Validate a business idea in a week, with money not opinions

An idea is validated when someone who has no personal reason to be kind to you has paid for it, at the price you mean to charge, before you have built the thing. Not signed up. Not said it was a great idea. Paid. Every other test on this page is a proxy for that one, useful in proportion to how close it gets to a card being charged.

The reason to be this strict is that the alternative is the most common way a business dies. CB Insights went through 431 companies that shut down from 2023 onwards and could assign causes to 385 of them: 70% ran out of capital, and 43% failed on product-market fit — of which about two-thirds never found a market at all. Running out of money is what it feels like from the inside. Not having a market is usually what actually happened, eighteen months earlier, in a fortnight nobody spent asking.

The asymmetry is the whole argument. A week of validation costs you a week. Skipping it costs you the year you spend building for people who were being polite. And the cost of finding out has collapsed — a page that takes a real payment is an afternoon now, not a fortnight — so the only remaining reason not to run these tests is that they can say no, which is exactly what you want from them.

This page is the fortnight where money changes hands or doesn't. The order everything else goes in — when to form a company, when to build, when to spend on ads — is on the ninety-day sequence, and this fortnight is its second gate.

These aren't weights from a regression — nobody has run a controlled trial on business ideas, and anyone who claims otherwise is selling a course. They're the four things that are cheap to check now and expensive to be wrong about later, chosen because each one, when it's missing, kills a business in a different and predictable way.

The scorecard below marks each one out of two — missing, half there, solid — which gives a total out of eight. Two points rather than ten because the honest resolution of the evidence you have in week one is three states, and a scale of ten invites you to award yourself a 7 for a feeling.

The fourth signal isn't a judgement at all, it's arithmetic, so the scorecard asks for three numbers instead of an opinion. Ten customers a month or fewer is solid: one person can find ten and serve ten. Up to thirty is half there — real, but it needs a system and most of your week. Past thirty and the price is wrong, because at that point you are betting the business on a volume of strangers you have no route to yet. That threshold moves the answer more than any amount of enthusiasm does, which is why it's the one input founders should be most honest about — including their own pay in what a month costs.

One more rule the scorecard follows: whichever signal scores lowest decides your next test, and ties break towards the earlier signal, because the earlier ones are the harder ones to fix. A strong idea with one unknown leg is a single experiment away from an answer, and running the other three tests will not give it to you.

In order of cost and in order of how much they prove. Run the one your weakest signal points at, not all three — three half-run tests in a week produce three anecdotes, and you'll believe the friendliest one.

Almost nobody fails these tests and then says so. What happens instead is a founder gets an ambiguous result, tells themselves a story that is true enough to survive a week, and builds anyway. Here are the stories, next to what the result actually was. Read this before you run the tests, not after — the point is to have named the trap while you still have nothing invested in walking into it.

There's a fail worth arguing with, though, and it's the mirror image: three ways a decent idea flunks a test that was never run properly. The checkout was broken and nobody told you. You asked five people who have the problem mildly rather than five who have it badly, and mild problems don't have prices. Or you asked about the future — “would you use this?” — which measures how much someone likes you, while “what did you do last time?” measures what happened. If any of those three describes your week, you don't have a result yet. Run it again properly; it costs two days.

Northline Bike Repair is the example business that runs through these pages: one founder, a van, mobile bike repair in Denver. On day 8 he had a plausible idea and no evidence. On day 14 he had three deposits. The figures below are this example's own — not benchmarks — and what's worth copying is the shape: one weak leg, one test, one answer.

Two things in that table are the actual lesson. The first is that $60 of deposits from three people outweighed a week of research, because it converted an opinion into a fact — and that nine of twelve saying “sounds useful” would have been read by most founders as a nine-out-of-twelve pass rather than the three-out-of-twelve it was.

The second is that even after passing, his weakest signal was the maths — 27 jobs a month is six or seven a week, which one person can do but which leaves no slack for a rainy fortnight or a broken van. By week six the site was live and taking cards, $1,286 had been collected that month, $300 of Facebook ads had brought back $1,486 counted from Stripe charges rather than from Meta, and the Money page showed 8.4 months of runway. Then, in week eleven, he ran the test the scorecard names for a weak fourth leg and priced it upwards: the tune-up went from $60 to $66 on new bookings, nobody objected, and 27 jobs a month became 24. None of that would have been worth building on day 8, and none of it needed to be.

Three deposits is a licence to build the smallest version, not proof of a business. Be clear about what you still don't know, because the gap between those two is where the second wave of expensive mistakes lives.

It doesn't tell you what a customer costs. Your first three came from a Saturday club, a neighbour, or a conversation you were already in. Customer eleven has to come from somewhere repeatable, and the price of that is a separate experiment with its own numbers — that's the first-ten-customers problem, and it's the one that decides whether the margin you just validated survives.

It doesn't tell you whether you can deliver ten at once. One job in a driveway is not ten jobs on a Saturday. The tests measure demand; capacity is your problem, and it's the reason the fourth signal asks whether the monthly customer count is one you could plausibly serve yourself.

It doesn't tell you the price holds. Early buyers are unusually forgiving and unusually motivated. The price is validated at customer twenty, not customer three, which is why the sensible move at ten is to put it up on new work and watch.

And the harder question: when do you stop? A reasonable ceiling is three cycles — price, then offer, then customer, a fortnight each, six weeks in total. If nobody has paid after all three, you have learned something in six weeks that most founders take two years and their savings to learn, and almost nothing is wasted on the next attempt: the page, the checklist, the list of names and the conversations all carry over. That is a much better outcome than a fourth cycle run on stubbornness.