How to Create an AI Transformation Office

Chief of Staff (Government-Linked Enterprise)

Our chairman has asked us to establish an AI transformation office. I have seen transformation offices become reporting factories. What should this one actually do, and how big should it be?

AI Leadership Agent — Strategic Advisory

Small, senior and unusually specific about its remit. Five to eight people is enough for most enterprises; beyond that the office starts generating work to justify itself.

Give it four jobs and explicitly deny it a fifth:

1. Prioritisation. Maintain one enterprise-wide list of AI work with visible criteria, so funding stops being decided by proximity to the chairman.
2. Standards and enablement. Common tooling, evaluation practice, data access patterns, reusable components — the things that make the tenth project cheaper than the first.
3. Governance coordination. Not owning risk, but making sure every system has an owner, a tier and monitoring, and that the committee has what it needs.
4. Adoption measurement. Track whether delivered systems are used and whether the decisions they support changed. This is the number the chairman should see monthly.

The denied fifth job is delivery. The moment the transformation office builds things itself, it stops being able to hold anyone else accountable for building them.

Who would the office report to in your structure?

Chief of Staff (Government-Linked Enterprise)

Probably to me, which means effectively to the chief executive. What roles do I need in the first six months, and where do I find those people here?

AI Leadership Agent — Strategic Advisory

Reporting to the chief executive is the right call; an AI office inside IT will be treated as a technology programme and will never move an operating model.

1. A lead with delivery scars — someone who has taken an AI system into production somewhere and can tell when a plan is fiction.
2. A data and platform architect who understands your estate, ideally hired from inside.
3. A risk and governance specialist, seconded from risk or legal, not recruited externally, because relationships matter more than credentials here.
4. Two business-side change leads drawn from operating units, rotating on twelve-month terms so knowledge flows back out.
5. An analyst who owns the measurement and the reporting, so the senior people are not writing decks.

In this market, the secondment route usually beats external hiring for three of those five. Local operating knowledge is scarcer than AI knowledge, and harder to buy.

Staff the office with people the business already trusts. Authority in a transformation office is borrowed from relationships, not from the org chart.

What would success look like to your chairman after twelve months?

Chief of Staff (Government-Linked Enterprise)

Visible progress, honestly. Which worries me, because visible is easy to fake. How do I set targets that are real without setting us up to fail in year one?

AI Leadership Agent — Strategic Advisory

Set targets on the two things that are hard to fake and early to appear: systems in production with named business owners, and measured change in the work those systems touch. Avoid counting pilots, trained employees or use cases identified — all three can be inflated indefinitely.

It also helps enormously to establish the baseline before you start. A shared, queryable model of how the organisation currently operates and where readiness is weak gives you an honest starting line, so year-one progress is measured against reality rather than against a memory of it.

Set the review date for the office itself at the outset. A transformation office that is still the same size in year three has become the thing it was meant to fix.

Build it to be handed back to the business, then plan the handover from the first week.