We built the marking assistant we wanted to be able to trust
Marky marks GCSE, AS and A-level papers in maths, biology, chemistry and physics against the official mark scheme, and returns the marks with the reasoning attached. A class set takes about five hours to mark by hand. On the Instant lane most classes come back in 20 to 30 minutes; on the Standard lane, typically 60 minutes.
That is the core of the product. This page is about the things a marketing page usually leaves out: what we actually measure, where the software is still wrong, and who is accountable when it is.
Our mission
To make fast, high-quality feedback something every department can afford to give on every piece of work, rather than a luxury rationed to the sets there is time to mark.
Give teachers their time back
Marking is the single biggest call on a teacher's evenings and weekends. Five hours of a class set should not be the price of knowing how a cohort is doing.
Close the gap between sitting and knowing
Feedback a fortnight late lands as a grade. Feedback the same week lands as something a student can still act on, while they remember what they were thinking.
Earn the trust rather than assume it
Software that touches student work has to show its reasoning and publish its error rate. We would rather be checkable than impressive.
Why we started
Marking is the part of teaching that scales worst. A department that wants to run frequent mocks has to weigh every round against the weekends it will cost, and the honest answer is usually to run fewer of them. So the assessment that would tell a teacher the most gets rationed, and it gets rationed hardest in the term when it would matter most.
The second cost is slower to show up. By the time a set comes back a fortnight later, the class has moved on, the student has forgotten what they were thinking, and the feedback lands as a grade rather than as something to act on. Turnaround is not a convenience feature. It is most of whether the feedback works at all.
We built Marky so a department can mark a full cohort in an evening's gap rather than an evening, and run the extra mock rather than talking itself out of it.
Who we are
Marky is a British software company based in West Sussex. We make marking software for secondary school departments, and the revision tools that sit alongside it for their students. It is not a side product of something else, and there is no other business it has to fit around.
The same people design the marking pipeline, measure its accuracy and answer the support email. When a head of department writes to say a mark looks wrong on question 6(c), it reaches someone who can open the script, see what the model read, and change how it reads it. That loop is the reason the product improves, and we have built the company around keeping it short.
We build for UK secondary schools specifically: GCSE, AS and A-level, marked against the official mark schemes of AQA, Pearson Edexcel, OCR and WJEC/Eduqas. The grade boundaries, the mark scheme conventions and the language in the feedback are the ones your department already uses.
Schools are the customer
Schools pay for marked papers and, if they choose, student accounts; families will be able to pay for the individual revision plan when it opens. Student data is never a revenue stream.
Built and run in the UK
A UK company working to UK GDPR, for UK qualifications and UK exam boards.
Narrow on purpose
Four subjects, three levels, one job done properly. We say no to a great deal in order to keep saying yes to that.
How we think about the problem
Four commitments shape most of the decisions we make, including the ones that cost us features.
A number without its denominator is not evidence
Every accuracy figure we publish carries the sample it came from and the method that produced it. Ours is 1,830 question sections on 60 real GCSE maths scripts, compared against the marks the class teacher had already written, with nobody adjudicating between the teacher and us. We publish where the software agrees exactly, within one mark and within two, because the tolerance you choose changes the headline. We would rather quote a lower number we can defend than a higher one we cannot.
The teacher holds the pen
Marky proposes a mark and explains it against the mark scheme. A teacher can change any of it, and the marks that go to students are the ones the teacher approves. Work the checks pick out, such as a total close to a grade boundary or a section the system could not mark, is flagged for review rather than passed through silently, so attention lands where it is worth spending. Nothing here is designed to mark a cohort while nobody looks.
Student work is the school's, not ours
We are the data processor for student data and act under the school's instructions. Scripts are processed to mark them and, as the privacy policy sets out, to measure marking accuracy; they are never used to train AI models. Schools can export everything through the subject access request tooling, and our privacy policy sets out the lawful bases and retention in full rather than in summary.
One price, printed on the page
A marked paper is £1.20. There are no per-teacher seats, no feature tiers and no monthly caps, and a department buys a year's allowance sized to its own mocks. Instant marking is faster because the whole set is marked at once, and each script draws two papers from that same allowance instead of one, which is why it costs £2.40 a script. It is not a plan, and it does not change the price of a paper.
What we can show you
These are the figures behind the product, each drawn from the same comparison: 1,830 question sections on 60 real GCSE maths scripts, marked in red pen by the teacher who taught the class before Marky saw them, and then by Marky. We also measured an AS maths cohort, but that teacher's marks were collected afterwards, with Marky's mark in view, so it is reported separately on the accuracy page rather than in these figures.
The accuracy page gives the method, the per-cohort splits, the direction of every disagreement and the limits of what this evidence supports. We would rather you read it than take the headline.
What we will not claim
A tool used on student work should be honest about its edges. These are ours.
It is not an examiner, and it does not replace one. Marky is a first pass that a teacher checks. It is not moderation, it carries no exam board endorsement, and no mark it proposes is a prediction of an awarding body outcome.
We do not mark our own homework. The reference we publish against is the mark a real class teacher gave on their own class set, with one exception the accuracy page describes, and with nobody adjudicating between the teacher and the software. Where we do read a disagreement against the mark scheme ourselves (the reading is done by AI models working for us, not by an examiner), that is labelled as our reading in the same sentence as the number, and the unadjudicated figure is published beside it. The accuracy page gives the method, the denominators and the limits.
Most of what we get wrong is judgement, not handwriting. Measured in July 2026 on an earlier version of the marking model, our own reading of 108 disagreements on maths scripts, made by setting the software’s transcription of each answer beside the note in our reference set rather than by re-reading the script, puts 66% down to work that was read correctly with the mark scheme applied too strictly or too loosely and 30% down to the script being misread, with the rest unclear. That 66% is an upper bound, because a misread can dress itself as a method mark when the software is the one reporting what it saw, and the reading was done by AI agents, not by an independent human. The two fail in opposite directions. Judgement errors tend to be stingy, usually on method and process marks part way through a long answer. Misreads tend to be generous, because a tidier reading of a slip can earn credit the student did not. That is why every mark is shown with the working it was read from: it is the fastest way to tell which kind you are looking at.
Four subjects marked, two measured. We mark maths, biology, chemistry and physics at GCSE, AS and A-level. The accuracy figures above are maths, because maths is the only subject where a teacher-marked cohort exists to measure against. Physics is measured against our reading of the mark scheme, set out on the accuracy page. We publish no figure for biology or chemistry rather than extrapolate one, and we would rather say that here than let another subject’s number stand in for them.
Turnaround is typical, not guaranteed. Most classes come back in 20 to 30 minutes on Instant and in about 60 on the Standard lane. Busy periods are slower: when several Instant classes are sent at once they wait their turn, and the run's page shows where yours is in the queue. We would rather say so here than in a footnote after you have bought.
The most useful thing you can do with any of this is not believe it. Mark ten papers yourself, run the same ten through Marky and compare. That is the test we would apply, and it is the one we recommend to every department that asks.
Who you are dealing with
There is no call centre between you and the people who wrote the software: an email to hello@markyedu.com reaches them.
- General and sales enquiries
- hello@markyedu.com
- Data protection and accessibility
- hello@markyedu.com
Our privacy policy, terms of service and accessibility statement say what we are bound to, and the changelog records what has actually shipped.
Try it on papers you have already marked
The fastest way to judge any of this is against your own marking. The trial runs for 14 days and covers ten papers, and schools do not need a card to start.