AI Centre of Excellence: Should You Build One — and What Should It Do?

As enterprise AI programmes mature from isolated experiments to strategic capability, a recurring organisational question emerges: who owns AI across the organisation? Who is accountable for AI strategy, for the quality of AI governance, and for the cross-business-unit decisions that affect the entire AI portfolio?

For many organisations, the answer is a Centre of Excellence. The AI CoE is a dedicated function that provides centralised expertise, standards, and oversight for AI investment and deployment across the enterprise. It is also one of the most misunderstood and misimplemented structures in enterprise AI.

What an AI CoE Is — and What It Is Not

An AI CoE is not an AI implementation team. It does not build AI systems for business units. When it does, it becomes a bottleneck — a centralised resource that every business unit competes for, and a dependency that slows AI adoption rather than enabling it.

An AI CoE is a standards and oversight function. Its value is in the things that every AI deployment needs but no individual business unit should build from scratch: governance frameworks, risk assessment approaches, operating model standards, talent development, and the institutional knowledge of what AI investments have worked, what has failed, and why.

The Three CoE Models

Centralised CoE — A single team owns AI governance, standards, and oversight for the entire organisation. Business units operate under direct CoE governance. This model produces high consistency and strong governance, but creates a central bottleneck that limits the speed and autonomy of AI adoption at scale. Appropriate for early-stage programmes or organisations with significant regulatory risk requiring consistent governance.

Federated CoE — AI capability is distributed across business units, each maintaining its own AI team. A thin central function sets standards and facilitates knowledge sharing. This model produces high speed and business-unit alignment but often at the cost of governance consistency — business units interpret standards differently and reinvent shared capabilities independently. Common in organisations where AI adoption outpaced governance.

Hybrid (Hub and Spoke) — A central hub owns governance standards, enterprise AI strategy, and shared services. Business-unit spokes own implementation and domain-specific AI development, operating within the standards the hub defines. This model captures most of the benefits of both centralised and federated approaches. Most mature enterprise AI programmes converge on this model as their AI portfolio grows.

The Six Functions the CoE Must Own

Regardless of model, the CoE should own six functions that genuinely require centralisation:

1. 1AI strategy alignment — ensuring AI investment is consistent with enterprise strategy and sequenced correctly
2. 2Governance standards — defining risk classification, documentation, and oversight requirements for all AI deployments
3. 3Risk oversight — maintaining a portfolio-level view of AI risk and escalating material issues to executive leadership
4. 4Operating model standards — defining how human–AI workflows are designed and what accountability structures are required
5. 5Talent and capability development — building AI literacy across the organisation and developing the specialist roles the programme needs
6. 6Knowledge management — capturing and distributing lessons from AI deployments across the organisation

Functions the CoE should not own: delivery of individual AI projects, management of technology vendors, operational support of deployed AI systems. These belong with business units and technology functions — the CoE's role is to set the standards they operate to, not to do the work for them.

When to Build the CoE

The timing of CoE establishment matters. Building too early — before there is meaningful AI activity to govern — produces a team with no mandate and no work. Building too late — when uncoordinated AI deployments have already created governance debt — means the CoE's first task is remediation rather than governance.

The right time is when the organisation has multiple AI deployments in production, when governance questions are emerging that individual business units cannot resolve alone, and when absence of coordination is visibly producing duplicated effort or inconsistent standards. For most large enterprise organisations in 2026, that moment has arrived or is imminent.