What is an AI champion in an organisation?

An employee in a normal role, in a business team, who receives part of their working time and an explicit mandate to help colleagues use AI tools in their day-to-day work. It isn't a technical role and it isn't a new position on the org chart. The difference from an internal trainer is that a champion works on the real workflows of their own team, in its context, and is present after the training, which is exactly the period in which people abandon new tools.

How do you choose AI champions in an organisation?

Three criteria matter more than enthusiasm. First: they should be professionally credible in their team, because adoption travels on trust, not on information. Second: they should have patience with those who don't understand, which excludes some of the most advanced users. Third: they should be spread across departments and seniority levels, not concentrated in the technical area. A useful exclusion criterion is pure volunteering: if the list is made only of those who raise their hand, you will get enthusiasts without influence.

Why do AI adoption programmes fail?

Because they treat a problem of people and processes as a technology problem. BCG research shows a split of roughly 10% technology, 20% data and 70% people and processes, while budgets go in the opposite order. Concretely, adoption gets stuck in three places: at direction, when the leadership team has no shared position; at competence, when people saw a generic demo and did not rewrite their own workflow; and at fear, when employees suspect automation will cut their job and smile through the training without changing anything afterwards.

13 October 2026 · the Facilitare.ro team

AI champions: the 10% who change the rest of the organisation

You bought the licences, ran the training, and three months later 10% of people use them. The next move, in most organisations, is another training. The move that works is a different one: you choose 10% of the people and make them responsible for the rest.

They are called AI champions, ambassadors or an internal network; the name matters less. What matters is who is in it, what they receive and what is measured. Three things almost all organisations get wrong, in the same way.

Three colleagues at a desk in an open-plan space, one of them explaining something to the other two and pointing at a laptop. — Adoption travels on professional trust, not on information.

Why another training doesn't work

The figure everybody cites is that 70% of the effort of an AI transformation lies in people and processes, 20% in data and only 10% in technology. Budgets are distributed exactly the other way round. The practical consequence is that the organisation buys technical capacity and is left without capacity for change.

A generic training loses to routine for three banal reasons. People see examples that don't resemble their work. They aren't allowed to experiment on real data. And, most important, when they get stuck the next day, they have nobody to ask without looking as though they couldn't cope.

A network of champions solves precisely the third reason, which is also the most ignored.

How to choose, and why enthusiasts are the wrong choice

The first list any organisation makes is the list of those who raise their hand. It is the worst possible list, for two reasons.

The first: the most advanced users usually have the least patience with those who don't understand. A person who has already integrated three tools into their workflow no longer remembers what wasn't obvious six months ago, and the colleague who gets stuck at the first step feels foolish and doesn't ask a second time.

The second: enthusiasm doesn't spread without credibility. Adoption travels on professional trust, not on information. If the person recommending the tool isn't someone whose work is respected in the team, the recommendation remains pleasant noise.

The criteria that matter: professional credibility in their own team, demonstrated patience with slower people, and coverage across departments and seniority levels. Add an uncomfortable criterion: at least one of the champions should have been sceptical at first. A converted sceptic convinces more than ten enthusiasts.

What they must receive, otherwise the role dies in three months

The most frequent way to miss the programme is to give the title with nothing behind it. A champion who has the same targets as before, plus a new responsibility, will drop the new responsibility. Character has nothing to do with it. It is simple arithmetic.

Time, written down somewhere. A few hours a month, acknowledged by the direct manager. Without the direct manager's agreement, the programme is a fiction from the first busy week.
An explicit mandate. What they can decide alone, what they must escalate, what data must not leave the organisation. Clear limits don't hold back experimentation, they make it possible.
A community. One hour a month together, moderated, in which real problems are brought, not presentations. A champion alone in their department loses heart within six weeks.
Visibility. Someone from leadership who says publicly that the role matters and who asks about it in meetings. Without that, it is volunteering with extra work.

The part about fear, which no tool solves

In any room where AI is discussed there is a question nobody asks: how many of us will still be here in two years. As long as the question stays unasked, every discussion about efficiency is translated in people's heads into a discussion about staff cuts.

The effect shows exactly where you don't expect it: people attend the training, give positive feedback, then change nothing. They aren't sabotaging out of ill will. It is just that nobody is in a hurry to automate the part of their work that makes them visible.

An organisation that has no honest answer to the question about jobs will buy licences forever.

The answer doesn't have to be reassuring, it has to be honest and consistent. If you don't know yet, say that you don't know yet, say what changes and what doesn't in the next six months, and repeat it. Adoption grows with predictability, not with optimism.

What to measure at 30, 60 and 90 days

The number of active licences is the worst measure available, because it can rise while nothing about the work changes. The measures that tell you something are of a different kind.

How many workflows were actually rewritten, with steps eliminated or changed. One documented rewritten workflow is worth more than 200 active accounts.
How many questions reach the champions each month. A sudden drop doesn't mean everyone has understood, it usually means people have given up.
How many use cases were deliberately stopped, because they weren't worth it. An organisation that has stopped nothing has evaluated nothing.

Where it actually starts

The champion network is the third move, not the first. Before it comes diagnosis, meaning where adoption is stuck: at direction, at competence or at fear. Then alignment at leadership level, because without a shared position there, any internal programme becomes someone in HR's personal initiative.

What the three interventions look like and who facilitates them is on the AI adoption page. And if your question comes earlier than that, namely whether you need training, consulting or facilitation, we wrote about that separately.

The AI Champions programme

Selection and training of the internal network, a moderated community of practice, facilitated redesign of key workflows and an adoption pulse at 30, 60 and 90 days. Accompaniment for 3 to 6 months.

See the programme