There is no shortage of education ideas. The scarce thing is an idea that fits its market, means something to the learner, and has someone who can carry it. These are the six checks we run before we'd commit to co-building, written so you can run them yourself first. None of them is about how clever the technology is.
1. It's system-native
The first question is the one most imported tools fail: is it built for the system its learners actually live in? Their syllabus, their languages, their price point, their connection. We run the Four Mismatches Test on every venture, and a single failure is usually disqualifying for a core product. A brilliant maths tutor mapped to the wrong exam is teaching confidently in the wrong direction.
Good signal: it names the exam, the languages, and the price a local family pays, and it was designed around them. Bad signal: "standards-aligned" to someone else's standards, with localisation as a later phase.
2. It has a real wedge, not a feature
We look for one painful, specific job done completely, for one kind of learner, in one system — not a broad platform that does ten things adequately. A wedge is defensible because it earns trust by being the best thing for that one job; a feature is copied in a sprint.
The tell is in how a founder describes it. "It's a whole learning ecosystem" is usually a demo. "It gets a Form 4 student through SPM Add Maths, and nothing else, better than anything they can buy" is a company.
Good signal: you can say in one sentence who it's for and what it does, and the sentence is narrow. Bad signal: the pitch needs a diagram with five boxes.
3. It measures mastery, not engagement
This is a values screen as much as a product one, and we hold it firmly. A learning product should measure whether someone actually learned — cold recall, real before-and-after — not how long they stayed in the app. The two goals pull in opposite directions: the engagement-maximising version of an education product is often a worse education product.
A student who finishes their session in twenty-two minutes and logs off has won. If the metric on the wall is daily active minutes, the incentives are already wrong. We pass on products built to be sticky rather than effective, even when the sticky version would grow faster. (On why mastery is the right target, the underlying research goes back to Bloom's two-sigma work — what good one-to-one teaching actually does.)
Good signal: the product can tell you what a learner couldn't do before and can do now. Bad signal: the headline number is time-in-app or streaks.
4. The founder is the right shape for it
We co-build with two kinds of people, and both bring something we can't: subject-matter experts with real domain authority — teachers, curriculum people, child-development specialists — and entrepreneurial founders with operator drive and a market they understand. What we look for in either is someone who can carry the thing for years. Education is a long relationship; a child who is served well at eleven can be served at twenty-eight. We're wary of tourists — people for whom this is the idea of the month rather than the work of a decade.
Good signal: the founder has either deep domain credibility or a hard-won understanding of the buyer, and they talk about the learner more than the technology. Bad signal: the founder picked "education" because the market is big.
5. There's a real path to learners
A product with no route to the people it's for is a hobby. We look for distribution that already exists or a budget that already exists: a relationship with schools or tuition groups, a parent audience, or — for training — the demand-side levy employers already hold. Building the thing is half the work; getting it in front of learners is the other half, and it cannot be an afterthought.
Good signal: there's a concrete first hundred users you can name. Bad signal: "once it's built, we'll figure out distribution."
6. It uses AI honestly
AI has changed exactly one thing in this field for certain: it has collapsed the price of good teaching. That is enormous, and it is also specific. We look for ventures that use AI to remove a genuine constraint — usually the supply of one-to-one instruction — not ones that bolt "AI" onto a brochure. The word in the pitch deck means nothing; what the AI actually removes means everything.
Good signal: "AI lets us give every student something only the wealthy could afford." Bad signal: "AI-powered" with no sentence explaining what it makes possible that wasn't before.
What we pass on
The inverse of the six is a useful list in its own right. We tend to pass on: a tool built for a system it doesn't fit; a broad platform with no wedge; a product optimised for engagement over learning; a chatbot presented as the whole product rather than a feature; a founder who is a tourist; and anything whose pitch leads with the technology instead of the learner.
Run it on yourself
Before you talk to us, or to anyone, run the six on your own idea. Honestly. Most ideas fail two or three, and knowing which is most of the early work — it tells you what to fix before you build, not after. The ones that clear all six are rare, and they're the ones worth years of your life.
Frequently asked questions
Do I need all six to be fundable? No — early ideas rarely clear all six on day one. We look for ventures that clear fit and the values line, and have a credible plan for the rest. Failing two or three is normal; not knowing which is the problem.
Is this only for technical founders? No. We co-build specifically with non-technical founders and subject-matter experts — the build is what we bring. The six checks are about fit, wedge, values, founder, distribution, and honesty, not code.
What's the single most common reason you pass? No wedge — a broad "learning platform" that does many things adequately and nothing exceptionally. Depth is the whole game.
A note on why we wrote this. Addestra co-builds EdTech ventures from zero to one, and these are the criteria we'd hold a build to — ours included. Publishing them is the point: a founder should be able to judge their own idea against our standards before any conversation. How we work: editorial standards.
Sources
- The Four Mismatches Test and System-Native EdTech (our framework): /articles/the-four-mismatches-test, /articles/system-native-edtech
- Education Next — "Two-Sigma Tutoring: Separating Science Fiction from Science Fact" (on mastery and one-to-one teaching): https://www.educationnext.org/two-sigma-tutoring-separating-science-fiction-from-science-fact/