10 UK EdTech Companies to Watch in 2026: AI Tutors, Assessment and Independent Learning

Edtech companies

Oct 17

Imagine a Year 10 pupil revising GCSE science after dinner. An AI tutor is available immediately, a quiz reveals a misconception the pupil did not know they had, and a teacher can view the pattern in class the following morning. Meanwhile, another school is trialling AI-generated planning resources; a training provider is considering automated marking; and a sixth-form student is exploring careers through a virtual work-experience project. These are illustrative scenarios, not reports about specific customers.

The United Kingdom is a particularly revealing place to ask what such technology actually changes. British EdTech now stretches from exam preparation to university access, teacher workload, skills assessment and workplace development. In April 2026, Beauhurst reported 1,561 active UK EdTech companies in its dataset. It also recorded £131.6 million raised in the sector during 2025, far below the pandemic-era peak of £697.8 million in 2021. These are figures from Beauhurst's defined company database, not a measure of the total value of every British education product.

A tougher investment climate has not eliminated experimentation. It has made the questions more demanding: Does a product improve the work of teachers? Does it make a learner more capable without the screen? And what evidence supports the promise?

The UK's Department for Education opened a programme in April 2026 inviting technology companies to develop and test safe AI tutoring tools under teacher supervision, especially for disadvantaged pupils. Separately, the EdTech Evidence Board pilot findings, published in September 2026, concluded that independent review of educational technology evidence is feasible and valued, while stronger evidence and clearer communication are still needed.

This article selects ten UK-based or clearly UK-rooted companies and platforms to watch in 2026, from young entrants to established but evolving businesses. It is not a league table by investment, market share, academic performance or future growth. The order reflects different educational problems, not the relative quality of the companies. Where a business provides its own performance numbers, those are identified as claims rather than automatically treated as independent proof.

At a glance: ten British EdTech companies and what they do

Company	Main educational problem	Primary users	Key question
Medly AI	Responsive exam tutoring	GCSE and other exam students	Does improvement persist independently?
Eedi	Hidden maths misconceptions	Pupils and teachers	Can diagnosis become lasting understanding?
Tugo	Affordable, supported small-group tuition	Families and tutors	Is human feedback strengthened by AI?
Chalkie	Teacher planning and resources	Teachers and school leaders	Does saved time produce better teaching?
sAInaptic	Slow assessment and marking	Assessment and training providers	Are automated judgements dependable?
Carousel Learning	Knowledge that is quickly forgotten	Schools and pupils	Can retrieval become self-directed?
Gizmo	Turning revision notes into active practice	Students and independent learners	Do gamified habits deepen recall?
Springpod	Unequal career-exploration opportunities	Young people and employers	Do experiences lead to informed decisions?
Atom Learning	Personalised practice and exam preparation	Children, parents and schools	Is adaptation equitable and meaningful?
HowNow	Workplace learning detached from skills needs	Employers and adult learners	Do completed courses change performance?

1. Medly AI: a personalised tutor for exam preparation

The problem. Many pupils need a second explanation, targeted practice or immediate feedback outside a teacher's available hours. Private tuition can be costly, and an ordinary search engine rarely knows precisely where a learner is stuck.

Medly AI is an AI-powered tutoring platform founded by University College London alumni Kavi Samra and Paul Jung. In a September 2026 university announcement, UCL reported that the company had secured $8 million in fresh investment and that more than 400,000 students had used the platform for UK exam preparation. These are a named university's reported company milestones, not independently audited active-user figures.

Medly's educational proposition is to adjust explanations and feedback to the student's needs. UCL also described a randomised trial in which participating GCSE science pupils using Medly achieved scores around 22% higher than a comparison group. That makes the evaluation worth examining, but a press summary alone is not enough to establish general effectiveness: readers should ask for the full study, the outcome measure, the intervention duration and whether effects persist after access ends.

What deserves attention: AI tutoring is more educationally promising when it elicits a first attempt, offers a hint before a solution and asks a student to reconstruct the answer later. A strong classroom pilot would compare supported performance with a delayed test completed without AI. The success metric should not simply be conversations generated or minutes spent.

2. Eedi: diagnosing the maths misconception behind the mistake

The problem. A wrong answer does not tell a teacher why a pupil is wrong. Two pupils can select the same incorrect option for completely different reasons.

Eedi uses carefully designed diagnostic questions to identify mathematical misconceptions and suggest follow-up practice. Its teacher documentation describes how teachers can inspect answer patterns before and after assignments rather than treating all errors as equal.

There is also a substantive UK evaluation to consider. In November 2025, Eedi reported findings from an independent randomised controlled trial involving 2,901 Year 7 pupils in 20 schools during the 2023–24 academic year. The company described gains equivalent to roughly two months of additional maths progress compared with usual alternatives; the claim of four months concerned a higher-engagement subgroup and should not be read as a guaranteed result for every pupil. Evaluators should examine the underlying paper and its statistical assumptions.

Eedi has also discussed human-supervised AI tutoring experiments. That matters because diagnosing an error and helping a child understand it are different tasks. A plausible coaching question is: What led you to choose this answer, and what would change your mind?

The limit: Even a well-designed diagnostic platform should not turn the learner into a passive receiver of remedial exercises. Teachers need to check whether misconceptions disappear in unfamiliar, non-digital problems.

3. Tugo: putting an AI co-pilot beside a real tutor

The problem. Individual human tuition is personalised but expensive to scale. Group tuition is more accessible, yet it can be hard for one tutor to monitor every student's work at once.

Tugo offers live small-group tutoring with a human teacher assisted by software for feedback and lesson analytics. In an August 2026 company update, Tugo described a December 2025 launch, early school partnerships and more than 500 student-hours of tuition. Its inclusion on the 2026 UK Business Tech Awards EdTech shortlist is a genuine visibility signal; it does not establish educational effectiveness.

Tugo is interesting precisely because it resists the familiar binary of teacher versus chatbot. Its stated model gives the tutor responsibility for relationships, motivation and intervention while AI handles some work-checking. The company's September 2026 terms say that some AI feedback may reach a pupil before a tutor reviews it, an important detail that families and schools should understand.

The best study of Tugo would measure not only GCSE grades but the distribution of attention within a group: Which pupils receive meaningful teacher interaction? Does automatic feedback correct misunderstandings or occasionally reinforce them? Does the technology make good tutors more effective, or merely increase the size of the group?

What to watch: This is an early-stage company; the reported operational milestones are company statements. Strong independent attainment evidence remains a separate question.

4. Chalkie: reducing the teacher's preparation burden

The problem. Planning a sequence of lessons, preparing differentiated activities and adapting resources can consume time that a teacher might otherwise spend on feedback and individual support.

Chalkie develops AI tools for lesson creation, worksheets, activities and rubrics. Its schools product page describes curriculum alignment, export formats and resources for teachers and school leaders. UK Companies House records show Chalkie AI Ltd was incorporated in July 2024.

This is a powerful workflow proposition, but efficiency is not the same as educational quality. A generated worksheet may look polished while containing weak sequencing, inaccurate examples or tasks that do not reveal student thinking. Teachers still need subject expertise to recognise those flaws.

For a UK school, the strongest pilot would let teachers compare three lesson sequences: one planned conventionally, one prepared with AI and minimal review, and one prepared with AI but critically revised. Time saved is a useful measure. Pupil understanding, teacher confidence and accessibility are more meaningful companion measures.

The autonomy test: Does Chalkie help teachers spend more time listening to how pupils reason? Or do faster resources simply produce more tasks without deeper conversations?

5. sAInaptic: faster marking with a serious question about trust

The problem. Open-ended answers, vocational reports and professional assessments are expensive and slow to mark well. Learners may wait for feedback long after the reasoning process has faded from memory.

British assessment start-up sAInaptic uses AI to assist marking and feedback in education and training. Cornerstone VC reported a £1 million funding round on 13 October 2026, backed by investors including Ufi Ventures, as well as a partnership with assessment-software company Skilltech Solutions. Education Week's Market Brief independently covered the funding announcement.

The investor statement includes striking claims about reductions in marking time and agreement with expert assessors. Those are investor- or company-reported performance figures. Without public details of the sample, qualification types, scoring rubric and disagreement cases, they should not be generalised to every examination.

This company illustrates the difference between assessment speed and assessment validity. A system can mark quickly and still miss an unconventional but correct argument. A sound adoption process must include human moderation, appeals, checks for group differences and clear rules about where automation is advisory rather than decisive.

Educational promise: Faster, reliable formative feedback could create more room for reflection. But the moment a high-stakes judgement is involved, transparency matters as much as efficiency.

6. Carousel Learning: retrieval practice made systematic

The problem. Pupils often recognise material in a lesson yet cannot retrieve it weeks later. Teachers need an efficient way to revisit important knowledge without repeatedly rebuilding every quiz.

Carousel Learning specialises in retrieval practice, question banks and systems connecting homework with classroom teaching. Its July 2026 teacher guide explains how schools can integrate regular recall into existing routines. In April 2026 it described Study Packs designed to connect classroom retrieval and independent revision.

The underlying principle is plausible: remembering information requires active retrieval, not just rereading. But evidence for the broad retrieval practice approach should not be confused with proof that a specific subscription or feature produces a precise grade increase.

A teacher might use Carousel to test whether a class can recall vocabulary one week, one month and a term after it was first introduced. The more interesting follow-up is to ask pupils to design their own recall prompts, identify weak topics and choose the intervals at which they revise.

The autonomy test: When the platform is removed, can students still organise retrieval practice by themselves? If not, the habit belongs to the timetable or software rather than the learner.

7. Gizmo: transforming notes into quizzes that students will actually use

The problem. Students collect notes, screenshots, videos and PDFs but often mistake organising material for knowing it.

UK-rooted Gizmo turns imported material into flashcards and quizzes. Its official help centre describes importing notes, PDFs and video links; spaced-repetition study; an AI tutor; multiplayer quizzes; and gamified features. UK Companies House lists the active London company as Gizmo AI Ltd, formerly Save All Ltd. This learning app should not be confused with an unrelated mini-game product also called Gizmo.

Its strength is convenience: the distance between a messy set of notes and the first practice question becomes much shorter. But generated flashcards can reproduce mistakes in a source, prioritise trivial details or make factual recall seem like complete understanding.

A useful school trial would ask learners to correct inaccurate cards, explain answers orally and tackle an unfamiliar application question at the end of a revision cycle. These activities protect against the temptation to optimise only for streaks, rewards or rapid multiple-choice performance.

What to watch: Gamification may increase practice frequency, but streak length is not an achievement measure. The question is whether knowledge transfers into written work, problem solving and independent explanation.

8. Springpod: career learning beyond the school gate

The problem. Work experience and career exploration often depend on geography, family contacts and the willingness of local employers to host young people.

Springpod provides digital career exploration and virtual work-experience opportunities. Its 2026 partner resources describe programmes connecting employers, educators and young people, while its opportunities site lists ways learners can explore industries and next steps.

This is EdTech in a wider sense than AI tutoring. The product attempts to make professional environments legible to students who might otherwise have limited access. Springpod also reports reach and social-value figures on its own partner website; these should be examined alongside the underlying methodology rather than repeated as if they were verified academic outcomes.

The educational opportunity lies in reflection after exposure. Completing a virtual simulation is not the same as choosing a career with confidence. A learning coach or school careers adviser might ask: What surprised you? What skill did you use? What else do you need to investigate? Which real person could answer your remaining questions?

The limit: Virtual experience can widen access, but it cannot fully replace workplace relationships, actual responsibilities or hands-on experience. It is most powerful as a gateway to informed choices, not as a complete substitute for real placements.

9. Atom Learning: adaptive preparation and the fairness challenge

The problem. Pupils preparing for selective entrance examinations or core subjects do not all need the same questions. Families want to see progress without guessing what should come next.

Atom Learning provides adaptive practice and exam-preparation services. In an April 2026 explainer, the company described a system that adjusts question difficulty according to learner responses. Its Atom Access initiative says that qualifying pupil-premium children can receive free access through school partnerships.

Atom is not a newly launched business, and its place here reflects ongoing relevance and evolution rather than startup age. The most distinctive question is how to combine adaptive practice with equity. Personalised difficulty may help a pupil remain appropriately challenged, but it does not guarantee that the content addresses deep reasoning or that every family can afford the broader offering.

Company-stated acceptance rates and success stories are not causal evidence that Atom produced those school offers. Family background, prior attainment and selection effects matter.

What to watch: A transparent adaptive platform should help children notice which strategy worked, explain their errors and attempt unfamiliar problems without a hint. It should also make the difference between a paid product and an inclusion programme easy to understand.

10. HowNow: helping adults turn training into useful skills

The problem. Employers can buy large content libraries while employees struggle to connect courses with the skills their jobs actually demand.

London-based HowNow brings workplace learning, skills mapping and learning recommendations into one platform. Its current proposition links training with performance needs rather than merely tracking completed courses. In the 2026 UK Business Tech Awards, HowNow won EdTech Company of the Year, with judges highlighting a business case involving skill development and operational outcomes.

This is another established, growing EdTech, not an overnight newcomer. Including it makes the ten-company comparison less narrowly focused on schools. Learning is a lifelong process, and the ability to apply a skill at work is an essential educational outcome.

A useful test is whether someone who completed a module can now perform a real task, teach the procedure to a colleague or solve a novel problem. Completion rates, badges and recommendation clicks are intermediate signals rather than final evidence.

The autonomy test: Does workplace technology help employees diagnose their own learning needs, find people who can support them and apply new knowledge independently? That is a stronger objective than simply expanding the training catalogue.

What the UK EdTech market reveals in 2026

The ten companies approach at least five different bottlenecks: access to feedback, diagnostic information, teacher time, durable memory and transitions into work. Comparing them as though they offered the same educational service would be misleading.

One major policy change is particularly significant. The Department for Education's January 2026 generative-AI product safety standards ask developers to minimise cognitive offloading and to avoid giving full answers by default, instead favouring learner attempts, progressive hints and appropriate human oversight. These are England-focused safety standards, not proof that any of the ten named companies has been certified against them.

This changes the quality question. Previously, the marketing centrepiece might have been a faster answer. Now educators can reasonably ask whether the system makes it harder for a pupil to avoid thinking. They can look for age-appropriate safeguards, clear logging, data governance, teacher controls and evidence beyond usage statistics.

The September 2026 EdTech Evidence Board pilot points in the same direction: evidence should be appraised openly rather than inferred from the reputation or investment history of a supplier. In practice, this means distinguishing an independent randomised trial from an internal case study, an investor presentation, a school testimonial and a product feature page.

How a teacher or school leader should evaluate these tools

Consider a hypothetical Year 8 maths department with limited budget and twelve weeks to test a product. Instead of asking staff which platform looks most impressive, the department might define a small, observable instructional problem: pupils recognise a method when shown it but cannot select it independently on a new problem.

A useful evaluation could follow six questions:

Define the learning task. What exactly should pupils be able to do without help?
Record a baseline. Include a task completed without the new technology.
Inspect the teaching mechanism. Does the tool elicit an attempt, reveal misconceptions, support practice or merely provide polished explanations?
Protect teacher judgement and learners. Review safeguarding, accessibility, privacy, bias, age suitability and contracts before uploading pupil work.
Test transfer. Repeat a comparable task later, in a different format and without digital assistance.
Review real cost. Include training, staff moderation, subscriptions, integration and support, not just a headline price.

This is an evaluation framework, not a claim that we tested these ten products. A school might choose no additional software at all if a change in feedback practice or homework design answers the problem more effectively.

For parents, the principle is similar. Ask your child to explain what a digital tutor helped them notice. Encourage them to solve a fresh question on paper. If their confidence and understanding disappear when the app closes, more screen time is not automatically the answer.

Personalisation is not the same as learner independence

Many EdTech products promise personalisation. They tailor difficulty, pace, explanation or recommended activities to data about a learner. That can be useful. Yet a personalised pathway can still leave the child dependent on a system that chooses every next step.

Learning Coach approaches the question from another angle: Can the learner progressively recognise their own goal, resources, mistakes and effective strategies? A documented Learning Coach lesson on coaching posture emphasises clear goals, awareness, the learner's own resources and autonomy rather than the helper doing the work in their place.

Applied to an EdTech pilot, those principles become practical checks. Students should be able to name what they are working towards, justify a strategy, identify when assistance is necessary and gradually work without it. A teacher operating as a coach observes and asks questions; technology can provide information, but it cannot replace the professional relationship or the pupil's responsibility for thinking.

This is why our comparison does not announce an overall winner. Medly and Eedi tackle different kinds of support. Chalkie may help teachers prepare; Springpod expands career exploration; HowNow speaks to adult capability. The appropriate tool is the one that addresses a defined need while protecting what learners must ultimately do for themselves.

For those developing this practice in schools, the Learning Coach training and specialities provide a starting point for exploring the teacher-as-coach approach. Readers interested in the broader philosophy of learning can also browse the English Learning Coach book collection. Neither resource should be treated as an independent evaluation of the companies profiled here.

Frequently asked questions

Which are the fastest-growing UK EdTech startups in 2026?

The publicly accessible evidence used here does not support a reliable ordering of all UK EdTech firms by year-on-year growth. This article is an editorial watchlist. Medly's September investment, Tugo's early rollout and sAInaptic's October funding are identifiable developments, but they measure different things and cannot be used to construct a sound growth ranking.

Is an AI tutor better than a human teacher?

That question has no single answer. An AI system may deliver more immediate practice or feedback, while a teacher contributes judgement, group understanding, safeguarding and a human relationship. Product-specific trials, subject matter, age, implementation and independent transfer tests are needed before making comparisons.

Are UK EdTech products regulated for safety?

Education providers must consider safeguarding and data-protection obligations, among other applicable requirements. The Department for Education publishes product safety standards for generative AI intended to inform developers, suppliers and institutions in England. Listing a company here does not certify its compliance.

How should a school choose between an EdTech company and a traditional intervention?

Begin with the learning problem, not the supplier. Compare credible evidence, teacher effort, inclusion, cost and whether students can repeat the target skill independently. Sometimes better-designed human feedback or retrieval practice is the more suitable investment.

The real frontier is not smarter software. It is stronger learners.

British EdTech in 2026 is diverse enough to resist simplistic predictions. Some companies are using AI to give pupils a chance to practise at home. Others are helping teachers notice misconceptions, making marking faster or connecting young people to careers they had never imagined.

The important achievement will not be the number of tasks a system completes. It will be the number of learners who can eventually complete meaningful tasks without needing the system to think for them.

That is the test educators, families, investors and founders should keep in view: when the technology disappears, what new ability remains?

Sources

UK EdTech companies to watch 2026

Faysal Hafidi