Most schools I talk to have the same AI problem. Somebody has seen a demo, somebody else has a free trial running, and nobody is quite sure who decides whether it stays.

It starts as a buying problem. And the evidence I've been reading this term says schools are mostly working it out on their own.

In the US, district leaders told Stateline they're carrying the burden of vetting AI products themselves. One superintendent said the onus lands entirely back on the individual district. Another said most tools on the market do very little that changes anything for students.

The UK picture isn't far off. So here's what I'd want any head, trust leader or IT lead to think about before signing.

The licence is the cheap bit

A UNESCO background paper by Keith Krueger and Frankie Jackson of CoSN makes a point every IT lead already feels in their bones. The purchase price is a small slice of what a tool really costs.

They cite UNESCO's 2023 GEM Report: initial investment is typically a quarter or less of technology's eventual cost. The same report found around two-thirds of education software licences in the US went unused.

AI makes this worse. A textbook doesn't change after you buy it. An AI tool gets updated continuously, so it can behave differently a month after the contract is signed.

Some tools also bill by usage, so every prompt draws on compute you're paying for.

The paper sets out seven layers of cost to plan for:

  1. Governance
  2. Applications
  3. Infrastructure
  4. Data
  5. Cybersecurity
  6. Workforce capacity, including ongoing CPD
  7. Day-to-day operations

The one I'd underline is workforce. AI training is never a one-off. The tools change, so the CPD has to keep going.

And be wary of free. The paper describes Kazakhstan receiving 165,000 education licences at no charge for three years. Acquisition cost: zero.

But someone still pays for the training, the data protection work and the integration. And what happens in year four, when thousands of teachers depend on a tool nobody bought?

I've watched the same pattern with M365 rollouts across schools and trusts. The licence was rarely what hurt. The real bill came after, in configuration and training.

Ask for the evidence, then read it properly

Trust in vendor claims is low. The EdTech Evidence Board report cites BESA research putting it at 8% of educators back in 2023, and more recent Teacher Tapp data at around 6%.

The Evidence Board was funded by the DfE and run by the Chartered College of Teaching to test whether EdTech evidence could be judged independently. Its September 2026 report is worth an hour of any leader's time.

The headline surprised me. Most providers do have evidence. What they rarely have is a joined-up story that runs from the educational need, through the product design and how it's used, to actual outcomes. Lots of case studies and user data, very little connecting them.

The pilot also found AI products were mostly too early-stage for the full review. And no formal awards were given, because the evidence wasn't consistent enough to rate fairly.

The baseline checks were just as telling. Providers often believed they met accessibility, data protection and online safety expectations but couldn't show it in writing. Accessibility and online safety were the weakest areas.

At the same time, the DfE's Science Advisory Council has published eight principles for decisions about fast-moving technology. Two jump out for anyone buying:

Keep options open. Avoid lock-in, single-supplier dependence and long commitments while the evidence is still thin.

Evaluate conditions alongside products. Whether a tool works depends on training, oversight and how it's deployed, and that's on you as much as the vendor.

They also suggest providers should hand over usage data as a condition of public sector procurement. I'd start asking for it now.

The contract is your best governance tool

A second UNESCO paper, by Kalervo Gulson and colleagues, calls procurement the missing lever in AI governance. Most guidance focuses on how AI gets used in class. Far less looks at the moment you buy, which is the one point where you can set conditions.

Their comparison with interactive whiteboard software made me smile. Whiteboard updates came as named versions: a new pen tool, a bug fix. An AI tool's behaviour can shift with every model update behind the same screen. What you tested in June may not be what's running in March.

They also point out that AI tools take data as well as deliver content. So the contract has to govern data flowing both ways.

Watch for two back doors. The first is donated tools, where the school becomes a recipient instead of a buyer and loses the chance to set terms.

The second is scope creep: AI features switched on inside platforms you already license. Plenty of schools are living with that one right now in Microsoft 365 and Google Workspace.

Some US districts have drawn hard lines. Allentown requires that all user data and chat logs stay the district's property, and that vendors state in writing that student data won't be sold or used to train models. New York City now bans vendors from training on student data.

And the cautionary tale: Los Angeles paid $3 million for an AI chatbot from a company that later went bankrupt, with its founder charged with fraud. The superintendent resigned.

Ask these before you sign

  1. What problem are we solving, and could we solve it without new tech?
  2. What will this cost over three years once training, support and integration are counted?
  3. If it's free now, what happens when the free period ends?
  4. Can the vendor show evidence that links our need to their design and to real outcomes, or just case studies?
  5. Can they show, in writing, how they meet accessibility, data protection and online safety expectations?
  6. Will our pupils' or staff data be used to train their models? Get the answer in the contract.
  7. Who owns the chat logs and outputs, and can we export everything if we leave?
  8. How will they tell us when the underlying model changes?
  9. Will they share usage data so we can see who's using it and whether it helps?
  10. Have staff tested it properly before it goes anywhere near pupils?
  11. Who in school owns the decision to keep it, change it or switch it off?

Anyone on your leadership team can ask these. The hard part is making the time, and being willing to walk away from a slick demo.

The tools will keep changing. Good buying habits protect your pupils while they do.

Which of these questions does your school already ask, and which one would trip up your current suppliers?

Sources