Every consulting firm recommends an "AI Center of Excellence" — a centralized AI team to drive enterprise adoption. In practice, most CoEs become bottlenecks that slow adoption rather than accelerate it. Here is why and what to do instead.
AI CoE = centralized team that owns AI strategy, tool selection, governance, and often AI implementations across the business.
The appeal is obvious: consistency, governance, economies of scale, vendor relationships managed centrally.
The reality: most CoEs become approval bottlenecks. Functions wait for the CoE to bless deployments. Adoption slows. Innovation moves outside the CoE through unsanctioned tools.
1. They are too far from the work. Effective AI deployment requires deep understanding of the specific function's workflows. CoE generalists know AI; they do not know how marketing or CS actually operates.
2. They become governance, not enablement. CoE staffing inevitably tilts toward policy, vendor management, and approvals — not building. Functions hit walls instead of getting help.
3. They cannot scale to organizational pace. When every function wants AI now, a centralized team cannot match demand. Shadow AI deployments proliferate.
Pattern: light central governance + embedded function leads. Small central team (1-3 people) sets policy, approves vendors, runs training. Embedded function-level AI leads (one per function) drive workflow deployment.
Function leads report to function heads. Not to the CoE. This keeps them close to the work and accountable for outcomes.
The central team is enabler, not gatekeeper. Their KPIs are function-level AI usage and outcomes, not approvals processed.
This pattern produces 3-5x faster deployment than centralized CoE models.