An idea is not yet a hypothesis

AI for startups · Lesson 1 / 22

The difference between an intention and a testable claim

"A service that helps small businesses with bookkeeping using AI" is not a hypothesis. It is a description of intent. It cannot be disproved: any outcome can be explained after the fact. A hypothesis is a claim for which you know in advance which observation would make it false. If no such observation exists, you are not testing an idea, you are collecting arguments in its favour and calling that research.

Building a prototype, a landing page and copy now costs close to nothing. The expensive part of the work has moved: the hard thing is not building, it is finding out in time that you should not build at all. A written hypothesis is the cheapest instrument you have before you spend six months of your life.

What a usable claim is made of

  • A segment, not a market. Not "small business", but "the owner of a three to eight chair salon who files their own reports".
  • The problem as behaviour. What the person does today, how many hours and how much money it costs, who else is involved.
  • A claim with a number. One sentence containing a quantity rather than an adjective.
  • A disproof condition. What has to happen for you to honestly say it did not hold.
Hypothesis: [segment] spends [N hours or dollars] per [period] on [action]
because [current workaround] fails to cover [condition].
Test: [method], sample [how many people], window [how many days].
Disproved if fewer than [threshold] of the sample name this among their
top three problems without being prompted.

Set the threshold before you collect data, never after. A threshold invented after the interviews is not a threshold, it is a justification. A reasonable early bar: the topic comes up unprompted with at least half the people in your chosen segment.

Insight. A good hypothesis is written so that a colleague can run the test instead of you and reach the same conclusion. If interpreting the result requires you and your belief, the wording is broken.
Common mistake. A hypothesis containing the word 'and' is two hypotheses. 'Salons lose time on reporting and will pay for automation' splits in half: the first part is almost always confirmed, the second almost never, and you score the win on the first part.
Pro tip. Keep a hypothesis log with the date, the threshold and the verdict: confirmed, disproved, inconclusive. Three months in, that file is the only evidence that you were moving rather than drifting.

Cheat sheet

  • No disproof condition means no hypothesis.
  • Describe the segment so it can be turned into a list of names.
  • Fix the threshold before collecting data.
  • One "and" in the claim means cut it in two.
1. What turns an idea into a hypothesis?
2. When is the confirmation threshold set?
3. Why is a hypothesis containing 'and' dangerous?

🔒 Answer the question correctly to move on to the next lesson.

An idea is not yet a hypothesis — AI for startups — Skilvy