The £23m EdTech Testbeds: England's Bet on Evidence Over Hype
The DfE is funding a four-year, 1,000-school trial to test AI and edtech claims properly before procurement — schools should watch the methodology, not just the headline figure.
A different kind of AI announcement
Most AI-in-education news this year has been about mandates, bans or vendor launches. On 21 January 2026, at the BETT UK conference, education secretary Bridget Phillipson announced something narrower and, for once, less flashy: a £23 million expansion of the Department for Education's EdTech Testbeds programme, running for four years and involving more than 1,000 schools and colleges across England from September 2026.
The pitch is not "here is a new AI tool." It is "we do not actually know which AI tools work, so we are going to find out properly." For a sector that has spent two years fielding vendor claims about workload savings and attainment gains with little independent verification, that is a meaningfully different starting point.
What the programme actually does
The expansion builds on an earlier nine-month pilot and is delivered with the Education Endowment Foundation, with RAND Europe supporting the DfE in deciding where testing will add the most value. According to RAND Europe's published project description, the research focuses on three policy priorities: reducing teacher workload, improving teaching and learning outcomes, and increasing inclusion, including for pupils with SEND.
In practice that means recruiting primary, secondary and further education settings to trial edtech and AI products in real classrooms, then running a mix of rapid evaluations and more robust studies — described by the DfE as building "a stronger evidence pipeline" rather than a single verdict. The department says it has already had more than 280 expressions of interest from edtech companies wanting products included.
The programme's premise is blunt: schools have been asked to adopt AI tools faster than anyone has been able to test whether they work.
Why this matters more than another product launch
NeuralClass has covered plenty of tools that promise to cut marking time or personalise learning, and plenty of research showing the evidence base for those claims is thin — from mixed results on AI tutoring platforms to surveys showing high student use of AI study tools paired with low trust in the output. The Testbeds programme is the DfE's attempt to close that gap systemically, rather than leaving individual schools to run their own informal trials with no comparison group and no independent oversight.
That matters for procurement. Multi-academy trusts and school leaders currently have to weigh vendor marketing against very little independent UK evidence, often under pressure to be seen to be "doing something" on AI. A national evidence pipeline, even a slow one, gives buyers something to point to other than a supplier's own case study.
The caveats worth holding onto
Four years is a long commitment period for a technology category that changes every few months — a tool tested and found workload-negative in year one may be unrecognisable by year three, and the evaluation cycle needs to keep pace or the findings risk describing products that no longer exist in that form. The jump from 280 vendor expressions of interest to a curated testbed also raises an obvious question: who decides which products get tested, on what criteria, and how conflicts of interest with participating suppliers are managed. Neither the DfE nor RAND Europe has yet published the selection methodology in detail.
There is also a timing coincidence worth flagging. The Testbeds pilot begins in September 2026 — the same month schools must have fully implemented the KCSIE 2026 safeguarding rules covering AI and deepfakes. Two separate DfE-driven AI workstreams landing on school leaders in the same month, one about safeguarding compliance and one about voluntary evaluation, is a lot to coordinate for a senior leadership team that may have one person covering both briefs.
What to do
- If your school or trust wants to take part, look for recruitment routes through the DfE, the Network of Excellence (NEN) and Nesta, which have previously run related testbed and demonstrator programmes — apply with a specific problem you want tested (marking time, SEND support, a particular subject) rather than a general interest in AI.
- If you are not participating, use the programme as leverage anyway: when a vendor pitches a tool, ask whether it is part of the Testbeds cohort or has independent evaluation data, and treat the absence of either as a reason to pilot cautiously and locally before wider rollout.
- Keep procurement decisions reversible. Build short review points into any AI tool contract so you are not locked into a product for years before evidence on it exists.
What to watch
Watch for the DfE and EEF to publish the testbed selection criteria and the list of products and use cases chosen for the first cohort — that will show whether the programme is testing what teachers actually need help with, or what vendors are most eager to have validated. Watch too for early findings on the SEND and inclusion strand specifically, since that is the area where evidence is currently thinnest and where the stakes of getting an AI tool wrong are highest.