Skip to main content
Why AI Centres of Excellence Often Stall
Decision Architecture

Why AI Centres of Excellence Often Stall

February 23, 2026

In the early stages of an organisation's engagement with artificial intelligence, the creation of a Centre of Excellence has become something of a ritual. It signals seriousness. It gathers expertise. It provides a visible anchor for ambition that might otherwise feel rather abstract. One could be forgiven for assuming the hard part is over.

The logic is, on the surface, impeccable. New technologies introduce uncertainty. A dedicated unit promises focus, coordination and clarity of ownership. It reassures boards and executive teams that AI is not being left to the adventurous instincts of individual departments. It suggests momentum, which is always preferable to suggesting the alternative.

The Ritual of Readiness

And yet, in a surprising number of organisations, something curious happens over time. The Centre of Excellence produces pilots. It generates proofs of concept. It circulates presentations outlining potential use cases with considerable fluency. The internal narrative remains positive, occasionally enthusiastic. But the operational footprint of AI across the wider organisation grows slowly, if it grows at all.

The temptation, at this point, is to look for familiar culprits. Insufficient budgets, perhaps. Talent gaps. Or a CEO who publicly champions innovation but privately prefers predictability. These explanations are not wrong, exactly. They are simply incomplete.

The difficulty rarely lies in competence. The individuals involved are often highly capable, frequently more so than the organisation quite knows what to do with. Nor does it typically lie in intent. Senior leadership continues to speak of transformation with what appears to be genuine conviction. Budgets, at least initially, are allocated.

The friction tends to emerge elsewhere, in places less amenable to a well-constructed slide deck.

The Perimeter Problem

A Centre of Excellence, by design, sits adjacent to the core business. It concentrates expertise in one place, which sounds efficient until one considers the implication. What it does not automatically alter is how decisions are made outside its perimeter. Business units retain their own priorities, their own calendars, their own quietly entrenched definitions of what constitutes a good idea. Risk committees continue to operate according to protocols that were perfectly serviceable before anyone mentioned machine learning at a strategy offsite. Performance metrics remain calibrated to models of efficiency and predictability that predate the current conversation by a comfortable margin.

In such an environment, experimentation becomes episodic. A promising prototype may be celebrated within the Centre, perhaps even demonstrated to a visiting board member, yet struggle to find a natural path into production. Integration requires cooperation from functions whose incentive structures were not designed with AI in mind, and whose leaders may reasonably wonder whether cooperation will be rewarded or merely create additional complexity for which they will be held responsible.

Accountability, meanwhile, acquires a new and rather unwelcome ambiguity. When algorithmic outputs begin to influence operational choices, the question of who ultimately signs off becomes less straightforward than anyone had anticipated. The traditional answer, that the person with the budget decides, starts to feel inadequate when the decision in question was substantially shaped by a system that no one in the room fully built.

Coordination Is Not Architecture

None of this suggests that Centres of Excellence are misguided. They often serve a genuinely valuable coordinating role. They can accelerate learning, prevent the kind of fragmentation that leads to seven departments independently licensing the same tool, and provide a home for expertise that might otherwise drift towards organisations willing to use it. But coordination is not the same as architectural change, however much one might wish it were.

Artificial intelligence alters the velocity and texture of decision-making. It expands the informational substrate on which choices are based. It introduces probabilistic reasoning into domains long accustomed to deterministic accountability, where outcomes were expected to follow inputs with reassuring linearity. When these shifts encounter organisational structures built for slower cycles and clearer lines of responsibility, tension is not merely predictable. It is inevitable.

A Centre of Excellence may illuminate that tension. It is not, however, equipped to resolve it on its own. To expect otherwise would be rather like hiring an excellent navigator and then declining to adjust the ship's course.

From Mandate to Environment

What appears, at first glance, as a problem of execution frequently turns out to be a question of structure. The deeper issue is not whether the organisation possesses an AI unit, but whether its underlying decision architecture, the system governing how information flows, how authority is distributed, how risk is assessed and how accountability is assigned, has evolved sufficiently to absorb what that unit produces.

Where authority remains tightly held, experimentation depends on exception rather than invitation. Where incentives reward stability above learning, adoption will be cautious, and rightly so from the perspective of those whose careers are at stake. Where accountability frameworks have not been re-examined in light of algorithmic co-creation, scaling becomes politically delicate in ways that no amount of executive sponsorship can entirely smooth over.

Under such conditions, the Centre of Excellence risks becoming a symbolic centre rather than an operational one, a place where AI is discussed with sophistication and practised with enthusiasm, but from which it struggles to escape into the broader organism.

The challenge, then, is less about refining the mandate of the Centre and more about examining the wider environment in which it operates. If artificial intelligence is to become embedded rather than adjacent, the structures governing information, authority, incentives and accountability must gradually, and in some cases rather less gradually, adjust.

Centres of Excellence can catalyse attention, but they rarely, by themselves, reconfigure architecture. The distinction matters more than it might appear, particularly for those currently wondering why their Centre is producing excellent work that nobody quite seems to use.

Sources

  1. Pisano, G.P., "You Need an Innovation Strategy," Harvard Business Review, June 2015. On why dedicated innovation units often fail to produce systemic change.
  2. Christensen, C.M. and Overdorf, M., "Meeting the Challenge of Disruptive Change," Harvard Business Review, March-April 2000. On organisational capabilities and architectural constraints.
  3. Tushman, M.L. and O'Reilly, C.A., "Ambidextrous Organizations," Harvard Business Review, April 2004. On the structural tension between exploitation and exploration.
Decision Architecture - Book Cover

Decision Architecture

How Organisations Decide in an Age of Technological Acceleration

The essays collected here served as the conceptual foundation for a book of the same name, recently published and now available in physical and digital formats. Those seeking a more sustained examination of the structural conditions under which decision remains possible may find the book a useful companion to these shorter explorations.

View book details

More from Decision Architecture

When Decisions Refuse to End
Decision Architecture

When Decisions Refuse to End

March 20, 2026

Many organisations believe they struggle because they make poor decisions. In reality, a more subtle pattern often emerges. Decisions simply never reach completion.

The Pace Mismatch
Decision Architecture

The Pace Mismatch

March 10, 2026

Modern organisations operate within informational environments that move faster than the organisations themselves. This essay examines the structural tension between systems that generate knowledge in real time and institutions that coordinate action through deliberation.

When Knowledge Outpaces Decision
Decision Architecture

When Knowledge Outpaces Decision

March 9, 2026

Organisations have never possessed more information or analytical capability. Yet many leaders experience growing complexity instead of clarity. This essay explores the paradox of knowledge advancing faster than the structures through which organisations turn it into decisions.

When Signals Multiply
Decision Architecture

When Signals Multiply

March 5, 2026

AI does not eliminate uncertainty. It transforms its texture. When machine intelligence expands the field of detectable patterns, interpretation becomes the scarce resource.

The Quiet Redistribution of Judgement
Decision Architecture

The Quiet Redistribution of Judgement

March 3, 2026

AI does not merely introduce new tools. It reshapes the conditions under which judgement is formed, exercised and recognised. An essay on authority, accountability and the redistribution of cognitive labour.

The Incentive Paradox in AI Transformation
Decision Architecture

The Incentive Paradox in AI Transformation

February 26, 2026

Organisations declare AI ambition while rewarding the avoidance of visible failure. The resulting paradox produces activity without acceleration. An essay on incentives, architecture, and what actually becomes real.

AI and the Question of Decision Architecture
Decision Architecture

AI and the Question of Decision Architecture

February 21, 2026

AI adoption is less a technology problem and more a question of decision architecture. An essay on the structural layer that determines whether capability translates into durable impact.

Back to Decision Architecture