EdTech investment is quietly moving away from general-purpose AI and towards specialised AI trained on domain-specific data — and the clearest evidence sits in the DfE's June 2026 EdTech market assessment, Assessment of the education technology market in England. EdTech here means the use of technology to support and enhance education: tools designed to improve learning, reduce staff workload and make operations more efficient. The report does not use the phrase "vertical AI" on its cover, but its forward look says the quiet part plainly, and for any multi-academy trust (MAT) — a group of schools run under one legal structure — or IT director planning a 2026/27 roadmap, it is the most consequential signal in the document.
What does the DfE mean by "specialised AI"?
Specialised AI — often called vertical AI — is artificial intelligence built for one narrow domain rather than for open-ended conversation. Instead of a model that has read a little of everything, it is a system engineered around a single body of rules: in education, that means curricula, exam board mark schemes, examiner reports and special educational needs and disabilities (SEND) requirements. The DfE treats artificial intelligence and machine learning as a horizontal capability (p.7) that cuts across every EdTech vertical — which is precisely why the interesting question is no longer "does it use AI?" but "what is the AI specialised to do?"
The contrast is with general-purpose GenAI — generative AI such as ChatGPT or Claude that produces fluent text from any prompt. General models are broad and improvisational; specialised models are narrow and accountable to a defined standard. That accountability is the whole point in a school, where so much of the work is rule-bound and verifiable rather than open-ended.
Why is general-purpose AI losing its EdTech edge?
The DfE's forward look is unusually direct about where the value is heading.
The economics are straightforward. A general-purpose model is, increasingly, a commodity — every school can reach ChatGPT or Claude, so it confers no lasting advantage on the platform that merely wraps one. What is defensible is proprietary domain data: curriculum mappings, mark schemes, examiner reasoning and pedagogy. The DfE found UK EdTech to be a substantial market — 1,123 companies, £6.5bn turnover and 8.8% average annual growth (DfE, June 2026, p.7) — but growth and capital are not spread evenly across it. They are concentrating where the data moat is deepest, which is why an "AI-wrapped" product and an "AI-native" one are starting to diverge sharply in the eyes of buyers and investors alike.
Where is the £187m in personalised-learning investment going?
The single most telling number in the report is a paradox. According to the DfE (June 2026, p.8), the personalised-learning vertical attracts £187m in annual investment while turning over just £77m a year — and grows at only 3.1%, the lowest rate of any vertical. Investors are not pricing today's revenue; they are pricing a bet that domain-trained, specialised AI will unlock future demand. Set that against the assessment vertical, where investment of £24.3m sits on £470m of turnover and the highest growth rate in the market.
| EdTech vertical | Annual turnover | Annual investment | Annual growth |
|---|---|---|---|
| Personalised learning | £77m | £187m | 3.1% (lowest) |
| Assessment technologies | £470m | £24.3m | 18.1% (highest) |
| Management & admin | £1.39bn | £219m | — |
| Immersive learning | £287m | £158m | — |
Source: DfE (June 2026), Assessment of the education technology market in England, p.8.
Read together, these rows describe the same thesis from two directions: personalised learning shows investors paying a premium for specialised-AI potential, while assessment shows a vertical already converting that potential into revenue and growth. The capital is chasing domain depth, not chatbot breadth.
What does specialised AI look like in practice?
Stripped of the abstraction, specialised AI in schools falls into a few recognisable categories:
- Marking against mark schemes. Assessment tools that take a student's answer, retrieve the correct exam board mark scheme, apply the level descriptors and return examiner-style feedback. ReMarkAble AI is one example in this category: it marks work against AQA, Edexcel, OCR and WJEC mark schemes from KS1 (the first key stage of primary, ages 5–7) through to A-Level, returning a grade, strengths, areas for improvement and next steps. It is one tool among several in a fast-growing field — not a substitute for a teacher's professional judgement.
- SEND-specific tools. Systems tuned to particular needs — adapting reading level, scaffolding writing, or supporting communication — where a general model's one-size answer is exactly the wrong instinct.
- Subject-specific content generation. Generators that produce questions, worked examples or resources tied to a specific scheme of work and specification, rather than generic material a teacher must then re-fit.
What unites them is constraint. A general chatbot will cheerfully invent a mark scheme that sounds plausible; a specialised tool is held to the real one. The schools getting value from general-purpose GenAI tend to use it for the open-ended layer — drafting and planning — and reach for specialised tools where the output must be right against a published standard. We explore that division of labour further in our piece on ChatGPT and general-purpose GenAI in schools, and the accountability question specifically in examiner-quality AI marking and moderation.
How should school leaders think about this in their 2026/27 roadmap?
The practical takeaway is not "buy specialised, drop general" — it is to be deliberate about which job each kind of AI does. A few principles for a MAT digital lead or IT director planning the year ahead:
- Ask what the AI is specialised to. "It uses AI" is no longer a differentiator. Probe whether a tool is wired to real mark schemes, specifications and SEND frameworks, or whether it is a thin wrapper around a general model.
- Match the tool to the risk. Use general-purpose GenAI for low-stakes, open-ended work; insist on specialised, auditable tools where outputs are standards-bound, such as marking or reporting.
- Favour defensible data. Tools built on proprietary curriculum and assessment data are likelier to keep improving and to survive the shake-out the DfE's figures imply.
- Plan for a widening gap. Over the next 18 months the distance between AI-wrapped and AI-native tools will grow; today's "good enough" general tool may not keep pace in a specialised role.
For trusts already running established assessment platforms, the migration question is its own topic — we cover it in moving from Exampro to AI-led classroom assessment. The headline, though, is simple: the DfE's data is telling buyers where the market is going long before the marketing does.
See specialised AI marking in practice
ReMarkAble AI is built for one job: marking student work against real UK exam board mark schemes, from KS1 to A-Level, with examiner-style feedback on strengths, areas for improvement and next steps. Try it on a real piece of work and see what AI-native — rather than AI-wrapped — assessment feels like. Free to start, no card required.
Frequently Asked Questions
What is specialised AI in education?
Specialised AI — sometimes called vertical AI — is artificial intelligence built and trained for a single, narrow domain rather than for general conversation. In education that means a system shaped around curricula, exam board mark schemes, examiner reports and SEND requirements, so its outputs are constrained by the rules of the task. A general-purpose model has read a little of everything; a specialised model has been engineered to apply one body of rules reliably. The distinction matters in schools because assessment and curriculum work are rule-bound and verifiable, which is exactly where a general chatbot tends to improvise.
How is specialised AI different from ChatGPT?
ChatGPT is a general-purpose GenAI platform — GenAI meaning generative AI that produces fluent text or images from a prompt. It is excellent at drafting, summarising and rephrasing across any subject, but it does not know that a GCSE History 16-marker is assessed against specific assessment objectives, nor which mark scheme an OCR paper uses. A specialised education tool is wired to that structure: it retrieves the correct mark scheme, applies the level descriptors and returns feedback in the format an examiner would recognise. In short, ChatGPT is broad and improvisational; specialised AI is narrow and accountable to a defined standard.
Why is investment moving from general-purpose to specialised AI?
The DfE's June 2026 EdTech market assessment found investors pouring £187m into personalised-learning EdTech that turns over only £77m a year — money priced on future demand for domain-trained AI rather than current revenue. The report's own forward look states that investment is shifting towards specialised AI solutions and that general-purpose AI may erode the advantage of legacy platforms. The logic is that a general model is a commodity available to everyone, whereas proprietary curriculum data, mark schemes and pedagogy are defensible. Capital follows defensibility, so it is flowing towards tools that own a vertical rather than wrap a chatbot.
What are examples of specialised AI in UK schools?
Specialised AI in UK schools clusters into a few categories: assessment tools that mark student work against exam board mark schemes; SEND-specific tools that adapt reading, scaffolding or communication for particular needs; and subject-specific content generators that produce questions or resources tied to a scheme of work. ReMarkAble AI is one example in the assessment category — it marks work against AQA, Edexcel, OCR and WJEC mark schemes from KS1 to A-Level and returns examiner-style feedback. The DfE report identifies assessment as the fastest-growing vertical, so this category is where the trend is most visible. None of these tools replace a teacher; they handle the rule-bound, repetitive layer.
Will ChatGPT replace specialised AI tools?
It is more likely the two settle into different jobs than one replacing the other. General-purpose platforms are already embedded in schools for lesson planning and admin, and they will keep improving as broad assistants. But the DfE notes that general-purpose AI may reduce the competitive advantage of legacy platforms — not of purpose-built vertical tools whose moat is domain data and accountability. For high-stakes, standards-bound tasks such as marking against a mark scheme, a tool engineered for that single job will tend to be more consistent and auditable than a general chatbot. The pragmatic view: use general AI for breadth, specialised AI where the answer must be right against a defined standard.