Pillar · AI Centre of Excellence
AI Centre of Excellence — Build, Operate, Transfer
The operating model that scales enterprise AI from one-off projects to portfolio-level capability. Sovereign platform, mixed-source delivery team, accelerator library, governance substrate, structured BOT lifecycle. Most CoEs we audit ship strategy decks; ours ship production AI from week 4.
BOT modelMixed teamSovereign12-24 monthsCentralised → Federated
Definition
AI Centre of Excellence, defined.
An AI Centre of Excellence is the operating model that scales enterprise AI from one-off projects to portfolio-level capability. It combines a platform (sovereign LLM + observability + governance), a delivery team (mixed senior engineering + AI specialists), an accelerator library (reusable components), evaluation discipline (regression harness + drift monitoring), and the governance substrate (Articles 9–15 evidence, model lifecycle management, audit trail). Without a CoE, AI deployment stays project-by-project and never compounds.
Build → Operate → Transfer
The BOT lifecycle for AI CoE engagements
Build (Months 0-6)
Stand up the CoE
Establish the platform, recruit/train the team, ship the first 2-3 production use cases that prove the operating model. MindMap leads with a 4-8 FTE delivery pod; customer team joins from week 4.
Operate (Months 6-24)
Run the portfolio
Deliver the use-case pipeline. Maintain the platform. Manage the evaluation harness and post-market monitoring. Scale the customer-side team to a sustainable steady-state. MindMap runs the show with customer-side leadership oversight.
Transfer (Months 18-36)
Hand over the keys
Structured transfer of platform, runbook, and certified team to the customer's permanent organisation. Knowledge-transfer artefacts. Operational handover. Optional retainer for continued advisory.
Team structure
Roles in a typical 8–12 person AI CoE
Mixed customer-side + MindMap-supplied roles. Customer-side capacity grows progressively; MindMap-supplied roles transfer out during months 12–24.
CoE Director / Head of AI
Customer-side leadership; reports into CIO / CDO / CDIO. Owns AI strategy alignment, prioritisation, executive comms.
Platform Engineering Lead
MindMap-supplied initially. Owns the sovereign LLM platform, observability, infrastructure. Transfers to customer-side senior engineer by month 18.
Solution Architects (2-4)
Split: customer-side for domain knowledge and integration to existing systems; MindMap-supplied for AI architecture patterns and accelerator integration.
ML / AI Engineers (3-6)
Mixed model. MindMap-supplied senior engineers + customer-supplied engineers learning patterns through pair-work.
Eval + Quality Engineer
Owns the evaluation harness, regression discipline, drift detection. The most under-staffed role in most CoEs we audit.
AI Governance Lead
Customer-side. Owns Articles 9-15 evidence, model lifecycle artefacts, FRIA (Article 27), interfacing with risk + compliance + legal.
Change Management Lead
Customer-side. Owns the business-side adoption — comms, training, dual-running with the existing process, exception-handling SOPs.
Four CoE patterns
Pick the structure that matches your maturity
Centralised CoE
Single team builds and operates all AI for the enterprise. Best for early-maturity organisations. Risk: bottleneck at scale.
Best for: Years 1-3 of enterprise AI
Hub-and-spoke CoE
Central hub builds platform + governance + reference architecture; line-of-business spokes own their use cases. Best for mid-maturity organisations.
Best for: Years 2-5
Federated CoE
Light-touch central function for standards + governance; LoBs own everything else. Best for late-maturity organisations with strong AI skills distributed.
Best for: Years 4+
Embedded CoE
AI specialists embedded into each LoB; no central platform team. Best when LoBs are large enough to support their own AI engineering capacity.
Best for: Specific large LoBs at scale
FAQ
AI CoE — the questions buyers ask
What is an AI Centre of Excellence?
An AI Centre of Excellence (CoE) is the operating model that scales enterprise AI from one-off projects to portfolio-level capability. It combines a platform (sovereign LLM + observability + governance), a delivery team (mixed senior engineering + AI specialists), accelerator library (reusable components), evaluation discipline (regression harness + drift monitoring), and the governance substrate (Articles 9-15 evidence, model lifecycle management, audit trail). Without a CoE, AI deployment stays project-by-project and never compounds.
What is the Build-Operate-Transfer (BOT) model for AI CoEs?
Build-Operate-Transfer is the engagement pattern MindMap uses for AI CoE work. We build the capability (people, process, platform), operate it long enough to demonstrate stability and train the customer's team (typically 12-24 months), then transfer the going concern to the customer at a pre-agreed date and price. The customer ends up with a fully-staffed, operating CoE without the multi-year ramp risk of building from scratch.
How is MindMap's CoE different from generalist consulting AI CoE work?
Three differences. (1) We ship production AI from week 4, not month 12 — the 117-accelerator library means the first use cases hit production while strategic planning is still happening. (2) Sovereign deployment is the default architecture, the model regulated industries actually accept. (3) The team that builds the CoE is the team that runs it during operate phase — there's no hand-off from a strategy team to a separate delivery team. Most CoEs we audit ship more decks than production AI.
What does an AI CoE engagement cost?
BOT model: typically €1.5-5M annually during Build + Operate phases (months 0-24), depending on team size and scope. Transfer phase: structured wind-down over 6-12 months, with a smaller retainer for advisory. Customer-side hiring + platform costs sit on top. Most customers run the full BOT lifecycle for €4-12M total over 24-36 months and end up with an operating CoE that would have cost more to build from scratch and would have taken twice as long.
What CoE structure works best for our maturity?
Four structural patterns. Centralised CoE works for years 1-3 of enterprise AI (early maturity, single team builds and operates everything). Hub-and-spoke works for years 2-5 (mid maturity, central platform + LoB delivery). Federated works for years 4+ (late maturity, distributed capability with light central coordination). Embedded works for specific large LoBs at scale (their own AI engineering capacity). Most customers cycle through these patterns as they mature.
How long does it take to set up an AI CoE?
Build phase: 6 months from contract to first production AI shipped. First steady-state CoE delivery rhythm is established by month 12. Operate phase typically runs 18-24 months. Transfer phase 6-12 months. Total BOT lifecycle: 24-36 months. Customers who skip the operate phase ('we'll just build it ourselves') consistently take 4-5 years to reach equivalent steady-state and bear materially higher staff-turnover risk during the build years.
What's the role mix in a typical AI CoE?
Indicative 8-12 person CoE: 1 CoE Director (customer), 1 Platform Engineering Lead (MindMap initially, transferred by month 18), 2-4 Solution Architects (mixed), 3-6 ML/AI Engineers (mixed), 1 Eval + Quality Engineer, 1 AI Governance Lead (customer), 1 Change Management Lead (customer). At scale (20+ FTE) the engineering side splits into platform team + use-case delivery teams and the governance side adds a model-risk specialist.
Ready to build your AI Centre of Excellence?
20-minute scoping call. We'll map the build-operate-transfer plan to your maturity and timeline before we leave the call.