Beyond the CIO: Who Should Really Lead AI Transformation

You are currently viewing Beyond the CIO: Who Should Really Lead AI Transformation

Mark Blackwell, Arkaro

Executive Summary

  • The default reflex, handing AI to the CIO, treats it as another enterprise software rollout. That reflex is understandable but wrong, because AI is a general purpose technology, not a SaaS licence.
  • Harvard’s Joseph Fuller estimates 60% of companies still treat AI as a technology adoption problem, repeating the mistake factory owners made when they bolted electric dynamos onto steam-driven infrastructure rather than redesigning around what electricity made possible.
  • AI’s clearest near-term value sits in debottlenecking decisions, supply chain, risk, forecasting, which puts the ownership question in the hands of whoever is accountable for that constraint, not the technology function.
  • For asset-heavy industries, Boston University’s Venkat Venkatraman makes the industrial case directly: the winners will build intelligence into the core of the business, not paint it on at the edges. The same category error shows up in innovation, per David Cropley’s research.
  • Charlene Li’s central argument, drawn from advising 49 of the Fortune 100, is that you do not need an AI strategy, you need a business strategy that AI enables, and that ownership belongs with a named individual, not a steering committee.
  • Governance and ownership are not the same thing. Ray Eitel-Porter’s research shows technical guardrails alone cannot close the gap between boardroom AI ambition and frontline AI trust, that gap closes through practice, not policy.
  • The evidence has moved with the argument. BCG’s 2026 AI Radar shows nearly three in four CEOs now call themselves the primary AI decision maker, roughly double the share a year ago.

When leadership, a CEO, a board, sometimes both, decides it is time to “do something about AI,” the instinct all too often is the same. Call the CIO, the person who already owns the servers, the software licences and the IT estate, or the CTO, who more often sits over product and innovation technology than over enterprise-wide decision-making. Either choice feels logical. Both are, in most cases, the wrong call.

Why 60% of AI Projects Fail: The Adoption Trap

The reflex is not stupid. For thirty years, the arrival of a new enterprise technology, ERP, CRM, cloud migration, has followed a predictable pattern: procure it, configure it, roll it out, train people on it. That is a technology function’s home ground, and the CIO, as the executive who typically owns enterprise information systems, is the natural owner of that kind of rollout.

Harvard Business School’s Joseph Fuller, speaking on Arkaro Insights, put a number on why that pattern fails with AI. By his estimate, around 60% of companies are still treating AI as a technology adoption problem, something that can be bolted onto an existing management process the way a new piece of software would be.(1) The consequence, he argues, is that companies run experiments without clean data or trained staff, get disappointing results, and then blame the technology rather than the decision to delegate it upward to the wrong function in the first place.

A general purpose technology, not a SaaS licence

Fuller’s deeper point is that AI belongs to a small and historically consequential category of innovation: a general purpose technology. Economists Timothy Bresnahan and Manuel Trajtenberg coined the term in their 1995 paper, naming the steam engine, the electric motor and semiconductors as the examples that had, in their view, driven whole eras of growth by spreading across nearly every sector, improving continuously, and spawning waves of complementary innovation downstream.(2) Later economists have extended the same label to electrification and computing more broadly, and Fuller places AI in that lineage, arguing that no executive alive has had to navigate the arrival of a general purpose technology before, at this speed and this scale.(1)

Fuller illustrates the point with the arrival of alternating current in American factories in the 1870s and 1880s. Most factory owners bolted the new electric dynamo onto their existing steam-driven infrastructure and captured only modest efficiency gains. The real transformation came later, when engineers realised electrical current, unlike steam, does not need to rise through a building. It follows wires. That single insight collapsed the four and five-storey mill, built to exploit low-pressure steam boilers, into the single-storey factory, freeing the fifth of every floor that manual elevators and work-in-progress inventory had occupied.(1) Companies bolting AI onto today’s processes, rather than redesigning around it, are repeating the mistake with a different technology.

Debottlenecking, not deployment

There is a second Fuller idea that speaks directly to Arkaro’s own territory, and it comes from an earlier interview, given in January 2022, well before the current wave of generative AI adoption. Asked how AI would affect work, Fuller pointed to decision-making as the most obvious application, ranging from debottlenecking the supply chain to evaluating risk.(3) That is a statement about where AI’s value shows up, not a licence to bolt it on. Read alongside his steam-engine argument, the two ideas fit together rather than conflict: a supply-chain bottleneck is exactly the kind of constraint that rewards redesigning the decision process around what AI can now do, and exactly the kind of constraint that yields only marginal gains if AI is simply added to the existing planning meeting as one more input nobody has re-examined.

That framing matters for Arkaro’s own territory, because it locates AI’s payoff exactly where integrated business planning and commercial excellence practitioners already spend their time: in the constraint currently capping throughput, forecast accuracy or service level, not in a shiny new capability looking for a use case.

A bottleneck has an owner already. It sits inside demand planning, supply, commercial or operations, and whoever is accountable for that constraint is the person best placed to judge whether an AI-enabled decision genuinely clears it or merely automates the existing, flawed process faster

Digital paint versus digital structure: the industrial case

For agriculture, food and chemicals businesses specifically, the sharpest recent articulation of this came from Boston University’s Venkat Venkatraman, another Arkaro Insights guest and co-author of Fusion Strategy. His argument is that roughly $75 trillion of global GDP sits in asset-heavy industries, chemicals, food, agriculture, industrials, where the last two decades of digital transformation have so far only touched the edges, in contrast to the asset-light economy of software, media and digital platforms, which was transformed at its core first.(4) Most of these companies are now applying what Venkatraman calls digital paint: adding AI at the periphery of an existing model, a smarter dashboard, a chatbot, without touching the core.(4) The alternative, digital structure, means designing intelligence into the product and the business model from the outset.

His standard illustration is John Deere’s engaged acre, the accumulated data graph of field performance, weather, agronomy and soil conditions across thousands of farms that gives the company a vantage point no individual farmer can match.(4) The bottleneck to building that kind of advantage, he is clear, is not data collection. It is the analytical capability to turn connected data into insight and the customer trust required to earn the right to hold that data in the first place, both business problems rather than technology ones.(4) Whether a mid-sized speciality chemicals or food ingredients company can compete on this basis has nothing to do with which function owns its servers, and everything to do with whether commercial and technical leadership together understand where the industry is heading by 2030.

Supply chain and commercial decisions are not the only processes this applies to. Professor David Cropley, PhD of the University of Adelaide, in a separate Arkaro Insights conversation, makes much the same point about innovation: most organisations are applying AI to it as a product, a tool added at the edge, when innovation is itself a process, and treating it as the wrong category of innovation produces the familiar failure pattern of overrun timelines and underwhelming results.(5) Innovation belongs on the same list as supply chain planning and commercial excellence, one more business process that only benefits from AI once someone accountable for how it actually runs owns the redesign.

Stop building an AI strategy

Charlene Li, adviser to 49 of the Fortune 100 and author of Winning with AI, put the ownership question most sharply in her own Arkaro Insights conversation: you do not need an AI strategy, you need a business strategy that AI enables.(6) If AI is treated as a technology initiative, she notes, it sits in IT. If it is treated as a business capability, it sits with the CEO.(6) Her recommended transformation team includes a technology expert, but she is explicit that this need not necessarily be the CIO, whose role is more often focused on procurement and risk management than on redesigning how the work itself gets done.(6)

Marco Ryan, former Global Chief Digital Officer at BP, makes the same distinction in leadership rather than governance terms. A leader cannot simply paint over the analogue cracks of the business and expect the disruption to pass.(7) Stephen Wunker, whose book AI and the Octopus Organization explores what adaptive structures look like under AI, supplies the statistic that both Ryan and Li’s conversations return to: roughly 80% of executives say AI is core to their strategy, while only 15% of employees believe it, a gap that opens up when the strategy is set at the centre without ever being tested against how value actually gets created by the people running the workflows it depends on.(8)

Ownership over sponsorship

A business sponsor who attends a quarterly steering committee is not the same as a business owner whose targets move with the outcome. Li’s argument is that eventually one person needs to own AI adoption inside the business, someone for whom it is their entire job to ask each morning how the organisation is using AI to create value.(6) That is accountability, and accountability is what produces adoption. Consultation alone produces polite agreement and a pilot that quietly dies.

Named ownership also creates the conditions for something else organisations tend to skip: building the organisational muscle to use AI well. Scott D. Anthony’s argument, explored further below, is that this judgement is built through repeated, low-stakes practice rather than handed down from a lecture or a licence, and that kind of practice only happens once someone is genuinely accountable for the outcome it is meant to improve.(9)

The Data: CEOs Are Taking Over AI Decisions

This is no longer a contrarian view; it is where the wider data has moved too. BCG’s 2026 AI Radar survey of more than 1,250 executives found that nearly three in four CEOs now identify themselves as the primary decision maker on AI, roughly double the share a year earlier.(10) The World Economic Forum’s read on the same research is blunt: companies most likely to succeed treat AI transformation as a business-led undertaking rather than an IT initiative.(11) Separate governance research traces stalled deployments back to the same root cause, accountability sitting with a team that owns platforms and infrastructure but not the business outcome the AI is meant to improve.(12)

Governance is not the same as ownership

This is where the argument connects to a fourth Arkaro Insights conversation, with Ray Eitel-Porter, former head of Accenture’s global responsible AI practice and co-author of Governing the Machine. His central claim is that governance is not what slows AI down, it is what allows an organisation to scale AI with confidence.(13) That claim only holds, though, if governance is understood correctly: as the platform, the guardrails and the escalation routes a technical function provides, not as a substitute for business engagement.

Eitel-Porter’s evidence for why technical controls alone cannot carry AI adoption is the scale of shadow AI. Research he cites shows up to 90% of employees already use personal AI accounts for work while corporate tool usage lags as low as 40%, and he is blunt that pure technical guardrails cannot close that gap.(13) The organisation he advises that moved workforce trust in AI from around a quarter to over 90% did it with a mandatory, in-person, four-hour workshop, not a policy document or a licence rollout.(13) That is the organisational muscle Scott Anthony describes, built in practice rather than declared in a policy, and it is why the tooling question matters less than it feels like it should.(9) Handing the mandate to whichever function is best at procuring software skips the harder, more valuable work of building that judgement inside the teams who own the decisions AI is meant to improve. The division of labour this article has been building towards follows from both: the technology function owns the platform and the guardrails, the business owns whether people actually trust, practise with and use what sits behind them.

Where this leaves the CIO question

So who should lead? The mandate matters more than the title. The right leader is whoever is accountable for the constraint AI-enabled workflow redesign is being asked to relieve, supported by technical colleagues who make the capability trustworthy and safe to depend on. In a food or chemicals business, that is more often the head of supply chain, the commercial director, the innovation leader accountable for how ideas actually get generated and tested, or the IBP process owner than it is the CIO. Their job is not to become AI experts. It is to redesign the decision process the technology now makes possible, and that redesign always runs through people, incentives and culture as much as it runs through a model.

That is precisely why we do not believe AI leadership is a technology appointment at all. It is a change leadership appointment with a technology component, which is why our own approach starts with understanding the business and its context before a line of code or a licence gets discussed, moves through co-creating the redesigned process with the people who will run it, enables the capability and the skills to sustain it, and only then treats the tool itself as settled. Get the ownership question right first. The technology choice becomes far easier once it is answered.

Further reading and listening on Arkaro Insights

This article draws directly on conversations from the Arkaro Insights podcast. For the fuller argument in each guest’s own words:

📺 YouTube: www.youtube.com/@arkaro 🎧 Audio: arkaroinsights.buzzsprout.com

Sources

  1. Joseph Fuller, in conversation on Arkaro Insights, “The Steam Engine Mistake Companies Are Repeating with AI,” published 5 May 2026. https://arkaro.com/joseph-fuller-ai-implementation-strategy-harvard/
  2. Bresnahan, T. F. and Trajtenberg, M. (1995). “General Purpose Technologies: ‘Engines of Growth’?” Journal of Econometrics, 65(1), 83-108.
  3. Fuller, J., interviewed by Devjani Roy, GrowthPolicy / Mossavar-Rahmani Center for Business and Government, January 2022. https://www.hks.harvard.edu/centers/mrcbg/programs/growthpolicy/joseph-fuller-work-twenty-first-century
  4. Venkat Venkatraman, in conversation on Arkaro Insights, “Fusion Strategy: The $75 Trillion Industrial AI Opportunity,” published 10 June 2026. https://arkaro.com/fusion-strategy-75-trillion-industrial-ai-venkat-venkatraman/. See also Venkatraman, V. and Govindarajan, V. (2024). Fusion Strategy: How Real-Time Data and AI Will Power the Industrial Future. Harvard Business Review Press.
  5. David Cropley, in conversation on Arkaro Insights, “Innovation Is Not a Light Bulb Moment, It’s an Engineering Discipline,” published 9 April 2026. https://arkaro.com/engineering-innovation-discipline-david-cropley/. See also Cropley, D. H. and Cropley, A. J. The Psychology of Innovation in Organizations. Cambridge University Press.
  6. Charlene Li, in conversation on Arkaro Insights, “Why AI Transformation Fails, and What Leaders Must Do in 90 Days,” published 3 March 2026. https://arkaro.com/why-ai-transformation-fails-leaders-90-days-charlene-li/. See also Li, C. and Walsh, K. Winning with AI: The 90-Day Blueprint for Success.
  7. Marco Ryan, in conversation on Arkaro Insights, “Rewire or Retire: Why AI Is a Leadership Issue, Not a Technology Problem,” published 14 January 2026. https://arkaro.com/rewire-retire-ai-leadership-marco-ryan/
  8. Stephen Wunker, in conversation on Arkaro Insights, “AI and the Octopus Organization,” published 17 October 2025. https://arkaro.com/ai-octopus-organization-stephen-wunker/. See also Wunker, S. and Brill, J. AI and the Octopus Organization.
  9. Scott Anthony, in conversation on Arkaro Insights, “Why Playing Games at Work Isn’t Childish, It’s a Competitive Advantage,” published 19 May 2026. https://arkaro.com/playing-games-at-work/
  10. Boston Consulting Group, “AI Radar 2026,” survey of 1,250+ executives, published May 2026.
  11. World Economic Forum, “CEOs are all in on AI but anxieties remain: What leader confidence indicates for 2026,” January 2026. https://www.weforum.org/stories/2026/01/ceos-are-all-in-on-ai-but-anxieties-remain/
  12. iQuasar Software, “AI Transformation Is a Problem of Governance: What Every Business Leader Must Understand,” March 2026; and The AI Insider, “Why AI Transformation Is a Problem of Governance,” April 2026.
  13. Ray Eitel-Porter, in conversation on Arkaro Insights, “AI Governance: How to Use AI Without It Going Wrong,” published 23 December 2025. https://arkaro.com/ray-eitel-porter-on-ai-goveranceai-governance-ray-eitel-porter/. See also Eitel-Porter, R. Governing the Machine: How to Navigate the Risks of AI and Unlock Its True Potential.