What does the DfE's 2026 data tell us about classroom assessment use?
Classroom assessment in England runs on two layers, and only one of them is solved. The DfE's June 2026 report, Assessment of the education technology market in England, found that EdTech — technology used to support and enhance education — is now a £6.5bn turnover sector spanning 1,123 UK companies (DfE, June 2026, p.7). Within it, assessment technologies are the fastest-growing vertical, expanding 18.1% a year (DfE, June 2026, p.8). The report describes question banks as deeply embedded in day-to-day teaching, with classroom and assessment tools the second-highest area of engagement among schools (DfE, June 2026, p.9). What it also shows, between the lines, is where the embedded layer stops: at the question. The marking and feedback that turn a practice paper into learning are still done by hand.
The report singles out one platform by name to make the point about embeddedness:
That is a striking signal. A platform held under multiple contracts within a single school, across departments, is not a pilot — it is infrastructure. The DfE also notes that this kind of tool is used cyclically, spiking around assessment periods (DfE, June 2026, p.9). Teachers reach for the question bank when mocks and end-of-topic tests loom — precisely the moments when the marking burden is heaviest. The questions arrive instantly; the feedback does not.
Why did question banks dominate the last decade?
Question banks won because they solved a real, narrow problem extremely well: sourcing high-quality, board-aligned questions at scale. Before platforms like Exampro, building a differentiated mock or a targeted topic test meant a teacher hunting through past papers, photocopying, and reformatting by hand. A question bank collapsed that into a search-and-filter task — by board, by topic, by mark tariff — and made the same library available to every department in the school. The cross-departmental adoption the DfE describes is the natural result: once one faculty proves the time saved, others follow.
But the value stopped at distribution. A question bank hands you a question and a mark scheme; it does not read a student's thirty-line History answer and tell them their argument lacked sustained evaluation. That second half — the marking and the feedback — remained stubbornly manual, and it is the half that actually drives improvement. For a fuller account of why that workload sits where it does, see our piece on how many hours UK teachers spend marking.
What's the missing layer — and why AI fills it
The missing layer is feedback, and it is no coincidence that it is where the money and the growth are now moving. The DfE attributes the assessment vertical's sector-leading 18.1% growth (DfE, June 2026, p.8) to a clear driver:
Reducing workload is, concretely, about marking. Generative AI — GenAI, AI that produces text rather than just retrieving it — can now read an extended written answer, weigh it against a mark scheme and return examiner-style feedback in minutes. That is the capability the previous generation of tools never had. The two layers are best seen not as rivals but as a stack, each supplying what the other cannot:
| Layer | Question-bank era (e.g. Exampro) | AI-feedback era |
|---|---|---|
| Sourcing questions | Search, filter and assemble board-aligned papers | Not its job — consumes the questions you already have |
| Setting the assessment | Strong — topic, tariff and board controls | Neutral — works with any paper the teacher sets |
| Marking the scripts | Manual — teacher applies the mark scheme by hand | Automated — reads the answer against the mark scheme |
| Returning feedback | Limited to a mark scheme the student self-checks | Grade, strengths, areas for improvement and next steps |
| Turnaround | Days to weeks per class set | Minutes per script |
Read across the rows and the relationship is obvious: there is almost no overlap. The question bank owns the top of the table; the AI marker owns the bottom. ReMarkAble AI is one example of this newer layer — a UK-built platform that marks student work against AQA, Edexcel, OCR, WJEC and Eduqas mark schemes from KS1 through A-Level and returns examiner-style feedback rather than the questions themselves. The point is not the brand; it is that the row labelled "returning feedback" now has something in its right-hand column for the first time.
How does the practice-to-feedback loop change with AI?
The shift is from "here's a question" to "here's a question, marked, with feedback and next steps" — and that completes a loop the question-bank era left half-open. The improvement cycle that actually works is: set a practice answer, mark it, return specific feedback, have the student act on it, then set the next one. Question banks made the first step instant and left the rest to a teacher with a marking pile. When marking takes minutes instead of a fortnight, the loop can turn several times in the run-up to a mock rather than once.
This does raise a fair question about whether AI feedback can be trusted to the standard of an examiner. It is a question worth taking seriously, and we treat it head-on in our piece on examiner-quality AI marking and moderation. The short version for a head of department: AI marking is best used as formative feedback that a teacher moderates, not as an unsupervised summative grade.
What's the right way to introduce AI feedback alongside existing tools?
Do not rip anything out. The whole argument of the DfE's data is that question banks are embedded for good reason — so the sensible move is to add the missing layer on top of the one your department already trusts. A practical sequence for a head of department:
- Keep Exampro for what it's good at. Carry on sourcing and setting board-aligned papers exactly as you do now.
- Run one class set through an AI marker. Pick a single essay-based assessment and feed the completed scripts through, so the comparison with hand-marking is concrete.
- Moderate the output. Have the teacher sample a handful of AI-marked scripts against their own judgement before trusting the feedback at scale.
- Measure the loop, not the tool. Track how many practice-and-feedback cycles students complete before the next assessment — that is the metric the new layer actually moves.
Close the feedback loop — try one class set
Keep the question bank you already trust. Take one essay-based assessment, run the class set through ReMarkAble AI, and see board-aligned feedback — grade, strengths, areas for improvement and next steps — come back in minutes instead of a fortnight. Free to try, no card required.
Frequently Asked Questions
What's the best AI marking software for UK schools?
The right tool is one built specifically for the UK curriculum, rather than a repurposed general-purpose chatbot. Look for marking against the exam boards your department actually uses — AQA, Edexcel, OCR, WJEC and Eduqas — and coverage across the key stages you teach. A purpose-built platform marks against the real mark scheme and returns examiner-style feedback (a grade, strengths, areas for improvement and next steps), not a generic opinion on the writing. Just as important is fit with your existing workflow: the best choice complements your question bank rather than asking you to abandon it.
How does AI marking work alongside existing assessment tools?
Question banks and AI marking sit in different layers, so they stack rather than compete. A platform like Exampro supplies the questions, past papers and topic filters; an AI marking tool takes the completed scripts and returns structured feedback against the mark scheme. In practice a teacher sets a past paper from their question bank, students complete it, and the scripts go through the AI marker — closing a loop the question bank alone leaves open. Nothing about your current set-up needs to be torn out; the AI layer slots on top of the practice you already generate.
Can AI mark KS2 SATs practice papers?
Yes, for the writing-led elements where teacher judgement and the Teacher Assessment Framework matter most. AI marking is most useful for extended writing and reasoning tasks where a child's work has to be weighed against descriptors rather than simply ticked right or wrong. For arithmetic and short-answer reasoning papers, a standard mark scheme and a tick are perfectly adequate and AI adds little. The value at KS2 is in giving children faster, criteria-aligned feedback on writing so they can practise, revise and improve before the assessment period rather than after it.
Is AI marking better than peer marking?
They do different jobs, and the honest answer is that AI marking is more consistent while peer marking builds metacognition. Peer marking helps students internalise what a good answer looks like, but novices apply mark schemes unevenly and rarely give reliable feedback on analytical depth. AI marking gives every student the same criteria-aligned read in minutes, which is hard to achieve across thirty peers. The strongest classroom routines use both: AI for fast, consistent feedback on every script, peer marking as a deliberate exercise in understanding the standard.
How long does AI marking take per paper?
Minutes per script, rather than the days or weeks a teacher-marked class set can take. A student uploads a photo or scan of handwritten work, optical character recognition converts it to text, and the examiner-style feedback comes back in a single sitting. The practical effect is on the feedback cycle, not just the clock: instead of one marked attempt per fortnight, a class can run several practice-and-feedback cycles in the same window. That speed is exactly why the DfE identifies workload-driven demand as the engine behind assessment EdTech's growth.