How to validate a startup idea before you build it

Start by writing the idea as a list of claims that could each turn out false. An idea like “an app that chases late invoices for freelance designers” quietly asserts that designers are paid late often, that today's workaround is not good enough, that enough of them exist, that they would pay, and that nothing blocks you. Each of those is a hypothesis.

A complete idea makes a claim on 6 lenses. If one of them has no claim written under it, that is the part of the idea nothing will test.

Then mark the critical ones: the claims the idea dies without. Those get tested first, because good news about anything else does not matter if one of them is false.

Before looking at any evidence, write down what result would make you stop. If you decide the bar after the results are in, you will set it wherever the results happen to land.

For each critical claim, state two things: what you would need to see to treat it as supported, and what you would need to see to treat it as false. “Most of the freelancers I interview already use a paid tool or a spreadsheet for this” is a bar. “People seem interested” is not.

A great deal of what you need has already been published, and reading it costs nothing. For each claim, look for:

• Who already sells an answer, what they charge, and who they sell to. An existing paid product is evidence the problem is real.
• What their users complain about in reviews and community threads. The gap in an existing product is often the idea.
• How people describe the problem in their own words, in forums and communities, and what they do about it today.
• Market data from a primary source: a statistics office or a research firm, not a blog quoting one.

Three habits separate research from reassurance:

1. Rank your sources. A government dataset outranks trade press, which outranks a vendor's blog. A company describing its own market has an interest in the answer.
2. Open the page. A link is not proof. Check that the page says what the claim says, including the number.
3. Keep what you could not find. “Nobody has published this” is a finding. It tells you what to test yourself.

Spend as much effort looking for reasons the idea fails as for reasons it works. Evidence you went looking for in order to feel better is worth less than evidence that survived an attempt to knock it down.

• Search for competitors until you find them. “Nobody does this” usually means the search was too short, or that people tried and stopped.
• Look for the predecessors that died, and for why.
• Check what you depend on. A platform, an API or a data source that can be withdrawn is a risk someone else controls.
• Check the rules. Law, regulation and platform policy block more ideas than competition does.
• Attack your best evidence. Take the strongest finding in your favour and search specifically for a reason it is wrong.

Published evidence rarely settles the claims that matter most: whether these particular people will pay this particular price for your particular answer. Those need a small real-world test with a result you committed to in advance.

Customer interviews test whether the problem is real. Landing pages, fake doors and ad tests measure demand. A concierge test or a prototype shows whether your answer works. Each is covered in Startup validation experiments: which test to run, and when.

One rule covers most interviewing mistakes: ask people what they did the last time the problem happened, not what they think of your idea. People are generous about ideas and accurate about their own past. Rob Fitzpatrick's The Mom Test is the standard handbook for these conversations.

Compare what you found with the bars you set in step 2, and choose one of three things: continue, pivot, or kill. Continue if the critical claims held. Pivot if the problem is real but your answer to it is not. Kill if a critical claim is contradicted and no version of the idea survives without it.

Keep two questions apart: how strong is the evidence, and how good is the idea. A claim nobody has published anything about is untested, not false. The full decision is in Continue, pivot or kill: how to decide on a startup idea.

The reading takes hours to days by hand, depending on how crowded the space is. The experiments take days to weeks, because they wait on real people.

Software can compress the reading. Buteo, the AI idea validator, runs steps 1, 3 and 4 for you: one measured first pass ran 55 searches, kept 52 findings and mapped 8 competitors in 13 minutes. That is one real run of one idea, so yours will differ. It then proposes the experiments for step 5 and waits for your results, because no tool can run a customer conversation for you. The whole method is on the methodology page.

• Asking friends and family. They are answering a question about you, not about the product.
• Counting compliments as demand. “I would use that” costs nothing to say. A deposit, a signed-up email or an hour of someone's time is a commitment.
• Collecting only confirming evidence. If nothing you found argues against the idea, you did not look.
• Trusting a number because it has a link. Open the page and find the number.
• Validating the solution first. Establish that the problem is real before testing whether your answer to it works.
• Setting the bar afterwards. Decide what would stop you before the evidence arrives.
• Reading one score as the answer. A single number hides which claim is weak, and that is the only thing you can act on.