The Mom Test

You get off a customer call feeling great. They said the idea was interesting, that their team would definitely use something like that, that it would save them hours. You write an enthusiastic summary. Four months later the feature is live and that same customer has opened it twice. Nobody lied to you. You asked questions that could only produce nice answers.

Rob Fitzpatrick's framing is that a good question is one your mother could not give you a falsely encouraging answer to, however much she loves you. The test is not about mothers. It is about whether the question is anchored in something that already happened, because the past is a fact and the future is a guess people are happy to make flattering.

Three rules, and the middle one carries most of the weight. 'Would you use this?' asks someone to predict their own behaviour under conditions that do not exist yet, while looking at an enthusiastic person who clearly wants a yes. Everything about that setup produces a false positive. 'Walk me through the last time you had to do this' asks for a recording, not a prediction.

Notice that the right hand questions are harder to answer. That is the point. A vague answer to a specific past question is itself information: if someone cannot remember the last time a problem bit them, it is not biting them often.

The three things people say that mean nothing

• Compliments. 'That is really cool', 'nice idea', 'I like where you are going'. These are politeness, and they feel like data because they feel good. Note them as social noise and move on.
• Generics. 'We usually', 'typically', 'normally I would'. A generic is a summary someone constructs on the spot, and it drifts toward the version that makes them look organised. Follow with: tell me about the most recent time.
• Futures. 'I would', 'we might', 'if you built that we would probably'. Zero predictive value. Redirect to what they have already done about the problem.

The redirect matters as much as the recognition. When someone gives you a generic, you do not correct them. You say 'when did that last happen?' and let them go and find the episode. People are reliable narrators of specific events and unreliable narrators of their own patterns.

One thing this approach is bad at, and you should know it. Past behaviour tells you little about genuinely new categories, because there is no past behaviour to ask about. If you are building something people have never had an equivalent of, you can still ask about the underlying job and what they do instead, but be honest that you are reasoning by analogy. The Mom Testthe Mom Test: Asking about past behaviour rather than future intent, so people cannot flatter you. protects you from false enthusiasm. It does not give you clairvoyance.

The deeper habit underneath all of this is not mentioning your idea early. Once your idea is on the table, every subsequent answer is shaped by it, and the person is now trying to be helpful about your idea rather than accurate about their life. Save it for the last five minutes, if at all.