Eight suppliers, 53 bids, £300,000 each: Britain turned AI tutoring into a procurement line

Medly AI, a London exam-prep platform built by two former NHS doctors, has raised an $8m seed round led by Felix Capital, with existing backers Eka Ventures and Ada Ventures returning ^[1]. The more consequential paperwork is older and smaller: in June the company was one of eight suppliers picked from 53 bids for the AI Tutoring Tools Pioneers Programme, run by the Department for Science, Innovation and Technology with the Department for Education as delivery partner ^[2].

Each winner takes £300,000 excluding VAT on research contracts running to 31 March 2027, aimed at raising attainment among disadvantaged pupils in England ^[3]. That is roughly £2.4m across the cohort ^[1], which buys nothing at national scale and is not trying to. What it buys is a named customer, a named target population, and a date by which someone expects to see findings. Eight awards from 53 bids is a success rate of about 15% ^[2], which tells you the department had enough applicants to be selective and chose to spread the money thinly rather than back one horse.

The cohort composition is the other signal. Pearson, ElevenLabs and Eedi are also on the list ^[4]: an incumbent publisher, a voice model company, and a maths startup, in the same lot. The buyer is not procuring a product shape called "AI tutor". It is procuring tutoring capability from whoever can evidence it.

That matters because the evidence currently on offer from the sector is the kind that a research contract exists to replace. Medly says 74% of students who used it for last summer's GCSEs improved by at least one grade and 41% improved by two or more, figures drawn from a survey of its own users rather than an independent assessment ^[5]. Taken together those numbers imply about a third of respondents moved exactly one grade ^[4]. The company puts its largest subject gains at 46% and 43%, but its own website attributes the 46% to English Literature while the announcement attributes it to Physics ^[6]. It claims more than 400,000 users in the UK alone while its consumer site advertises "300,000+ students", a gap of a third ^[7][5].

The more serious work is unfinished. In June three members of the team published a preprint benchmarking 32,534 double-marked GCSE responses across 328 questions and five subjects, finding that leading off-the-shelf models agreed with examiners at least as closely as two examiners agreed with each other ^[8]. That paper tests general-purpose models rather than Medly's own marking stack and has not been peer reviewed ^[9]. A randomised controlled trial of more than 1,000 pupils in lower-attaining schools is complete but unpublished ^[10].

The product itself marks practice answers and explains where the marks went through a conversational interface ^[11], runs on what the company describes as a proprietary combination of seven large language models ^[12], and accepts stylus handwriting on a tablet, which matters in maths and chemistry where the working is the answer ^[13]. Medly intends to build its own models for UK and EEA marking and pedagogy rather than renting general-purpose ones indefinitely ^[14]. Meanwhile it has added the SAT in the US on top of GCSEs, A-levels and the IB, with APs and ACTs planned before year end ^[15].

Worth noting the proportions: £300,000 is about 18% of the £1.7m round Medly announced in February 2025 ^[16][3]. Nobody is taking this contract for the revenue.

Watch three things. Whether the completed RCT gets published, and whether it says what the user survey says. Whether the Pioneers Programme publishes comparable results across all eight suppliers before 31 March 2027, or whether the evidence stays proprietary. And whether the own-model plan ships before the contract ends, given the US expansion competing for the same engineers.