When AI adoption stalls, the first purchase is almost always the wrong one. The leadership team looks at the licences nobody is using, or the training day that changed nothing, and concludes that staff lack skills or that the tool was wrong. So they buy more training, or a different tool. Six months later the numbers look the same.
I have watched this cycle repeat in schools, trusts, colleges and companies, and the diagnosis is nearly always off by one level. The problem sits above the staff who are meant to adopt, not with them. It is a set of leadership conditions, and no amount of training fixes a leadership condition.
The Observation: Stalled Adoption Is Rarely a Skills Problem
Across the organisations I work with, stalled AI adoption is rarely a skills problem, and the evidence is usually in their own usage data: a small group using AI heavily, often well, and a large group who tried it once or twice and stopped. Both groups had the same training and the same licences. What separates them is whether anyone around them made it clear that using AI was expected, useful and safe.
Having trained more than 150,000 educators across more than 30 countries, I have watched the same session land in very different organisations. The content changes little from one room to the next. The outcomes change enormously: in one organisation the training becomes daily practice within weeks, in another a pleasant memory. What predicts which happens is not the trainer or the energy in the room but what the leadership team did in the month before the training and the fortnight after it.
Why the Misdiagnosis Matters
The misdiagnosis matters because it sends the organisation's effort to the one place the problem is not, and each round does quiet damage: staff learn that AI initiatives arrive, are announced and pass, and by the third attempt the sensible response is to wait it out. The skills story is also convenient. If the problem is skills, the next move belongs to somebody else. If the problem is leadership, the next move is theirs, which I suspect is a large part of why the skills story is so popular.
The Four Patterns Underneath a Stalled Adoption
Underneath almost every stalled adoption I find the same four patterns, and each one is a decision the leadership team has not yet made.
Nobody has decided what AI is for
The first pattern is that nobody has decided what AI is for in this organisation, so staff guess, and most guess conservatively. "Use AI to save time" is not a decision, it is an invitation to an argument. The decisions that move adoption are specific: AI is for first drafts of parent communication, for feedback on practice answers, for summarising case notes before a review. This is the first question of the Readiness Test, and if your team cannot yet answer it consistently from the top to the front line, it is worth reading what it actually means to be AI ready before doing anything else about adoption.
Nobody senior has been seen using it
The second pattern is that no senior leader has been seen using AI for real work, so staff conclude, reasonably, that it is for other people. I ask leadership teams one question: in the last month, has a member of staff watched you use AI to do part of your actual job? Not a demonstration, your job. The answer is usually no. The common version is the leader who mentions AI in every briefing and drafts every briefing by hand. Staff notice the second thing.
The thing it replaces is still required
The third pattern is that the work AI was meant to reduce is still required in full, so adoption becomes pure addition. A school introduces AI to lighten marking and keeps the policy that demands written comments in every book. A firm introduces AI for report drafting and keeps a review process that assumes a person typed every sentence. Adoption in those conditions is a second job, and nobody takes a second job voluntarily. Leaders own the policies, templates and review processes. Only leaders can subtract.
It is not safe to be seen using it
The fourth pattern is that staff do not know whether being seen to use AI will be read as initiative or as cutting corners, so they either hide it or stop. This is where most shadow use comes from: not defiance, but uncertainty. The teacher who plans with AI at home and never mentions it at school, in case a colleague or a parent thinks less of them, is not resisting. They are protecting themselves from a judgement nobody has ruled out. Visible practice is what spreads adoption, and it only happens where it is safe.
Permission Is Not Sponsorship
Permission is not sponsorship, and most organisations that believe they have supported AI adoption have only granted permission. Permission removes a prohibition: you may use these tools, within these limits. It is necessary, and it lives in a policy. Sponsorship adds a commitment: a named leader who uses the tool, protects time for others to learn it, takes the first mistakes on their own shoulders, and changes what the organisation asks for so the new way of working has room. Permission is a sentence in a document. Sponsorship shows up in a leader's diary, in what they ask for in meetings, and in what they stop asking for.
The test of which one you have is simple. Permission can be delegated to a policy. Sponsorship cannot be delegated at all.
Why the Middle Layer Decides Adoption
The middle layer decides adoption, because the people staff actually watch are not the executive team but the heads of department, team leaders and line managers they see every day. Senior leaders decide whether AI is allowed. Middle leaders decide whether it is expected, through what they ask for, what they accept, what they praise and what they quietly correct. A head of department who says "show me how you did that" when a teacher mentions AI-assisted feedback changes more behaviour in a week than the AI strategy did in a year.
It is also the layer most organisations forget to equip. Asked to check AI output, coach their teams and keep delivering with nothing removed from their load, middle leaders default to the old way, and the old way is what their teams copy.
What the Research Says
The research now points the same way: adoption follows the behaviour of leaders and managers far more than it follows access to tools.
Microsoft's 2026 Work Trend Index, published in May 2026, found that where managers actively modelled AI use, employees reported a 17-point lift in the value they got from AI and a 22-point lift in critical thinking about their own use, and where managers made experimentation feel safe, employees were 1.4 times more likely to be frequent users. Only one in four AI users said their leadership was clearly and consistently aligned on AI, and organisational factors accounted for more than twice the reported AI impact of individual ones. That last finding is this article in a number: the organisation, not the individual, decides most of what AI delivers.
In education the pattern is the same. A Gallup and Walton Family Foundation survey of 2,069 US teachers, reported by Education Week in May 2026, found that 82 per cent had received no formal guidance on applying AI to their work and a third had received none at all. Teachers are not refusing to adopt. They are waiting to be told what their organisation wants.
On the middle layer, Julia Shin and Sandra Sucher's June 2026 research in Harvard Business Review, from interviews across two consulting firms, found senior leaders focused on strategy, junior staff enjoying large productivity gains, and middle managers absorbing the checking and coaching with unchanged delivery pressure and no formal support. Every adoption plan has to pass through that overloaded middle.
What I Have Learned From Working With Organisations
What I have learned from organisation-wide implementation work is that adoption follows what leaders visibly do, not what they announce, and that most of what gets labelled resistance is a rational response to unclear leadership.
The staff labelled resistant are often the ones taking the change most seriously. They can see that using AI for feedback changes what feedback is, and they want to know what happens when it is wrong and who carries the accountability. Those are leadership questions, and when leaders have not answered them the silence fills with caution. The teachers I have seen written off as resistant usually just needed time and space: time to see AI used well by someone they respect, and space to try it without an audience. Give them both and most of them move.
Change of this kind puts an organisation in a liminal space for a while: the old rules no longer quite hold and the new ones are not yet written, and leadership's job is to hold that space deliberately rather than pretend it has closed. That is why, before any organisation-wide training I run, I ask two things of the leadership team: that they can say in one sentence what AI is for here, and that at least one senior leader will show real use of it in front of their staff. Where either is missing, I have learned to expect the training to be enjoyed and forgotten.
The Sponsorship Test
The Sponsorship Test is four questions a leadership team must be able to answer with a yes before it blames a stalled AI adoption on skills or tools: Have we decided what AI is for here? Have we been seen using it? Have we removed what it replaces? Have we made it safe to be seen using it?
Each no points to a piece of leadership work rather than a purchase. A no to the first means the strategy is not finished, and how to create an AI strategy that people actually use is the place to start. A no to the second means the senior team's diaries change first: every senior leader picks one real task from their own role and does it with AI, in front of colleagues, within the month. A no to the third means a policy, a template or a review process is rewritten by the people who own it, and the change is announced. A no to the fourth means writing down what happens when AI gets something wrong, so the first mistake is handled by a rule rather than a mood.
Give yourselves one term to change what staff can see, brief middle leaders before anyone else, and then run the training. By then it will have somewhere to land.
Working Through This With Your Leadership Team
If your organisation has trained its people, bought the licences and still watched adoption stall, the work that usually helps is with the leadership team rather than the staff: deciding what AI is for, making sponsorship visible, and building the implementation plan around the middle layer. This is the kind of work I support through organisational development and implementation support for leadership teams.
Dan Fitzpatrick is the founder of The AI Educator and has trained more than 150,000 educators across more than 30 countries on AI adoption and implementation. More about Dan.
