Who Owns AI?

AI ownership often looks clear in meetings, then breaks down when decisions move into real workflows. This episode examines why fragmented authority turns promising pilots into slow, duplicated, and politically complex AI programmes across the enterprise.

It explores how decision rights shape AI scale.

TLDR / At a Glance

• Fragmented AI accountability
 • Decision rights over job titles
 • Centralised, federated, and hybrid models
 • Governance without slow consensus
 • Executive sponsorship and escalation routes
 • Ownership maps for enterprise scale

The key lesson is that AI scales when authority, accountability, and value ownership are explicit.

Transcript

Who Owns AI In Practice

SPEAKER_00 0:00

Who owns AI? AI often becomes everyone's priority and no one's responsibility. That is one reason progress becomes slow, fragmented, political, and hard to scale. The issue is not simply that AI crosses many functions. It is that most organizations have not designed clear decision rights for something that touches strategy, platforms, risk, legal, operations, data, and business execution. The first two articles explained why AI strategies fail before they scale and introduce the human AI operating system. This article turns to the decisions layer and asks the question every organization eventually faces. Who actually owns AI when it moves into real work? Pilot AI can survive on enthusiasm, local sponsorship, and improvised governance. Scale needs visible ownership, defined authority, escalation routes, and accountability that holds under pressure. Why AI ownership becomes fragmented? AI creates a structural ownership problem because the decisions that matter rarely reside in a single place. Technology may own the platform. Data may own access and quality. Legal and risk may own approval thresholds. Operations may own workflow change. HR may own workforce policy. Business units may own the use case and expected value. Finance may control the budget. The problem is that meaningful AI decisions cut across all of these boundaries. A use case is not just a technology choice, it is also a data decision, a risk decision, a workflow decision, a people decision, and a value decision. When ownership is fragmented, AI can move forward in activity but stall in accountability. That is where many organizations lose control. Not because no one is involved, but because too many functions are involved without a clear decision architecture. That is why organizations so often end up with shared responsibility without clear authority. McKinsey's 2025 global survey found that respondents reported two leaders in charge of AI governance on average. Only 28% said the CEO oversaw AI governance, and only 17% said the board did. In its 2024 survey, only 18% said their organization had an enterprise-wide council or board with the authority to make decisions involving responsible AI governance. Similarly, in UK financial services, the Bank of England and FCA found that while 84% of firms reported an accountable person for the AI framework, accountability for AI use cases was still commonly spread across three or more people or bodies. The result is predictable. AI becomes strategic in rhetoric but operationally diffused in reality. It is discussed by many people but not truly governed by a decision system that can move quickly when trade-offs appear.

The Hidden Cost Of Fragmented Ownership

SPEAKER_00 3:04

The cost of unclear decision rights. When ownership is unclear, the obvious costs are delay and duplication. The less visible costs are political complexity, weak accountability, and slow movement from pilot to production. Teams may feel they are moving quickly because experimentation is active. In reality, progress often stalls when shared decisions must be made. Which platform to use, which use cases to prioritize, which legal and risk thresholds apply, who funds the work, how workflows must change, and who owns the benefits. The UK's National Audit Office provides one of the clearest public examples of this problem. In its 2024 review of AI and Government, it warned that value for money was at risk if government did not establish which department had overall ownership and accountability for delivering the AI strategy and define roles and responsibilities of those contributing to it. That is unusually explicit. It does not describe a lack of ideas or technology. It describes a strategy at risk because ownership at the center was unclear. Sector evidence points in the same direction. McKinsey's work in banking found that highly centralized generative AI operating models were associated with materially better movement into production than fully decentralized ones. In life sciences, McKinsey also warns that highly decentralized models can allow early speed but later lead to problems with quality, cost, sustainability, and silos. The pattern is consistent. Decentralized AI can help organizations generate ideas quickly, but without shared decision rights, it often slows the move to scale.

What The Decisions Layer Controls

SPEAKER_00 4:51

What the decisions layer actually decides. The decisions layer is not about job titles. It is about who gets to decide what, under what authority, with what rights, and how disputes are escalated. In practical terms, it covers ownership, decision rights, authority, review, challenge, escalation, and accountability. That includes questions such as these. Who sets enterprise AI posture? Who chooses the platform and vendor approach? Who approves use cases? Who decides when a workflow is ready for production? Who owns the business case? Who can stop a deployment? Who reviews AI incidents? Who arbitrates when business speed and risk concerns collide? These questions are not administrative details. They are the conditions that determine whether AI remains local activity or becomes enterprise execution. Pilot AI can often survive with fuzzy answers because the stakes are contained and the boundaries are narrow. Scaled AI cannot. Once AI enters client service, compliance, risk management, workflow automation, or operations across functions, unclear decision rights stop being a governance inconvenience. They become a structural blocker.

Why Ownership Beats Tool Choice

SPEAKER_00 6:09

Why ownership matters more than tool choice? Organizations often ask which tool to use before they answer who gets to decide. That sequence is backwards. Tool choice matters, but ownership matters more because every major AI question is a decision problem before it becomes a technology problem. This is visible in the evidence on Senior Oversight. McKinse found that CEO oversight of AI governance was among the organizational elements most correlated with higher self-reported bottom line impact from generative AI. Stanford's 2026 Enterprise AI Playbook adds a practical explanation. Effective sponsors, clear blockers, connect business and technical teams, and tie AI adoption to real operating goals. Passive sponsorship is not enough. AI needs leaders with the authority to make decisions across silos. The same point appears in where resistance comes from. Stamford found that legal, HR, risk, and compliance were the most frequent source of resistance in its case, sample, ahead of end users. That does not mean these functions are the problem. It means they become de facto gatekeepers when participation models are unclear. If they are excluded early, they block late. If they are included only as generic approvers, they become bottlenecks. Clear decision architecture is what turns them from veto points into defined co-owners of specific decisions. Without that architecture, AI does not fail because the wrong people are involved. It fails because the right people are involved too late, with unclear authority, unclear ownership, and no agreed route for resolving conflict.

Build A Hybrid Decision Architecture

SPEAKER_00 7:56

What better decision architecture looks like? The evidence does not support a simplistic answer such as centralize everything or give AI to the business. It points to a more practical rule. Centralize the decisions that should be made once, federate the decisions that need to sit close to the work, and define escalation routes before conflict appears. A centralized model tends to work best early in maturity, especially where talent is scarce, platform choices are still fluid, or regulation exposure is high. A federated model works better later, when businesses have the capability to own local workflow, redesign, adoption, and value capture. For most large organizations, the practical destination is hybrid. A central enterprise layer sets strategy, standards, risk posture, architecture, and portfolio priorities. Business units then own domain use cases, adoption, workflow change, and business outcomes. Lloyd's Banking Group is a useful example of what that looks like in practice. It describes AI as a horizontal capability spanning business units and functions, supported by a dedicated AI center of excellence and a new AI operating model. It also describes a Gen AI control tower, a forum used to prioritize use cases, allocate resources, and ensure alignment with broader strategy. The important point is not the name of the mechanism, it is that decision architecture is explicit. The center owns the shared decisions while the business remains connected to value and execution. Alibaba provides a different ownership signal. Reuters reported that the company launched Wukong, an enterprise AI platform designed to coordinate multiple AI agents across tasks such as document editing, spreadsheet updates, meeting transcription and research, and automate multi-step business work. Reuters also reported that Wukong sits under Alibaba's newly established Alibaba token hub, signaling a company-wide push into enterprise AI agents. That kind of reorganization raises the right question for leaders. When AI becomes a horizontal enterprise capability, which decisions need to move to the center so that platforms, security, permissions, and coordination do not fragment? Standard Chartered makes the same issue visible from the user side. Its partnership with Alibaba is designed to deploy AI into client service, sales intelligence, risk management, compliance, and workforce upskilling. The moment AI crosses that many functions, ownership cannot remain vague. Someone has to define who owns the platform, who owns the workflow, who owns the control thresholds, and who resolves disputes when those interests collide.

A Leadership Checklist For Scale

SPEAKER_00 10:49

How leaders should think about AI ownership now. The question is not who owns all of AI. That question is too broad to be useful. The better question is do we have a visible decision architecture that can hold under real operating conditions? That is a much stronger leadership test. For boards and CEOs, that means demanding a clear ownership map. Who sets enterprise posture? Who approves risk thresholds? Who owns use case prioritization? Who owns value? Who has challenge rights? Who can pause or stop a deployment? If those answers exist only in meetings, informal relationships or individual judgment, the organization is not ready to scale AI with confidence. For transformation leaders, it means treating the decisions layer as the hinge between pilot and scale. Pilots tolerate ambiguity because their boundaries are narrow and their risks are usually contained. Scale does not. When decision rights are unclear, every other layer weakens with them, including governance, capability, workflow redesign, and value capture. For technology, data, legal, risk, and operations leaders, it means being precise about what is central, what is local, and what gets escalated. The real failure mode is not shared input. Shared input is necessary, it is shared responsibility without clear authority.

Closing Takeaways And What’s Next

SPEAKER_00 12:15

Conclusion. AI scale does not happen when tools improve. It happens when decision rights are clear, ownership is visible, and accountability is real. Pilot AI can survive with local sponsorship, informal approvals, and improvised governance. Scaled AI cannot. The next article moves from ownership to work itself. We will look at why AI does not scale through tools alone, why faster tasks are not the same as better workflows, and what leaders need to redesign if AI is going to create real organizational value. This concludes the article. You can also read this article on my LinkedIn page where I share regular insights on AI, strategy, and emerging technologies.