What Executives Get Wrong About AI  ## Metadata - Author: [[John Winsor, Sangeet Paul Choudary]] - Full Title: What Executives Get Wrong About AI - Category: #articles - Document Note: --- tags: [AI, Strategy, Business-models] author: [John Winsor, Sangeet Paul Choudary] source: [Harvard Business Review] reference: [https://hbr.org/2025/11/what-executives-get-wrong-about-ai] date created: 2025-11-10 00:00 --- # 🏷️ Title: What Executives Get Wrong About AI ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ Paraphrase](#️-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) ### 🔗 References - Include links cited in or relevant to the article. ## 📝 Summary - What happened. - The authors argue most firms treat AI as an add-on, not a catalyst to reinvent value creation and operating models. - Why it matters. - “With AI…reinvention, not adoption, should be your goal.” - Key stakeholders. - Executives, product/org leaders, ecosystem partners, creators, enterprise buyers. - Risks and tradeoffs. - Adoption without reframing erodes margins and power; reinvention requires capability redeployment and complement investments. - Bottom line. - Reframe objectives, redeploy distinctive capabilities, and build complements to shift power and capture durable advantage. ## 📖 Main Content - Core arguments - Adoption vs reinvention: “They’re buying the shiny new thing and bolting it on to yesterday’s business model.” - Technology shifts reallocate power across value chains; winners redesign architecture, not just tools. - Historical contrasts - Kodak adopted digital but kept a substitution mindset; failed to build complements (sharing, storage, editing). - Fujifilm redeployed chemistry/imaging capabilities into healthcare, cosmetics, electronics; reduced legacy dependence. - Kmart used barcodes for checkout speed; Walmart rebuilt information/logistics architecture and dictated supplier standards. - Contemporary example - Stock photo incumbents assume scarcity; genAI collapses scarcity and trust/rights. - Shutterstock reframed to governance (trust, rights, indemnity), partnered on data licensing, and created contributor compensation. - Key terms - Complements: ecosystem assets (platforms, partnerships, compensation/indemnity) that unlock value from the core tech. - Reallocation of power: who sets rules, standards, and captures margins post-shift. - Uncertainties - Long-term sustainability of compensation/indemnity models; regulatory evolution around training data and copyright. ### 🗒️ Notes - “Ninety-five percent of AI projects fail…if you scope your project using the wrong objectives…you might…still set yourself up to lose.” - Kodak: digital killed scarcity; margins migrated from capture to organize/share/store; Ofoto reframed as printing—missed platform play. - Fujifilm: redeployed capabilities to skincare, diagnostics, LCD/pharma; proactively shrank film reliance; profitable by 2012. - Stock photos: genAI undermines scarcity; real issue is trust, authorship, rights. - Shutterstock: reframed as governance engine; data-licensing with OpenAI/NVIDIA; contributor pay + customer indemnity. - Barcodes: Kmart local optimization vs Walmart system-level rewiring (EDI, satellite, centralized logistics) → supplier standard-setting. - Prescription: Reframe, redeploy distinctive capabilities, build complements; aim to shift power, not just efficiency. ## 📚 References - What Executives Get Wrong About AI (HBR) (https://hbr.org/2025/11/what-executives-get-wrong-about-ai) - Shutterstock data partnerships announcement (OpenAI) (https://openai.com/blog/shutterstock-partnership) - NVIDIA generative AI data partnerships (https://www.nvidia.com/en-us/industries/media-entertainment/) - Walmart retail barcode/EDI history overview (https://corporate.walmart.com/newsroom) ## 🗣️ Paraphrase - Paragraph A paraphrase. - Most firms chase AI adoption metrics; without reframing objectives, they optimize the old model and erode their future position. - Paragraph B paraphrase. - Advantage comes from reorganizing work, reallocating decisions, redeploying capabilities, and assembling complements that shift power. ## 📰 Newsletter - This week: Why AI ROI stalls. HBR’s Winsor and Choudary show wins come from reinvention—not pilots. Learn from Kodak vs Fujifilm, Kmart vs Walmart, and how Shutterstock became a governance layer for synthetic media. Resources: HBR analysis, OpenAI–Shutterstock licensing, NVIDIA media stack. Tip: Map your complements (trust, rights, indemnity) before shipping “AI features.” ## ✅ Action Items #### DevRel - Launch examples showing “complements” (governance, indemnity) not just models. - Build quickstarts for data licensing, audit trails, provenance. - Measure developer adoption of governance SDKs, not just inference calls. #### OSPO - Define policy for training-data licensing and contributor compensation. - Set review cadence for AI bill of materials and provenance signing. - Establish embargo/incident playbooks for model misuse. #### Community - Run roadmap clinics on “reframing outcomes.” - Pair contributors on trust/rights modules; publish impact reports. - Convene cross-project drills on coordinated disclosures. #### WordPress - Core AI: evaluation sets, red-team pilots, provenance hooks. - Docs: examples-first patterns for AI complements (logging, indemnity). - Plugins/Themes: provenance metadata, usage disclosures, opt-in telemetry. - Accessibility: add TTS summaries; caption quality checks for media AI. #### The WPCC - Fund governance artifacts (charters, operator agreements). - Sponsor contributor stipends for AI provenance and rights tooling. ### FAIR - Enforce signature checks and local metadata stores. - Publish funding transparency and sunset clauses for AI pilots. ## 🧪 Hypotheses - Firms that invest in complements (provenance, indemnity, compensation) capture higher post-AI margins than feature-only adopters. - Redeploying distinctive capabilities predicts new profit pools within 24 months. - AI programs with “power shift” KPIs outperform those with efficiency-only KPIs. - Data-licensing partnerships reduce legal incidents and increase enterprise adoption. ### 📊 Methodology - Comparative case studies (matched pairs) on margin shifts post-AI. - Event studies on partnership announcements vs enterprise uptake. - Governance-maturity scoring vs revenue mix changes. - Legal incident rate analysis pre/post indemnity launch. ### ❓ Questions - Which complements most reliably create switching costs in AI-heavy markets? - How should firms measure “power reallocation” leading indicators? - What compensation models sustain creator ecosystems long term? ### 🔮 Future Research - Standardized AI provenance schemas and cross-platform verification. - Regulatory impacts on training data markets. - Organizational designs that accelerate capability redeployment. ### 💡 Implications - Practice: Budget for complements before model rollouts. - Policy: Encourage transparent licensing and provenance. - Security: Treat AI governance as core control plane. - Education: Train leaders on architectural, not tool, thinking. - Funding: Tie investment tranches to governance and partner milestones. ### 🧠 Learning Styles Assessment - Reading level estimate: Flesch-Kincaid ≈ Grade 10–11. #### 👀 Visual learners - Present: Clear contrasts (Kodak/Fujifilm, Kmart/Walmart). - Improve: Architecture and value-chain power maps. #### 🔊 Aural learners - Present: N/A. - Improve: 3-minute audio abstract. #### 🗣️ Verbal learners - Present: Definitions (complements, power shift). - Improve: Glossary and analogies. #### 🧩 Kinesthetic learners - Present: N/A. - Improve: Step-by-step reframing workshop. #### 🧠 Logical learners - Present: Causal chains (tech → architecture → power). - Im - Summary: Most companies fail to realize that true advantage with AI and other transformative technologies comes not from mere adoption but from reinvention—rethinking business models, value creation, and organizational architecture. - URL: https://hbr.org/2025/11/what-executives-get-wrong-about-ai Jorg Greuel/Getty Images #### Summary. Most companies fail to realize that true advantage with AI and other transformative technologies comes not from mere adoption but from reinvention—rethinking business models, value creation, and organizational architecture. Historical examples such... Ninety-five percent of AI projects fail. [MIT’s Project NANDA](https://hbr.org/2025/08/beware-the-ai-experimentation-trap) has made the number famous, and it now opens countless strategy sessions and keynote addresses. Leaders naturally ask how they can join the 5% that succeed. But that question ignores the real issue, which is that if you scope your project using the wrong objectives, you might build the “right” tools and hit the “right” milestones—but still set yourself up to lose. That’s what’s happening with AI today: Companies are rushing to adopt it but aren’t thinking about it as a tool for reinvention. They’re buying the shiny new thing and bolting it on to yesterday’s business model. They’re thinking of it merely as an accessory, not as a powerful new lever for redesigning their whole system of value creation. To behave in that way is a recipe for failure. If you’re thinking about experimenting with AI, it’s time to stop obsessing with adoption and start thinking hard about reinvention. In this article, we’ll discuss historical examples of how these two very different approaches to new technology have played out—and we’ll show how they lay out a way forward for leaders wondering how best to engage with AI. #### **Image Capture** Consider the infamous story of Kodak, which is usually told like this: In 1975, the company invented the first digital camera but didn’t recognize the technology’s transformative potential and so only adopted it late, after other, more prescient companies had already taken over the space. But adoption wasn’t actually the issue for Kodak. By 2005, in fact, the company was the leading seller of digital cameras in the United States and a leading provider of digital-printing services. Nevertheless, just seven years later, it went bankrupt. To understand what happened, you have to remember that film photography was built on the scarcity of captured images, which were labor-intensive and expensive to produce. As a result, film and printing carried margins of 60% to 70%. Digital photography eliminated that scarcity. In a world where images could be produced without limits and digital cameras were a low-margin hardware business, value migrated away from the act of taking photos to the systems that organized, shared, and stored them. Kodak didn’t adapt to that new reality. After adopting digital cameras, the company treated them as a simple product substitution for film cameras—and in so doing dismantled the high-profit engine that had sustained it for decades. Even more important, it failed to invest in the value-generating complements that would define the new ecosystem: photo-sharing platforms, editing tools, storage services. Kodak did acquire Ofoto, an early online photo-sharing startup, but it reframed it as a printing service, rather than a social or sharing platform. All of this resulted in a double blow: The substitution mindset eroded profits, and the absence of complements left the company without new profit pools to migrate to. By adopting a new technology with the wrong objectives, the company accelerated its own decline. Kodak’s competitor Fujifilm responded very differently to the same challenge. Recognizing that the days of using film as an engine for profits were over, the company drew on its deep expertise in chemicals, imaging, and materials science and quickly expanded into adjacent domains, using antioxidants it had developed for film preservation to create cosmetics and skincare products, applying its imaging capabilities to medical diagnostics, and leveraging its materials-science expertise to develop pharmaceuticals and LCD screen components. At the same time, the company aggressively reduced its reliance on the legacy film business. By 2012, when Kodak filed for bankruptcy, Fujifilm had successfully reinvented itself: It was profitable and growing, with revenues exceeding $25 billion and with strong earnings from healthcare, document solutions, and electronic materials. The lesson that these stories teach is clear: Merely adopting a new technology does not guarantee advantage. The key is to determine where the new value lies, how to redeploy its distinctive capabilities toward capturing that value, and how to invest in the right complements. #### **Taking Stock** Today gen AI and its image models pose a threat to stock photography that’s as existential as the threat that digital cameras posed to film. Most stock-photo platforms that license photos for commercial use are built on the assumption of scarcity of high-quality images, but that scarcity no longer holds, because anybody using a gen AI model can generate the images they want with simple prompts at virtually no cost. Stock-photo platforms have responded mainly in one of two ways, either by adopting AI as a tool to more efficiently tag and categorize the photos in their archives, or by rejecting the technology altogether. Both of these responses are anchored in the old economics of scarcity, and both fail to recognize that the real challenge confronting stock agencies is not competition from cheap synthetic images but the way in which they lead to a collapse of trust, authorship, and rights, which are critical for content to be commercially licensed. Shutterstock, confronted with the same shift, didn’t treat gen AI as a tool to be adopted or a threat to be avoided. Rather than defending its traditional business in an era of synthetic content, the company shifted its focus to the management of trust, authorship, and rights. Next, it applied its strengths in metadata and rights management, along with its relationships with contributors and enterprises, to build a system that makes AI-generated content safe for enterprise use. It then built the complements required to make that system work: It established data-licensing partnerships with OpenAI and NVIDIA, and, most important, set up compensation models that ensure fair payment for contributors and indemnification frameworks that protect customers from copyright risk. Through that combination of reframing, redeployment, and complement-building, Shutterstock has moved from being an aggregator of photos to a governance engine. In a world of infinite photos, it sets the rules of participation for both AI-model developers and content users. Model developers, [OpenAI among them](https://www.theverge.com/2023/7/11/23791528/openai-shutterstock-images-partnership), now depend on Shutterstock for rights-cleared access to its photo library as a training set. Enterprise buyers depend on Shutterstock for [legal protection](https://www.shutterstock.com/blog/ai-generated-images-indemnification), while [contributors](https://investor.shutterstock.com/news-releases/news-release-details/shutterstock-unveils-trust-its-best-class-approach-ethical-ai) trust it for fair compensation and visibility. #### **A Reallocation of Power** In every technological shift, real advantage shows up as not just higher profits or new business lines but also a reallocation of power within the value chain. That’s what happened in the case of the U.S. retailers Walmart and Kmart. When barcode technology was adopted in retail in the mid-1970s, both companies embraced it early, but with very different results. Kmart used the technology to speed up checkout lanes but didn’t recognize that it could also make possible a new organizational model. Decision-making was decentralized, with store managers responsible for purchasing and inventory, and the company kept it that way. Walmart, by contrast, treated barcodes as not only an efficiency tool but also a system-level capability that could rewire information and coordination across its entire retail network. It invested in complementary infrastructure, including a satellite-communications network, an electronic-data interchange with suppliers, and centralized logistics. This new architecture allowed the company to restock proactively and forecast demand with precision, which gave it the power to standardize supplier practices. Suppliers now had to operate on Walmart’s terms, conforming to its delivery standards and pricing expectations. ##### . . . Most companies today have not learned the lesson that these stories teach about new technologies, which is that unless you get the framing right, adoption alone may simply accelerate your decline. Each technological shift redistributes power, as we noted in the examples above, but winners rarely emerge merely by adopting technology first. The winners, all too often, are those who use the technology to reimagine the architecture of their business—how decisions are made, how capabilities are deployed, and how value and control are redistributed across the value chain. Kodak and Kmart adopted the right technologies but aimed them at the wrong problems. Fujifilm, Shutterstock, and Walmart, by contrast, reimagined their game in light of what the new technology made possible. With the wrong objectives, a company can execute successful pilots and adoption programs but fail to create advantage. With AI, as with previous new technologies, reinvention, not adoption, should be your goal. What matters most is whether your experiments with the technology will help you reframe work, redeploy distinctive capabilities, and assemble the complements that create durable advantage. Unless you do that, AI will accelerate change—but not necessarily in your favor., What Executives Get Wrong About AI, John Winsor, Sangeet Paul Choudary, 01k9q4v15q14mcxzh08ka9qk4q.