Validate
Validate behavior, not approval
Useful evidence changes a founder’s next decision. Look for behavior that carries a cost, not compliments, survey answers, or clicks that are easy to give.
Primary question
What observable behavior would show that this problem is real enough to keep pursuing?
Why it matters
The decision behind the output
Validation is often treated as a search for encouragement. Interviews, landing-page visits, and enthusiastic replies are collected into a reassuring story even when none of them touches the risky behavior the product depends on. Useful evidence is narrower: it reduces a named uncertainty enough to authorize, revise, or stop a specific next investment.
The strength of a signal depends on its distance from the claim. Attention can test whether a promise is understood. A live-input working session can test whether the outcome matters in context. A repeated request, operational commitment, or payment reaches further. The goal is not to demand the strongest signal immediately, but to stop promoting a cheap signal into proof of something it never tested.
Watch for
Common mistakes
- Counting praise, sign-ups, or cheap clicks as product evidence.
- Measuring activity without deciding what decision the result should change.
- Building an MVP before testing the riskiest assumption in the workflow.
Use this sequence
A working framework
- State the assumption that must be true for the opportunity to work.
- Choose the smallest test that asks a customer to spend time, money, data, or reputation.
- Define the behavior that would count as evidence before launching the test.
- Use the result to continue, narrow the problem, or change direction.
From the field
A field example
A founder exploring an AI research assistant may attract a hundred sign-ups with a clear landing page. That supports the message, not the workflow. A stronger next test asks several intended users to bring a live research question and source set, then watches whether they inspect the evidence, use the output in a real decision, and return with another case. Each behavior closes a different explanation that the sign-up left open.
Use this threshold
The decision rule
Before running a test, write the decision it may change, the behavior that would count, and the threshold for continuing, narrowing, or stopping. Accept the result only at the level it earned: attention authorizes a closer test, qualified commitment authorizes delivery, and repeated value may authorize product investment. Do not let a positive result answer a larger question than the experiment reached.
Field Notes
Related essays
- Cheap Clicks Are Not Product Evidence
How can founders distinguish cheap acquisition clicks from evidence of product demand?
- How to Validate an AI Startup Idea Without Building an MVP
How can founders validate an AI startup idea by testing risky behavior and a real transaction before building an MVP?
- Customer Interest vs. Product Evidence: What Actually Counts?
Which customer signals count as product evidence, and what decision can each signal honestly justify?
- When to Stop Validating and Start Building
When has validation reduced enough uncertainty that the next responsible experiment is a real product build?