AI Tutor for Schools UK 2026: The Headteacher's Procurement & Buying Guide

Audience: UK headteachers, deputy heads, SENCOs, bursars, school business managers, and MAT (Multi-Academy Trust) trustees. Also relevant for UK local-authority advisers and SEND-caseworkers evaluating assistive technology.

Hook: Every edtech vendor is now an "AI tutor." Most are chatbots with a curriculum skin. The 17 questions in §4 filter out roughly 80% of weak platforms in under an hour. The compliance checklist in §3 protects you from a parent-data incident that ends a career. Skip both and you'll sign a 3-year contract with a tool that helps 11% of pupils and exposes 100% of them to a US data pipeline.

Tone: Plain, vendor-neutral, willing to flag where Grademy is the wrong choice. We sit on one side of the trade, so we go out of our way to disclose it.

Word count target: 2,400–2,800

---

Intro — Why UK Headteachers Are Buying AI Tutors in 2026 (and Why Most Are Buying Wrong)

In a January 2026 UK schools survey, 41% of secondary headteachers said they were evaluating or had just purchased an AI tutoring platform — up from 14% in the same survey 12 months earlier. The drivers are well-rehearsed:

• Post-pandemic catch-up that conventional intervention tutoring hasn't closed (NAO 2024–25 reports show KS4 progress gaps widening, not closing, in 60% of LAs).
• Teacher workload (DfE 2025: 71% of secondary teachers report unmanageable workload; AI-tools-for-marking-and-practice is the second-most-requested support).
• SEND capacity: EHCP demand up 38% since 2021, with no commensurate rise in specialist staff.
• Cost pressure: per-pupil funding flat in real terms; an AI subscription is cheaper than a 1-to-1 human tutor, even before considering availability.

But the same survey showed 52% of headteachers regretted their first AI pilot within 6 months — typically because the platform couldn't evidence pupil progress, didn't meet UK GDPR, fragmented across year groups, or was abandoned after a leadership change.

This guide is the procurement playbook that fixes those failure modes. It is built for:

• Headteachers who have a budget line but no evaluation framework.
• Bursars / school business managers who need ROI paperwork for governors.
• MAT trustees considering a trust-wide rollout.
• SENCOs evaluating tools for a specific cohort (autism, ADHD, dyslexia, EAL).

We will not push Grademy as the default. Where Grademy is the right answer, we'll say so. Where the right answer is "no AI tutor at all," or "a different platform," or "a human tutor with a smaller caseload," we'll say that too.

---

1. The 2026 UK AI Tutor Market — Who Actually Exists

Before you can evaluate, you need to know which categories exist. Below is the taxonomy we use when advising MAT boards. It is not flattering to most vendors — most collapse into 2 of the 4 categories.

Category	What it actually is	Typical UK price (2026)	Honest use case	Failure mode
Adaptive practice engine	Question bank with per-pupil difficulty routing, no open-ended tutoring	£2–£8/pupil/year	Maths fact fluency, GCSE question drilling	Looks like an AI tutor; isn't. Pupils can't ask "why".
LLM-with-curriculum-prompt	ChatGPT / Claude / Gemini wrapped in a school-branded prompt + RAG over a syllabus	£3–£12/pupil/year	Subject-agnostic tutoring across KS3–KS5	Hallucinations. No mastery tracking. No teacher dashboard.
Full-platform AI tutor	Adaptive practice + LLM tutoring + teacher dashboard + parent reporting + UK-hosted + UK GDPR-grade	£18–£45/pupil/year (Grademy: £24/pupil/year at volume)	Trust-wide deployment; replacement for some 1-to-1 intervention	Cost; requires onboarding; vendor lock.
Human-tutor marketplace	AI on the matching/booking layer, real tutors do the teaching	£30–£55/hr per pupil	Exam-year GCSE/A-Level catch-up	Availability caps; quality variance; doesn't solve teacher workload.

The mistake 80% of first-time buyers make: they buy a Category 2 (LLM-with-prompt) thinking it's a Category 3 (full-platform). It looks the same to a teacher in week 1. By week 6 the teacher realises there's no analytics, no safeguarding log, no UK hosting, no mastery model. The "AI tutor" becomes a glorified search engine.

Action: In your first vendor call, ask: "Are you Category 1, 2, 3, or 4, and which capability from a higher category do you explicitly lack?" If the vendor can't answer that in one sentence, walk.

---

2. Define the Problem in Procurement Terms Before You Look at Vendors

A surprising number of UK AI-tutor pilots fail because the school never wrote down what it was trying to fix. The three most common (and useful) problem framings:

Problem framing A — KS4 maths catch-up

"We have 87 Year 11 pupils predicted a 4 or below in GCSE maths. We want to raise 32 of them to a 5+ by summer 2027. Budget: £14,000 for the year. We will measure by mock result and final GCSE grade."

Problem framing B — KS3 reading / EAL cohort

"We have 41 pupils across Y7–Y9 who are 18+ months behind in reading age. We want 24 of them to close the gap by 12+ months by end of year. Budget: £6,500. We will measure by NTS reading age + teacher assessment."

Problem framing C — SEN / ADHD assistive layer

"We have 12 pupils with EHCPs for ADHD or autism. We want each of them to access 20+ additional minutes of focused practice per day without burning out their TA time. Budget: £4,200. We will measure by attendance + engagement + EHCP outcome progress."

Each of these needs a different AI tutor. A Category 1 adaptive practice engine is fine for A. B needs a full-platform with reading-age tracking. C needs the platform to be soothing, low-stim, offline-tolerant — most Category 3 platforms fail this because they were built for neurotypical pupils.

Before you brief any vendor, write down:

1. The cohort (year group, SEND flag, prior attainment).
2. The measurable outcome (mock result delta, reading-age delta, attendance delta).
3. The budget (per-pupil + total for the year).
4. The kill criterion (what result would make you end the pilot early).

If a vendor can't tell you whether their tool fits your framing in their first demo, they're selling, not advising.

---

3. UK GDPR + DPA 2018 Compliance Checklist (the Non-Negotiable)

This is the bit that ends careers if missed and goes unnoticed if done. Most US-headquartered AI tutor vendors fail this category outright. UK-hosted is non-negotiable for any platform handling pupil personal data.

3.1 The seven mandatory checks

Every AI tutor you evaluate must, in writing, answer YES to all of these:

#	Check	Why
1	Is pupil personal data stored in the UK or EEA by default?	UK GDPR adequacy requires it for any data not under a valid IDTA / SCC.
2	Does the vendor have a signed Data Processing Agreement (DPA) compliant with DPA 2018 s.7 + UK GDPR Art. 28?	A verbal assurance is not a DPA. Without it, your school is the data controller of an illegal sub-processor chain.
3	Is the AI inference done without sending pupil data to a US LLM API for training?	Inference-only is OK; training-on is a UK GDPR breach.
4	Can you delete a pupil's full record within 30 days of leaving?	UK GDPR right-to-erasure (Art. 17) + retention limits.
5	Does the platform log every AI interaction with a pupil, timestamped, user-attributed, exportable?	Safeguarding + KCSIE 2025.
6	Is there a documented sub-processor list (AWS / GCP / Cloudflare / inference provider) and can you receive change notifications?	Art. 28(2) — must be in the DPA.
7	Has the vendor completed a UK GDPR DPIA template for schools and can share the relevant sections?	Required for "innovative technology" deployments under DfE guidance.

If the vendor answers NO to any of these, stop. Even if the price is a third of the alternatives.

3.2 FERPA / COPPA cross-mapping (for US schools considering the same vendor)

If you're a US-based independent school, or running a US-headquartered MAT, you'll also need:

• FERPA (20 U.S.C. § 1232g): vendor must sign a "school official" exception contract. Schools are exempt from parental consent for disclosure to school officials with legitimate educational interest.
• COPPA (for under-13s): direct parental consent required for any vendor collecting personal data from under-13s — unless the school itself is the operator.
• State laws: California SOPIPA, New York SHIELD, Illinois SOPPA all add further obligations.

Grademy is FERPA-aligned, COPPA-compliant (school-as-operator model), and ships with US-state addenda. We list this here because most UK-headquartered vendors don't.

---

4. The 17 Vendor Questions That Filter 80% of Platforms

These are the questions we put to AI-tutor vendors when advising UK MATs. If a vendor can't answer at least 14 with evidence, they're not ready for procurement.

Architecture & hosting (3)

1. Where is pupil personal data stored at rest, and where does inference happen? (Specifically, name the region.)
2. Are you using a foundation model from a third-party API (OpenAI / Anthropic / Google) for inference? If yes, is your contract with that provider inference-only, or does it allow their use of prompts for training?
3. Can the school choose on-premises or UK-only inference for any cohort?

Pedagogy & evidence (4) 4. What is your mastery model? (How does the platform decide a pupil has learned something?) 5. Can you show me per-pupil outcome data from at least 3 UK schools using the tool? 6. How do you diagnose a misconception, not just a wrong answer? 7. What is your proposition for pupils who are stuck after 3 attempts on a problem?

Safeguarding (3) 8. Where is the safeguarding log stored, who has access, and how is it exportable to school DSLs? 9. What is your filter for prompt-injection (a pupil trying to get the AI to discuss self-harm, sexual content, drugs, weapons)? 10. Have you ever had a UK schools data incident? What happened, what did you do?

Teacher / school workflow (4) 11. How many clicks from a class teacher logging in to "see which of my Year 9 pupils are stuck on quadratic equations"? 12. Can my SENCO mark a pupil's account as needing calm mode (low-stim layout, no timer, font/colour controls)? 13. Can I export per-pupil progress to a CSV my data manager can ingest for my MIS (SIMS / Bromcom / Arbor / ScholarPack)? 14. Does the platform support whole-class assignment and individual self-directed study, or only one of those?

Commercial & exit (3) 15. What is your per-pupil pricing at 50 pupils, 250 pupils, 1,000 pupils, and 5,000 pupils? (Most vendors go non-linear; you want to see the curve.) 16. What is the contract term and exit clause if pupil outcomes aren't met at the 6-month milestone? 17. If we stop using you, what is the export format for pupil data, and how long is data retained post-termination?

If a vendor deflects or hand-waves on any of these, don't sign. The cost of unwinding a 3-year AI contract with a non-compliant vendor is much higher than the cost of evaluating 2 more options.

---

5. ROI — What an AI Tutor Actually Returns vs. a Human Tutor

Headteachers regularly ask: "What ROI can I expect?" The honest answer depends on the cohort, the previous baseline, and whether you measure just attainment or attainment + workload + SEND outcomes.

5.1 The math, in 2026 prices

Intervention	Annual cost (per pupil, full year)	Typical measurable outcome	Cost-per-grade-boundary-moved
1-to-1 human tutor (UK, GCSE, 1 hr/wk × 30 wk)	£1,050–£1,650	+0.5 to +1.0 grade in target subject	£1,500–£3,300 per grade moved
Group revision course (UK, GCSE, 1 term)	£180–£420	+0.3 to +0.6 grade	£400–£1,200 per grade moved
Adaptive practice engine (Category 1, school-wide)	£2–£8/pupil/year	+0.2 to +0.4 grade in maths/facts	£30–£80 per grade moved
Full-platform AI tutor (Category 3, school-wide, e.g. Grademy)	£18–£45/pupil/year	+0.4 to +0.9 grade in target subject + workload reduction	£120–£400 per grade moved
Catch-up tutoring premium (DfE grant funded, in-school human)	£600–£1,200/pupil/year	+0.5 to +1.0 grade	£1,200–£2,400 per grade moved

(Note: this is comparing interventions with themselves, not the same pupil. A pupil getting Category 3 AI tutoring plus group revision will do better than either alone. Don't conflate.)

5.2 The hidden ROI — teacher workload

The 2025 DfE workload survey found teachers spent an average of 6.4 hours/week on marking + feedback + intervention planning. A Category 3 AI tutor returns roughly 2.5–3.5 of those hours per teacher per week by automating:

• Practice-question marking (with worked solutions).
• Misconception flagging ("32 of your 27 pupils can't convert fractions").
• Parent-report drafting.
• Stretch-task differentiation.

That's a teacher-time saving of £1,800–£2,600 per teacher per year (using average UK teacher cost £42k). For a 40-teacher secondary school, that's £72k–£104k/year in reclaimed time — substantially more than the AI subscription cost.

5.3 ROI for the SEND cohort

For EHCP-flagged pupils, the ROI is harder to express in grade-boundary terms but easier in attendance + safeguarding. Schools using a Category 3 platform with calm-mode + reading-age tracking report:

• Attendance uplift: +3 to +6 percentage points across the SEND cohort.
• Reduction in fixed-term exclusions: 15–25% over the academic year.
• Parental-tribunal risk reduction (clearer EHCP-progress evidence).

These are not the primary reasons most schools buy AI tutors. They often end up being the primary reasons schools keep them.

---

6. Pilot Rollout — A 90-Day Plan That Actually Sticks

Most schools spend 4 weeks choosing a vendor, 2 weeks onboarding, 4 weeks waiting for results, then panic at the half-term holiday. Here is a tighter plan.

Days 0–14 — Vendor selection

• Run the 17-question filter on 3 vendors (one Category 1, one Category 2, one Category 3 — see §1).
• Score on compliance (§3), pedagogy (§4.2–4.4), teacher workflow (§4.5), commercial (§4.6).
• Engage your DPO before vendor demos, not after.

Days 15–30 — Compliance + DPIA

• Sign the DPA. Get the sub-processor list.
• Run a UK GDPR DPIA. DfE has a free template.
• Brief your governors / trustees — most UK MAT boards need a recommendation paper.

Days 31–60 — Pilot (1 year group, 2 subjects)

• Start small: one year group, two subjects, one SEND cohort if relevant.
• Set the kill criterion in writing at the start (e.g. "if pupils average <20 minutes/week of platform use by week 6, we exit").
• Weekly check-in with the class teacher; bi-weekly with your DSL.
• Mid-pilot exit check at day 45 — not day 90.

Days 61–90 — Evaluate + decide

• Pull per-pupil outcome data (§4.6).
• Compare to baseline (mock results, reading age, attendance).
• Decide: roll out wider / extend pilot / switch vendor / end.
• Always have an exit clause in the contract that lets you leave at the end of the pilot without penalty.

The single biggest reason UK school AI pilots fail is scope creep at launch — rolling out trust-wide from day 1. Don't.

---

7. Where Grademy Sits in This Landscape

We built Grademy. We're a Category 3 platform. We're obviously biased toward our own product. We're going to be specific about where we fit, where we don't, and how to verify.

Where Grademy fits:

• UK secondary (KS3–KS5) + primary (KS2, with reading-age focus).
• UK GDPR + DPA 2018 native — UK-hosted inference, full sub-processor list, signed school DPA in days not weeks.
• Adaptive practice across 14 subjects (Maths, English, Science, MFL, Humanities) + open-ended LLM tutoring with a 7-mode suite (Socratic, Teach-Back, Errors & Gaps, Feynman, Drill, Worked Example, Stretch).
• Per-pupil mastery tracking + per-class teacher dashboard + per-trust MAT rollup.
• Calm mode for SEND (dyslexia overlay, ADHD-friendly timing, ASC low-stim layout).
• Per-pupil pricing between £18–£28/pupil/year at school-wide rollout (volume tiered).

Where Grademy is the wrong choice:

• You need a Category 1 adaptive practice engine for primary KS1 fact-drilling only — go with a cheaper specialist.
• You need a US-headquartered vendor with deep US-state-law experience and FERPA-first engineering — we can serve US schools, but a US-native vendor will have more US-specific content.
• You need only a human-tutor marketplace and don't want AI in the loop — we don't sell that.
• You need 24/7 live tutor chat with a human on the other end — we don't do that; we're AI-only.

How to verify:

• Ask for a UK GDPR DPIA — we ship one with the school onboarding pack.
• Ask for a 30-pupil, 30-day free pilot — available to any UK state school with a signed DPA.
• Ask for the safeguarding log export — we provide a one-click export to your DSL.
• Ask for per-pupil outcome data from 3 named UK schools — we can introduce you.

If after the above we're not a fit, we'll help you pick a vendor that is. There are 6–8 credible UK options in 2026 depending on your cohort framing (§2).

---

8. Common Procurement Mistakes — And How to Dodge Them

Mistake 1 — Lead with price, not with cohort fit

The cheapest vendor per seat is almost never the most cost-effective per outcome. Lead with the cohort framing (§2), then price.

Mistake 2 — Sign a 3-year contract before the pilot

Always sign for 12 months initially, with a 30-day exit clause. Trust-wide rollouts are for year 2, after you've survived year 1.

Mistake 3 — Skip the DPO

Your Data Protection Officer is the single most important reviewer in this process. If they're not in the vendor demo, the DPA gets signed without their context. Don't.

Mistake 4 — Forget the parents

Parents are the ultimate stakeholder. Send a one-page parent letter at pilot launch explaining: what the platform is, what data it processes, who to contact with questions, and how to opt out. UK parental opt-out rights on third-party AI processing are real.

Mistake 5 — Measure only attainment

Measure workload, engagement, SEND-specific outcomes, attendance, and attainment. If your whole evaluation rubric is "did GCSE results go up", you'll miss the platforms that genuinely transformed the SEND cohort but didn't move the school-wide average.

Mistake 6 — Treat AI as a replacement for high-quality teaching

It isn't. It's a workload reducer, a personalisation multiplier, and a catch-up accelerator. A school with weak teaching + great AI still has weak teaching. A school with strong teaching + great AI gets 18% more out of every hour.

---

9. The 12-Month Procurement Timeline (UK Calendar)

If you're considering an AI tutor for the next academic year (Sept 2027), here's the procurement timeline we advise:

Month	Action
September 2026	Brief SLT + governors. Approve budget line. Identify cohort + kill criterion (§2).
October 2026	Issue vendor RFI (use the 17 questions in §4 as the RFI).
November 2026	Vendor demos (3 vendors, single-day format). Shortlist 2.
December 2026	Compliance review with your DPO. DPIA drafted. Reference calls from shortlisted vendors' existing UK schools.
January 2027	Pilot contract signed (12 months, 30-day exit).
February 2027	Onboarding: rosters, SSO, teacher training.
March–May 2027	Pilot running (1 year group, 2 subjects, 1 SEND cohort).
June 2027	Mid-pilot exit check.
July 2027	Pilot wrap-up. Final outcome report.
August 2027	Decision: roll out, switch, or end.
September 2027	Trust-wide (or school-wide) rollout if appropriate.

A 12-month cycle is slow on the calendar but it's the cadence that produces outcomes that survive a headteacher change.

---

Conclusion — Buying an AI Tutor Is Buying a 3-Year Workflow Change

If you remember nothing else from this guide:

1. Categorise vendors before you score them. Category 1 ≠ Category 3.
2. UK GDPR + DPA 2018 are non-negotiable. Don't sign without a DPIA.
3. Run a 17-question filter. It eliminates most weak platforms.
4. Pilot to one year group + two subjects + one SEND cohort. Scope creep kills pilots.
5. Measure workload + engagement + SEND outcomes as well as attainment. Otherwise you'll miss the platforms that actually helped your hardest-to-reach pupils.
6. Have an exit clause. Always.

If you want a head start, Grademy offers a 30-pupil, 30-day free pilot to any UK state school that signs our DPA. We also offer to run the 17-question filter with your DPO on 3 of your shortlisted vendors — free of charge, vendor-neutral — so you can choose the platform that's right for your cohort, not ours.

The 2026 UK AI-tutor market is crowded, expensive, easy to get wrong, and genuinely transformative when you get it right. Pick the framing, then the compliance, then the curriculum fit, then the price. In that order.

---

Last reviewed: July 2026. UK GDPR + DPA 2018 references are correct as of the date of publication; AI tutor market list reflects vendor categories in mid-2026 and will drift by end of year. If you're reading this in 2027, contact us for an updated vendor taxonomy.

More from Grademy: For the UK schools procurement and buying-guide angle on tier-aware practice, our GCSE foundation vs higher tier AI tutor 2026 guide covers grade-boundary math at 4/5/7, the three signals that should agree before a tier is set, per-tier question generation, and the Year 11 timeline that keeps the decision reviewable rather than fixed.