📚 "The Mom Test"

Customers bring it up without prompting, speak about it with emotion or specific detail, and have already tried to solve it. Unprompted mention is the strongest shortcut to signal.

Enough to reach thematic saturation — typically 5–10 interviews with similar customers in the same segment. Single conversations are anecdotes; patterns across a segment are evidence.

Talk about their life, not your idea. Ask about specifics in the past, not hypotheticals. Talk less, listen more.

Interviewing champions (enthusiasts who love your idea) instead of decision-makers or typical users produces skewed data — you get enthusiasm without budget authority, or edge-case needs that don't generalise.

Have a dedicated note-taker so the interviewer can stay present and follow threads. Notes should capture exact quotes and specific facts — not summaries or interpretations — to preserve the raw signal for later analysis.

Define who you're talking to before you start — same role, same context, same problem domain. Mixing segments makes it impossible to distinguish signal from noise because different people have different problems.

Politeness mistaken for demand, feature requests mistaken for buying intent, curiosity mistaken for urgency, and stated preferences mistaken for actual behaviour.

Stopping interviews too early because the first few conversations seem consistent — before you've stress-tested across the full segment or found disconfirming cases. It mistakes early pattern-matching for validated insight.

When customers soften criticism with praise — 'I love the concept, but...' — founders remember the praise and discount the criticism. The compliment is noise; the 'but' is the signal.

The Mom Test is a set of rules for asking questions that even your mom can't lie to you about — by making conversations about the customer's life rather than your idea, you remove the social pressure to be encouraging.

Facts = observable past behaviour (what they did, paid, used). Opinions = speculation (what they think, want, or might do). Only facts should influence product decisions.

Bad conversations don't just fail to give signal — they actively manufacture reassurance. Compliments and vague enthusiasm feel like validation but are usually social niceties masking the absence of real pain.

Asking 'why' three times in response to a stated problem surfaces the root cause beneath the symptom — each layer moves you closer to the real job-to-be-done and further from the surface complaint.

When an answer is vague, follow up with a question about a specific past instance. 'It's a big problem' → 'What happened the last time it came up?' 'We do it manually' → 'Walk me through that step by step.' Vagueness is a prompt to go deeper, not a signal to move on.

Treat it as weak signal only. Follow up with: 'Have you tried to solve this before?' and 'What did that cost?' Stated willingness to pay is cheap; prior spending is expensive and therefore credible.

One outlier can distort strategy. Valid signal requires repeated patterns across a segment — similar pain, similar language, similar failed attempts, similar buying triggers.

A learning goal means you enter the conversation to understand their world. A validation goal means you enter to confirm your hypothesis. Only learning goals produce reliable data — validation goals introduce confirmation bias before the interview starts.

Warm conversations (via intros) reduce the pressure to pitch because context is pre-established. Cold outreach creates a sales dynamic by default. Warm intros consistently produce more honest, exploratory conversations.

It forces concrete recall: 'When did this last happen?' 'What did you do?' 'What did it cost you?' If they answer vaguely, keep digging — vagueness is a prompt to zoom in, not a signal to move on.

People accurately recall what they've done but poorly predict what they'll do — so asking 'what did you try?' surfaces real evidence; asking 'would you use this?' surfaces social performance.

Ask what they've already spent — time, money, or effort — trying to solve it. A problem with no prior attempts is probably not painful enough to buy a solution for.

It repositions discovery from a research activity into a bias-control mechanism — the goal is not to hear encouraging things, but to design conversations that make self-deception harder.

What you learned, whether the problem is real, how severe it is, whether the segment is worth pursuing, and which assumption changed.

Time (agreeing to a trial or follow-up meeting), Effort (sharing data, process detail, or workflow access), Reputation (introducing you to others), Money (pre-ordering or allocating budget). Enthusiasm without any of these is noise.

Sharper prioritisation (kills ideas with no real pain), better team alignment (anchors decisions in observable behaviour), and stronger system learning (repeatable assumption-testing over time).

Not from deception, but from politeness, desire to be helpful, and inability to predict their own future behaviour. This is why self-reported intent is structurally unreliable regardless of how honest the customer seems.

The moment you mention your solution, the conversation shifts from truth-seeking to validation-seeking. People react to your idea rather than describing their reality, and every subsequent answer is filtered through politeness. Keep discovery and selling in separate conversations.

Feature requests describe a customer's imagined solution, not the underlying pain. The pain may be real, but the feature they're requesting may be the wrong fix — and their willingness to request it doesn't mean willingness to pay.

Ranking and rating prompts people to engage with your framing rather than their own reality — it's still a hypothetical, just dressed up as quantitative. The result is structured noise.