A charter before a hire, not after
Picture a CoE with three members, a dozen workshops behind it, and a sponsor about to ask why it needs five more people. That's the moment most leaders realize they skipped a step: how to set up an AI center of excellence so its mandate is clear before anyone counts heads.
The common move is to staff first and write the charter later, once the team has 'proven itself.' Consultant Arjun Jaggi's July 2026 guide on enterprise AI centers of excellence argues the opposite: write the charter before the first hire. We think that's right — a team without a written scope spends its first year defending its existence instead of doing the work.
This article is an editorial exercise: a structured way to draft that first charter, scoped to what a CoE owns, what it advises on, and what it leaves alone for now.
Sources: The Enterprise AI Center of Excellence: Build It Right or Don't Build It · Arjun Jaggi
What to own when you set up an AI center of excellence
Editorial exercise. Task: draft a one-page charter that lists every item the CoE will either own outright or advise on, with nothing left ambiguous. This is the concrete part of how to set up an AI center of excellence that can credibly ask for more resources later.
Oracle's own guidance for its Cloud EPM center of excellence, published in October 2026, lists maintaining the AI use-case inventory and defining permitted autonomy among the CoE lead's core duties. Oracle's related CoE team guidance assigns a process-owner role as steward of enterprise AI governance.
Oracle's checklist for creating a CoE lists defining its scope as an early step, before staffing. That guidance was written for finance and EPM teams, so adapt the categories, not the mandate.
A predefined employee survey covering usage, skills and perceived barriers, like AI Adoption Score's nine survey types, is one way to gather this input without building a bespoke questionnaire from scratch.
Sources: Establish the charter and responsibilities for your CoE (Center of Excellence) for EPM (Enterprise Performance Management), CoE (Center of Excellence) checklist for EPM (Enterprise Performance Management), also called EPM centre of excellence or EPM competency center, Build the team, ambassadors, key stakeholders, and community of practice for your CoE (Center of Excellence) for EPM (Enterprise Performance Management)
What most centre of excellence guidelines miss
You need three people at the table: the CoE sponsor, one line-of-business leader who will inherit the 'advise' items, and whoever handles AI governance today, even informally. Most centre of excellence guidelines spend pages on structure and staffing and almost none on what evidence to bring to the first draft.
Before drafting the owned list, gather evidence on usage, skills confidence and perceived barriers. Deloitte's 2026 State of AI in the Enterprise survey of 3,235 leaders across 24 countries, conducted in August and September 2025, found that leaders cite insufficient worker skills as the biggest barrier to integrating AI.
The same survey found education — not role or workflow redesign — was the top way organizations adjusted their talent strategy for AI. Both figures describe what leaders reported, not a measured skills improvement, so treat them as context for prioritizing, not proof an intervention worked.
CompTIA's Corporate AI Adoption study found that 67% of respondents view technology teams as the primary owners of AI execution, and only 43% believe their workforce is enthusiastic about adopting AI. That gap belongs in your charter's evidence pack.
| Evidence type | What it tells you | Where to get it |
|---|---|---|
| Usage patterns | Which tools and tasks people actually use AI for | A usage survey across teams |
| Skills confidence | Where employees rate their own AI skills low | A skills survey |
| Barrier perceptions | What employees say is stopping heavier use | A barriers survey |
| Workforce sentiment | Whether the group is enthusiastic or wary about AI | A culture or sentiment survey |
Sources: The State of AI in the Enterprise - 2026 AI report | Deloitte US, Corporate AI Adoption | CompTIA Research
What to leave out of the first charter
Keep the owned list narrow in year one: standards, the use-case inventory, and governance definitions. Leave broad delivery capacity and blanket deployment approval for later, once the charter has survived contact with real projects.
Oracle's EPM center of excellence page, citing Gartner, names poor communication as one of the risks a cloud CoE runs into. A charter that tries to own too much multiplies the number of teams it has to coordinate with, which is exactly where that risk shows up.
In our view, a CoE that asks for deployment approval rights before it has a working use-case inventory is asking for authority it hasn't earned yet. We'd rather see a narrow charter renewed every quarter than a broad one quietly ignored.
How you'll know the charter is ready to circulate
Run through this before you send the draft to your sponsor.
If you're staring at a blank page, start smaller than feels comfortable. Our bet is that the charters which survive a first year read more like a constraint than an ambition — three owned items done well beat twelve on paper and none in practice.
That's the real answer to how to set up an AI center of excellence that earns more headcount later: own less, own it clearly, and let the evidence pick what's next.
Frequently asked questions
Who should hold the pen on the CoE charter?
The CoE sponsor should hold the pen, but draft it with input from at least one line-of-business leader who will inherit the advised items. A charter written in isolation tends to list ambitions the business never agreed to carry out.
How long should the first charter be?
One page is the right target. If the owned list needs more than a page to explain, it's probably too broad for a first version, and you're better off narrowing it before you circulate it.
Should the charter name specific AI tools?
No, keep it to categories and standards rather than named tools. Tools change faster than governance structures, and naming them ties the charter to decisions that belong in the advised column, not the owned one.
When should we revisit the charter?
Revisit it at the mandate-review trigger you set inside the document, typically tied to the 12-month success definition, not on a fixed calendar. Revisiting earlier than that signals the first draft was scoped too loosely to begin with.
