Enterprise AI Operating Model

How to Build an AI Center of Excellence
(AI CoE)

Definition

An AI Center of Excellence (AI CoE) is the standing team that owns how an enterprise adopts AI: it sets the charter, ranks use cases, writes acceptable-use rules, builds workforce literacy, and readies the platform and data. Most run as an executive council over a working taskforce.

AI Center of ExcellenceAI CoEArtificial Intelligence Center of ExcellenceGenerative AI Center of ExcellenceHow to Build an AI CoEAI Operating Model
The AI CoE at a Glance
2
Levels: an executive council over one working taskforce
4
Work streams: intake, governance, literacy, platform readiness
90d
From charter signature to a governed use-case pipeline
70%
Share of AI outcomes decided by people and process (BCG 10-20-70)

What an AI Center of Excellence Does (and Does Not Do)

An AI CoE exists because enterprise AI fails on coordination, not on technology. Gartner projects at least 30% of generative AI projects are abandoned after proof of concept (Source: Gartner, July 2024), and McKinsey reports more than 80% of organizations see no tangible enterprise-EBIT impact from generative AI, with workflow redesign the biggest determinant of whether AI moves the bottom line (Source: McKinsey, The State of AI, March 2025). Neither is a model problem. Both follow from every business unit running its own pilot, against its own rules, on its own data, with nobody accountable for the portfolio.

The CoE is the standing answer: a small, permanent team with a written mandate to decide what the enterprise builds, under what rules, on what platform, with which skills. BCG's 10-20-70 finding — roughly 10% of AI outcomes from algorithms, 20% from technology and data, 70% from people and process — is why this is an operating-model job rather than an engineering one. Iternal covers the split in the 10-20-70 rule.

The four things a CoE owns

Owns the intake queue

Every AI idea enters one queue, is scored the same way, and comes back funded, deferred or rejected with an owner.

Owns the rules

Acceptable use, data classification, model approval, human-review thresholds and incident handling, written once and applied everywhere.

Owns capability

Role-based literacy, a champion in each business unit, and a pattern library so the second project costs less than the first.

Owns readiness

The approved platform, the data pipelines behind it, the security posture, and the evaluation harness every use case must pass.

The three things it should never own

It does not build every use case

Business units build. The CoE sets the pattern, holds the gate and stays out of delivery, or it becomes the bottleneck it exists to remove.

It is not a second IT department

No parallel infrastructure budget, no shadow roadmap. It works through the platform team, not around it.

It is not a monthly review meeting

A committee that reviews slides and approves nothing is the most common failure mode. Decision rights and a clock are what make it a CoE.

The AI CoE Charter: Mandate, Scope and Decision Rights

Nothing about a CoE works until an executive signs a charter. It runs to one or two pages and is the only artifact that converts a working group into a body with authority. Without it the group recommends but cannot decide, and the queue silently becomes a backlog. A workable charter answers seven questions in writing:

  1. Mandate. One sentence on what the CoE is accountable for — typically "every AI system touching company data or customers."
  2. Scope boundaries. What is explicitly out: embedded AI in purchased software, personal productivity tools under a stated threshold, research that never touches production data.
  3. Decision rights. What the CoE decides alone (platform, model approval, acceptable use), what it recommends to the council (funding, headcount, risk acceptance), and what stays with the business unit.
  4. The decision clock. A published turnaround — ten business days from intake to decision is common — so teams stop routing around the CoE.
  5. Escalation path. Who breaks a tie, and how fast.
  6. Funding model. Central budget for the CoE, business-unit budget for delivery. Mixing them turns a CoE into a build shop.
  7. Review date. A twelve-month review with defined success measures, so the charter renews on evidence rather than inertia.

Compliance obligations land here too. The NIST AI Risk Management Framework 1.0 organizes AI risk work into four functions — GOVERN, MAP, MEASURE and MANAGE — and GOVERN is, in practice, a description of what a CoE charter establishes. For organizations in scope of the EU AI Act, Article 4's AI literacy obligation has applied since 2 February 2025 and belongs to the literacy work stream (EU AI Act Article 4 literacy).

The Two-Level Structure: An Executive Council Over a Working Taskforce

The structure Iternal prescribes has exactly two levels, and the reason is arithmetic: every extra layer adds a review cycle, and review cycles turn a ten-day decision into a ten-week one. Level one holds accountability and money; level two does the work. There is no third body.

Level 1 — meets quarterly

The executive AI council

The board-facing body: the CEO or COO, the CIO or CTO, the CFO, the general counsel or chief risk officer, and the leaders sponsoring the largest use cases. It approves the charter, sets risk tolerance, funds the portfolio and accepts residual risk in writing — not individual models or tools. McKinsey found that CEO oversight of AI governance is the element most correlated with EBIT impact (Source: McKinsey, The State of AI, March 2025), which is why this level stays small, senior and genuinely engaged.

Level 2 — meets weekly

The working AI taskforce

One cross-functional team — not three competing committees — drawn from technology, legal and compliance, security, data and the business units. It runs the four work streams below and sends only funding, headcount and risk-acceptance items up. A single taskforce prevents the fragmentation where several groups each hold partial authority and none can act. Iternal publishes the same two-level model on the AI governance framework.

The one-page rule. If you cannot draw the CoE on one page — council, taskforce, four streams, named humans in each box — it will not survive its first reorganization. Names, not job families: "the data office" is not an owner.

The Four Work Streams an AI Center of Excellence Runs

The taskforce runs four work streams, all reporting into one body so that strategic, ethical, technical and adoption perspectives are reconciled in a single decision. The third column maps each stream to the governance-taskforce counterpart Iternal already publishes, so a company that stood up governance first renames rather than rebuilds.

Work stream What it owns Governance-taskforce counterpart Cadence Standing deliverables
Use-case intake & prioritization One queue, one scoring model, one funded/deferred/rejected decision with a named owner Strategic Prioritization Weekly triage, monthly portfolio review Scored pipeline, funding recommendations, kill list
Governance & acceptable use Acceptable-use policy, data classification, model approval, human-review thresholds, incident handling Ethics and Fairness Monthly review, on-demand for incidents Acceptable-use policy, approved-model register, incident log, audit evidence
Literacy & the champion network Role-based training, a named champion in every business unit, community of practice, adoption support Business Implementation Continuous; quarterly cohort reporting Role curricula, champion roster, completion reporting
Platform & data readiness Approved platform, data pipelines and retrieval quality, security posture, the evaluation harness Technical Standards Sprint cadence with the platform team Reference architecture, eval suite, data-readiness scorecard, patterns

The champion network is the stream most often skipped

Three of the four streams have obvious owners. The literacy stream usually does not, and it is the one that decides adoption. What works is a named AI champion in each business unit — a practitioner, not a manager — who holds office hours, collects the use cases the intake queue never hears about, and translates central rules into their team's workflow. Without them the queue only receives ideas from people already comfortable filing forms.

Only 37% of organizations have AI governance policies in place (Source: IBM, 2025), and a policy nobody in the business can explain is functionally the same as no policy. The champion network is how a written rule becomes an applied one. Iternal covers the curriculum side on the AI literacy framework and the adoption side on AI change management.

Staffing an AI CoE: 500 Employees Versus 10,000

The most common staffing mistake is copying a Fortune 100 org chart into a 500-person company. The work streams are identical at both scales; the headcount is not. At 500 employees the CoE is a set of responsibilities assigned to people who already have jobs, held together by one accountable leader. At 10,000 it is a funded team with its own reporting line.

Role ~500-employee enterprise ~10,000-employee enterprise Notes
CoE lead 0.5 FTE — a fractional Chief AI Officer or a director with a protected half-week 1.0 FTE — a Chief AI Officer or VP of AI reporting to the CIO or CEO Never shared with a delivery deadline
Governance and risk 0.25 FTE — existing legal or compliance counsel, one day a week 1–2 FTE — AI risk manager plus privacy and security liaisons Scales with regulatory exposure, not headcount
Platform and data 0.5 FTE — a platform or data engineer shared with infrastructure 3–6 FTE — a pod owning the reference architecture and eval harness Under-staffed most often; causes the most rework
Literacy and enablement 0.25 FTE — one enablement owner plus champions 2–3 FTE — training lead, curriculum designer, champion-program manager Recognized part-time roles, not headcount
Business-unit champions 3–5 named part-time champions 20–40 named part-time champions across functions and regions One champion per 250–400 employees works
Total central cost ~1.5 FTE, from existing staff plus fractional leadership ~8–12 FTE, funded centrally; delivery paid by the business units Delivery engineers sit in the business units either way

The lead role is where the scales diverge most. A 10,000-person enterprise can justify a full-time Chief AI Officer; a 500-person one usually cannot, and hiring one badly is worse than not hiring. Mid-market organizations run the lead as a fractional engagement, then transition to an internal owner. Iternal publishes the economics of that choice on the fractional Chief AI Officer page and the decision itself on do you need a Chief AI Officer.

Under-funding the platform stream is the expensive mistake. Of the thousands of companies marketing agentic AI, Gartner estimates only about 130 are real, with the rest re-labelling existing chatbots and RPA (Source: Gartner, June 2025). Disciplined platform and evaluation work keeps that noise out of the portfolio, and it cannot be done in a monthly meeting.

Centralized, Federated or Hub-and-Spoke: Choosing the Operating Mode

The centralized-versus-federated question is the one most CoE debates stall on, and it is posed wrongly. It is a stage, not an identity: programs start centralized because skills are scarce, move to hub-and-spoke as demand outruns the central team, and federate only when the operating companies underneath are separate.

Operating mode How work flows Best when Characteristic failure
Centralized The CoE builds the first wave itself; business units request work Fewer than five live use cases, scarce skills, or the first 6–12 months Becomes the queue everyone waits in; shadow AI grows around it
Hub-and-spoke The hub owns standards, platform, governance and patterns; spokes build on them Several business units building at once on a shared platform An over-sized hub that reviews everything; spokes that fork the platform
Federated Each unit runs its own AI capability; a thin council aligns policy Genuinely separate operating companies or divergent regulation Incompatible platforms, duplicated spend, no enterprise view of risk
The decision rule. Centralize anything where being wrong is expensive and consistency matters — data classification, model approval, security posture, acceptable use, the evaluation harness. Federate anything where being close to the work matters more — use-case discovery, workflow design, delivery sequencing, change management. Anything ambiguous defaults to the spoke.
The AI Strategy Blueprint book cover
The Operating Model

The AI Strategy Blueprint

The operating model behind this guide — an executive council over one working taskforce, four work streams, and a charter that grants real decision rights — is set out chapter by chapter in The AI Strategy Blueprint.

5.0 Rating
$24.95

The 90-Day AI CoE Launch Plan

Ninety days is enough to stand up a working CoE and not enough to perfect one, which is the point. The plan front-loads the two artifacts everything else depends on — the signed charter and an accurate inventory of the AI already running — then converts them into a scored pipeline and funded decisions.

Window Work Exit condition
Days 1–30
Charter and baseline
Sign the charter; name the council and taskforce leads; inventory the AI already in use, including unsanctioned tools; classify the data; publish the intake form and clock A signed charter, a named human in every box, a verified inventory
Days 31–60
Rules and pipeline
Publish the acceptable-use policy and approved-model register; score 15–20 use cases; agree the reference architecture; recruit champions A ranked pipeline with funding recommendations; a policy an employee can read in five minutes
Days 61–90
First decisions and first delivery
The council funds two or three use cases and kills the rest; the evaluation harness runs against the first build; the first metric pack is published Funded builds in delivery, a published metric pack, a repeatable gate

A CoE that publishes rules without knowing what it is regulating writes policy for a company that does not exist. Iternal covers discovery on shadow AI risks, the scoring model on AI use case identification, classification on AI data classification, and the day-60 artifact on the AI acceptable use policy page.

Where the CoE sits in the broader delivery sequence is set out on the AI implementation roadmap, whose fifth phase is where CoE setup normally lands.

What an AI CoE Reports

A CoE earns its charter renewal on six numbers. Activity metrics are inputs, and a council that only sees inputs eventually asks what the team is for. Gartner found 45% of high-AI-maturity organizations keep AI projects in production for three years or more, versus 20% of low-maturity organizations (Source: Gartner, June 2025); durability, not launch volume, is the maturity signal worth reporting.

Intake throughput

Ideas received, decided, and median days to a decision. Missing the published clock costs the CoE its legitimacy.

Portfolio state

Funded, in delivery, in production, deliberately killed. Kills matter as much as launches.

Outcome per use case

The metric each production system was funded to move, against its baseline.

Risk and compliance

Completed risk assessments, incidents raised and closed, audit coverage.

Capability

Training completion, active champions, teams that shipped without central help.

Reuse

Share of new builds on an approved pattern — the clearest proof the CoE pays for itself.

Iternal's method for putting a defensible number on each use case is on AI ROI quantification.

Why AI Centers of Excellence Stall

Four patterns account for most of the CoEs that quietly stop meeting. Each has a structural fix that belongs in the charter, not in a later course correction.

It became a discussion forum

A standing meeting that reviews and decides nothing. The fix is in the charter: decision rights, a published clock, escalation when it is missed.

It became the bottleneck

Every build waits on central review. Move to hub-and-spoke early, converting repeated reviews into patterns spokes self-serve.

It optimized for pilots, not production

A pipeline of demos with no production gate. The fix is one gate: evaluation thresholds, security sign-off, a named owner.

It had no sponsor with money

A CoE chartered by a committee rather than a budget-holding executive cannot fund anything. The council needs whoever controls the budget.

The through-line is authority: Gartner expects over 40% of agentic AI projects to be cancelled by the end of 2027 on unclear business value and inadequate risk controls (Source: Gartner, June 2025) — both decisions a properly chartered CoE makes before a build starts. The related pattern of pilots that never graduate is covered on AI pilot purgatory and the AI execution gap.

Where Microsoft, IBM, Oracle and Deloitte Fit Around an AI CoE

A CoE is assembled from published work, not invented from scratch. Take the platform guidance from the cloud provider you use, the governance benchmarks from the research houses, and keep the operating-model decisions internal.

Microsoft

Platform-side companion

The Cloud Adoption Framework publishes dedicated AI Center of Excellence guidance for the platform half of the job: landing zones, workload governance, cost management and the Azure AI service estate. It pairs cleanly with the operating-model half described here.

IBM

Governance research and tooling

IBM publishes on AI governance operating models and produces some of the most-cited enterprise survey data, including the finding that only 37% of organizations have governance policies in place.

Oracle

Data-estate perspective

Oracle frames the CoE from the data and application-estate side — the right lens when platform readiness means getting governed data into a trustworthy state.

Deloitte

Hub-and-spoke authority

Deloitte is a leading authority on the hub-and-spoke operating model, has committed more than US$3 billion to generative AI by 2030, and launched a global AI Infrastructure Center of Excellence in September 2025.

U.S. General Services Administration

Public-sector precedent

The GSA Centers of Excellence model gives public-sector organizations a chartered, funded precedent for a shared capability team.

Iternal's contribution sits in the gap those references leave open: the charter your executives will sign, the decision rights that make the taskforce a decision-making body, the champion network that carries policy into the business units, and a CoE lead on demand while the internal one is recruited — alongside the platform and integrator ecosystem, not in place of it.

For a broader survey of the operating models enterprises choose between, see the roundup of enterprise AI strategy frameworks.

Running the AI CoE With Iternal

Most enterprises can describe the CoE they want and stall on the same two things: nobody senior enough is free to lead it, and the first thirty days of artifacts are a full-time job nobody owns. Iternal covers both and steps back once the internal owner is in place.

The stand-up

AI Strategy Consulting

The engagement that produces the artifacts a CoE needs before its first meeting: the readiness baseline, the scored use-case portfolio, the governance model mapped to NIST AI RMF and the EU AI Act, and the platform decision.

  • Readiness assessment and data-classification baseline
  • Scored portfolio with funding recommendations
  • Reference architecture and evaluation harness definition
Explore consulting engagements
The training arm

Iternal AI Academy

The literacy work stream needs a curriculum, not a lunch-and-learn. AI Academy is the training arm: 912+ courses, role-based curricula for marketing, sales, finance, HR, legal and operations, and certification tracks aligned to the EU AI Act Article 4 literacy obligation.

  • Role-based curricula champions can assign directly
  • Certification tracks producing documented literacy evidence
  • 7-day free trial, so a pilot cohort can start before the charter is signed
Explore AI Academy

The sequence a CoE fits into is published on internal technologies for business transformation; the book-length version is The AI Strategy Blueprint.

FAQ

FAQ: AI Center of Excellence

An AI Center of Excellence (AI CoE) is the standing, chartered team that owns how an enterprise adopts AI: use-case intake and prioritization, governance and acceptable use, workforce literacy and the champion network, and platform and data readiness. It runs as two levels: an executive council that funds and accepts risk, over a working taskforce that decides.

In three thirty-day blocks. Days 1–30: sign the charter, name the council and taskforce, inventory the AI already in use and classify the data involved. Days 31–60: publish the acceptable-use policy and approved-model register, score 15–20 candidate use cases, agree the reference architecture, recruit champions. Days 61–90: fund two or three use cases, run the evaluation gate and publish the first metric pack.

Roughly 1.5 full-time equivalents at a 500-person company, assembled from existing staff plus a fractional leader, and about 8 to 12 at a 10,000-person one. The work streams are identical at both scales; only the headcount changes. Delivery engineers sit in the business units either way.

Treat it as a stage, not an identity. Start centralized while AI skills are scarce, move to hub-and-spoke as soon as more than about five use cases run in parallel, and federate only when the operating companies underneath are genuinely separate. Centralize decisions where being wrong is expensive — data classification, model approval, security, acceptable use — and federate the rest.

The governance taskforce is the level-two body inside the CoE. The CoE is the broader operating unit: the executive council, the taskforce, and the four work streams it runs, including the literacy and platform-readiness work pure governance bodies rarely own. Organizations already running a governance taskforce extend it into a CoE rather than creating a second structure.

A single accountable executive with budget access — a Chief AI Officer where the portfolio justifies one, otherwise a fractional CAIO or a director whose time is genuinely protected. McKinsey found executive oversight of AI governance is the element most correlated with EBIT impact, so seniority is not a formality. The role is never shared with a delivery deadline.

On six numbers, not on activity: intake throughput and median days to a decision, portfolio state including deliberate kills, the business outcome each production system was funded to move, risk and audit coverage, training completion and active champions, and reuse — the share of new builds on an approved pattern.

Start With the Charter, Not the Org Chart

An AI Center of Excellence is not a box on a slide. It is a signed charter, two levels of people, four work streams, a decision clock, and a metric pack the executive council reads every quarter. Companies that assemble those six things in ninety days stop having the same AI conversation twice a year and start deciding — funding two or three use cases properly and killing the rest without ceremony.

The order matters more than the ambition: charter first, then an accurate picture of the AI already running, then the rules, then the pipeline. Structure that arrives before authority produces a standing meeting nobody can cancel.

Fractional Chief AI Officer

Get a CoE lead on demand.

Iternal takes the level-two seat as your fractional Chief AI Officer: charter written with your executives, taskforce chaired, intake queue running and the first production gate defined — then handed to your internal owner. Engagements run $5,000 to $30,000 per month.

See the Fractional CAIO Engagement
Advisory Engagements

Or stand the CoE up as an engagement.

AI Strategy Consulting produces the artifacts your Center of Excellence needs before its first meeting: the readiness baseline, the scored portfolio, the governance model mapped to NIST AI RMF and the EU AI Act, and the platform decision.

Explore Consulting Engagements
John Byron Hanby IV
About the Author

John Byron Hanby IV

CEO & Founder, Iternal Technologies

John Byron Hanby IV is the founder and CEO of Iternal Technologies, a leading AI platform and consulting firm. He is the author of The AI Strategy Blueprint and The AI Partner Blueprint, the definitive playbooks for enterprise AI transformation and channel go-to-market. He advises Fortune 500 executives, federal agencies, and the world's largest systems integrators on AI strategy, governance, and deployment.