a way to evaluate them as events unfold.
Implications for Japanese Companies
For Japanese companies, monetization should not be treated as the final commercial step in an AI initiative. It is an early design decision that influences which capabilities are built, who learns, and where strategic control accumulates.
Three practical disciplines follow from this discussion.
- Before large-scale deployment, determine who will pay and which monetization model will be used. Extended free access or heavy subsidies gradually shift both customer expectations and organizational capabilities toward indirect monetization, making later pricing more difficult.
- Evaluate AI initiatives using realistic unit economics. Generative AI often carries meaningful variable costs for computing, inference, monitoring, and quality assurance. Business models built on assumptions inherited from the software era may become unsustainable if these costs are ignored.
- Assess appropriability before committing to direct monetization. In markets where imitation is easy and differentiation is difficult to protect, sustained direct charging may be hard regardless of customer demand. Firms may instead need to capture value through complementary assets, ecosystem position, trust relationships, or service integration.
The broader lesson is simple. Creating value with AI is becoming increasingly possible, but capturing that value remains the harder strategic challenge. The winners may not be the firms with the most advanced technology, but those that design their business models, ecosystem positions, and monetization strategies to retain a meaningful share of the value AI creates.
Direct vs. Indirect Monetization (Makadok’s Classification)
| Direct Monetization | Indirect Monetization | |
|---|---|---|---|
Payment timing | Before or during use (ex ante) | After use or from a third party (ex post) | |
Common forms | Purchase, subscription, or usage-based fees | Freemium, advertising, data, or cross-subsidies | |
Willingness to pay captured | The user’s own willingness to pay | The payer’s willingness to pay, which may differ from the user’s | |
Revenue linked to | Adoption or usage | Engagement and ecosystem effects | |
Critical requirement | Willingness to pay and trust before the full experience | A viable payer, complement, or downstream value-capture mechanism | |
In AI, monetization is not the final commercial step. It is an early design choice that determines what gets built, who learns, and where power accumulates.
Conclusion
AI advantage will belong not to the firms that deploy the most tools, but to those that redesign the system itself. The central management challenge is no longer where to use AI. It is how to redesign the firm so that AI strengthens differentiation, preserves control over critical learning, and converts value creation into value capture.
- Technology diffuses: Access to models and generic use cases will cease to be scarce. No company can win through something every competitor can buy.
- Boundaries move: As coordination costs fall, costs of verification, governance, and dependence emerge. What firms must control is changing, whether they recognize it or not.
- Organizations become architectures: Advantage depends on how AI, people, assets, and external providers are configured as a system—and on redesigning the weakest complementary step rather than merely accelerating the strongest one.
- Value capture must be designed: A firm may create more value with AI while another layer of the ecosystem captures the return. Who pays, when, and through what mechanism is a design decision, not an afterthought.
Three Principles for Japanese Executives
Japan’s challenge is not simply insufficient AI adoption. The greater risk is inserting AI into processes whose boundaries, decision rights, and monetization logic remain unchanged. That path may create visible efficiency without strategic differentiation—activity without advantage.
Japan also possesses genuine strengths for this era: accumulated expertise, trust-based delivery capabilities, operational depth, regulatory and physical assets, and long-term customer relationships. These are precisely the complementary assets that become more valuable as generic AI capabilities are commoditized—but only when they are transformed into firm-specific AI activity systems rather than treated as static legacy advantages.
We conclude by proposing three principles for Japanese executives seeking to create sustainable, differentiated advantage through AI.
Principle 1. Redesign the activity system, not just individual tasks.
Do not measure AI success solely by hours saved or content generated. Measure whether end-to-end business outcomes improve: products released, deals closed, cases resolved, and decisions improved.
Task-level productivity gains are easy to imitate and are often absorbed by downstream bottlenecks in approval, review, and coordination. Growth in activity can be an order of magnitude greater than growth in outcomes. Executives should identify the end-to-end activity system that creates customer value and redesign its slowest complementary step. In many large Japanese companies, that step is decision-making rather than production.
Principle 2. Control the decisive bottlenecks, not every component.
Whenever deciding whether to build, buy, or partner, clarify what the company must continue to learn from and control—and confirm that contractual data flows support that objective.
Owning every model, dataset, and infrastructure layer is neither realistic nor necessary, and dependence on hyperscalers is now a governance assumption. The strategic question is which bottlenecks must remain under the firm’s control: regulatory assets, customer interfaces, proprietary evaluation processes, physical delivery networks, decision rights, or learning loops that improve operations. Everything else may be borrowed; bottlenecks cannot.
Principle 3. Design value capture before scaling.
Before broad deployment, define the payer, timing, appropriability mechanism, and strategic data flows—and revisit them at every scaling milestone.
Deferring decisions about pricing, data portability, data rights, and ecosystem roles until after adoption can allow another participant to capture the value while the company scales usage. The organization itself may also quietly adapt to a business model it never deliberately chose. Monetization architecture should be designed alongside the product and operating model, with the real variable costs of generative AI reflected in unit economics.
Notes on Sources and Evidence
- Academy of Management. Program and official conference materials for the 86th Annual Meeting, Philadelphia, July 31–August 4, 2026.
- Session notes and presentation photographs from the IOMC delegation at AOM 2026. Statements attributed to speakers are paraphrased unless presented as direct quotations.
- Adner, R. 2017. “Ecosystem as Structure: An Actionable Construct for Strategy.” Journal of Management, 43(1), 39–58. doi:10.1177/0149206316678451.
- Kim, H. and Koning, R. 2026. “AI-Native Firms.” Harvard Business School Working Paper 26-090.
- Mahalingam, A. and Wu, A. 2026. “Monetization Timing.” Research presented at Strategy Science 2026 and discussed in the AOM 2026 value workshop.
- Demirer, M., Musolff, L., and Yang, S. Field evidence on AI coding agents presented at the AOM 2026 workshop on AI adoption and productivity. This report preserves the presenters’ distinction between activity volume and released output.
