Most AI champions programs do not fail. They stop, and from a leadership meeting the two look identical.
The first year usually goes well. A few staff who were already curious get a title, a little protected time and some encouragement. They run drop-in sessions. Colleagues come. Practice spreads outward from them, person to person, the way good practice has always spread in schools and organizations. Then, somewhere in the second year, the sessions thin out. The same four faces keep appearing. And the champions start sounding less like enthusiasts and more like people asking for something they cannot give themselves.
The reason is structural rather than personal. A champions program is a mechanism for moving practice sideways between colleagues. It works well right up to the moment the next thing that needs to happen is not a demonstration but a decision. At that point the program has met what I would call the Champion's Ceiling, and no amount of enthusiasm below it produces what is needed above it. The response is not better champions or a bigger network. It is a leadership team that names the ceiling and does the four things a champion structurally cannot do.
What a champions program actually does, and what it cannot
A champions program transmits practice, not authority, and almost every stall traces back to that one distinction.
What a champion offers a colleague is a demonstration: here is a thing I do, here is how I do it, here is what it saved me. That is valuable, and it is the fastest way to move a technique between two people who trust each other. It is also the whole of what the role contains. A champion cannot tell a colleague that the technique is allowed on real student data. They cannot take the old task off that colleague's plate to make room for the new one. They cannot make the practice expected of the reluctant as well as the willing. And when the practice produces something wrong in front of a parent, an inspector or a regulator, they are not the person who answers for it.
So the program runs until it reaches the first colleague whose honest reply is "I would, but am I allowed?" Everyone below that line has already been converted. Everyone above it is waiting for something a peer cannot issue.
The evidence: individual gains are easy, organizational gains are not
The research now separates these two things cleanly, and the gap between them is the ceiling measured at national scale.
McKinsey's The state of AI in 2026: On the Road to ROI, published on 25 August 2026, reports that "eight in ten respondents say AI has improved their own productivity," while "about four in ten respondents (37 percent) report that AI has contributed positively to their organizations' EBIT, essentially unchanged from 2025." Individual benefit is now close to universal. Organizational benefit has not moved in a year.
What separates the organizations getting both is instructive. McKinsey finds that "nearly three-quarters of high performers report fundamentally redesigning workflows because of their AI use," against "just one-quarter of other respondents," and that high performers "are twice as likely as others to say that their senior leaders demonstrate commitment to AI initiatives." Redesigning a workflow and demonstrating senior commitment are not champion activities. They are the two items on the list that require a title.
Education shows the same shape in its own terms. RAND's AI Use in Schools Is Quickly Increasing but Guidance Lags Behind, published on 30 September 2025, found that "in 2025, 54 percent of students and 53 percent of English language arts, math, and science teachers indicated that they used AI for school," while only 45 percent of principals reported having school or district policies or guidance on the use of AI. Just over half the teachers are using it. Fewer than half the schools have said anything about it. The practice arrived through people. The permission did not arrive at all.
Ofsted watched this happen and wrote it down
The clearest account of the champion mechanism comes from the regulator, and it describes both the strength of the model and its limit.
Ofsted's "The biggest risk is doing nothing": insights from early adopters of artificial intelligence in schools and further education colleges, published on 27 June 2025 and drawn from 21 early adopter providers, found that AI use in these settings typically began with individual enthusiasts rather than a leadership decision, and it uses the term "AI champions" for them directly. One college champion describes a thirty-minute meeting with senior leaders that ran to ninety because the demonstration was compelling. That is the model working exactly as intended.
Two other findings in the same report show the ceiling. Leaders described short-term, pragmatic thinking rather than strategy, with one principal saying that "anybody who's telling you they've got a strategy is lying to you because the truth of the matter is AI is moving so quickly." And a digital leader, asked what the role actually requires, answered: "What you really need is someone with leadership responsibility. Someone who really has knowledge about what's going on in AI in education," plus "someone who can speak 'human' as well, rather than 'techie'." The people closest to the work were not asking for more enthusiasm. They were asking for the enthusiasm to be attached to authority.
I would push back on the principal. The pace of change is a reason not to write a five-year technology plan. It is not a reason to avoid deciding what AI is for in your organization, which is a decision about your purpose rather than about the tools, and which barely moves when the models do.
The Champion's Ceiling
The Champion's Ceiling is the point at which an enthusiastic individual can take an organization no further, because the next thing that needs to happen is not a demonstration but a decision. Every champions program meets it. Above the ceiling sit four things no champion can do for you: grant permission, take work away, change what is expected, and answer for a mistake. A leadership team that names the ceiling in advance can plan for it. One that does not will conclude it picked the wrong champions.
I offer this as a way of reading what is happening in your own organization rather than as a tested model. It is a description of a limit, not a maturity scale.
Grant permission
A champion can say what they personally do. Only a leader can say what everyone may do.
The question that ends a champion's usefulness is almost never "how does it work?" It is "am I allowed to put this in there?" A colleague who has watched a demonstration and wants to try it on a real report, a real reference, a real set of pupil data is asking a governance question, and the honest answer from a peer is "I do, but I am not the person who decides." Diagnostic: could a member of your staff read something you have published and know what they may put into an AI tool on Monday, without asking anyone?
Take work away
A champion can add a faster way of doing something. Only a leader can remove the slower way it was supposed to replace.
This is where most champion-led gains go. The new method is demonstrated, adopted, and then performed alongside the requirement it was meant to retire, because the template, the policy or the review process that mandates the old way belongs to somebody else. Adoption in those conditions is a second job, and the enthusiasts absorb it for a while because they enjoy it. Diagnostic: name one thing that has stopped since your champions started, with the date it stopped and the person who agreed it would.
Change what is expected
A champion changes what is possible for the willing. Only a leader changes what is expected of everyone.
A voluntary network is, by design, a self-selecting one. It recruits people who were going to do this anyway and it reaches people who like the people who were going to do this anyway. That is a real gain and it has a natural edge. Crossing the edge means saying that a particular practice is now simply how work is done here, which is an expectation, and expectations are set in job descriptions, in meeting agendas and in what leaders ask to see, not in a lunchtime session. Diagnostic: is there anything a member of staff would now be asked about if they were not doing it?
Answer for a mistake
A champion can be sorry. Only a named leader can be answerable.
AI produces confident, plausible, wrong output, and eventually one of those lands somewhere visible. If the only person associated with the practice is a volunteer with no formal accountability, two bad things happen at once: the volunteer is exposed to something they never agreed to carry, and the organization discovers it has no answer beyond "a colleague was trying something." Diagnostic: if an AI-assisted piece of work went wrong tomorrow in front of a parent or an inspector, could you name the person who answers, and would they already know?
Three signs you have hit the ceiling
The ceiling announces itself in three ways, and all of them are usually misread as a people problem.
The network stops growing and the same faces return. Attendance at the sessions is not falling because the sessions got worse. It is falling because everyone reachable by peer demonstration has been reached.
The champions start asking for rules rather than for tools. When your most enthusiastic people begin asking for a policy, a position statement, or a line in the handbook, they are telling you they have hit the edge of what enthusiasm can authorize. It is the most useful signal the program will ever produce and it is routinely heard as negativity.
Everything on show is still additional, and nothing has stopped. If a year of champion-led work has not removed a single requirement, the practice has been added to the organization rather than adopted by it. That distinction is the whole of what it takes to move AI from individuals to the organization, and it is worth testing before you renew anything.
The Leadership Question
The question I put to leadership teams about their champions is not "how are they doing?" It is: what have they made possible that we have not yet made permissible?
Almost every team I ask can answer the first half immediately. Champions are visible, likeable people and their work is easy to describe. The second half produces a silence, because it moves the subject from the volunteers to the table. That is the entire point of asking it. Champions generate a queue of decisions that only the leadership team can take, and a program that has been running for a year has usually generated a long one.
My own version of this came from running digital strategy in further education, where the enthusiastic-individual pattern long predates AI. Every wave of technology arrived the same way: a few people got good at it, a leadership team admired them, and the practice stayed exactly as wide as the people prepared to run lunchtime sessions about it. As a school trustee I have seen the governance end of the same story, where a board is shown impressive individual practice and is never asked to approve anything, which feels like good news and is actually a gap in the paperwork. The technology changes. The mistake does not.
I should be fair to the model here, because I am asking a lot of it. Champions are not failing at leadership. They were never given it.
What to do when you reach the ceiling
Three moves convert a stalled champions program into something that keeps working, and they are leadership moves rather than program tweaks.
Convert the demonstration into a decision. Take the practices your champions have proved and turn each one into a written line: this is approved, for this purpose, with these limits, owned by this person. A demonstration reaches the people in the room. A decision reaches everyone who was not. Skipping this step is the same failure that leaves AI pilots going nowhere after a promising first term.
Pair every champion with a named sponsor who holds authority. Not a mentor and not a line manager checking in. A senior leader whose job is to take the queue of decisions the champion has generated and get answers to them, and whose own name goes on the result. The digital leader in the Ofsted report was describing exactly this when they said what you need is someone with leadership responsibility. Most organizations have half of that person already.
Retire the program on purpose when its work is done. A champions program is scaffolding. Once the practices it proved are approved, expected and owned, keeping it running signals that AI is still a special interest rather than part of the work. Say what it achieved, say what has been absorbed into normal practice, and close it deliberately rather than letting it fade and calling that a failure.
Where none of the leadership conditions underneath this have been met, the champions program is not the thing to fix first. Stalled adoption is a leadership problem before it is a program problem, and the order matters.
When a champions program is exactly the right thing
One case makes the champions program the right answer, and the argument above is too tidy without it.
If your organization is at the beginning, has no internal evidence that any of this works in your own building, and has a leadership team that would otherwise be deciding on other organizations' examples, then a champions program is the fastest and least damaging way to generate real ones. In that phase its job is not adoption. Its job is evidence, and it should be judged on whether it produces a queue of decisions worth taking rather than on how many colleagues it converted.
The failure mode is not starting one. It is leaving one running for three years and reading its slow decline as a lack of appetite among staff. A champions program that is still the main AI mechanism in its third year is not a success story. It is a decision the leadership team has been avoiding, staffed by volunteers.
Where this usually goes next
If your champions keep bringing you questions rather than examples, the answer is rarely another round of recruitment. It is a session with the leadership team: turning what the champions have proved into decisions with owners, and building the expectations and the accountability that let the practice hold without them. That is the kind of work I support through AI strategy sessions and implementation advisory work with leadership teams.
Sources and further reading
- McKinsey & Company, The state of AI in 2026: On the Road to ROI, 25 August 2026.
- RAND, AI Use in Schools Is Quickly Increasing but Guidance Lags Behind: Findings from the RAND Survey Panels, 30 September 2025.
- Ofsted, "The biggest risk is doing nothing": insights from early adopters of artificial intelligence in schools and further education colleges, 27 June 2025.
Dan Fitzpatrick is the founder of The AI Educator, a former Director of Digital Strategy in further education and a school trustee, and works with leadership teams on AI strategy, adoption and governance. More about Dan.
