Interview Method

How to ask so you get evidence—not politeness, predictions, or a pitch meeting. This page is the how that runs through Who, Why now, Workflow, and Talk / search / buy.

What it is

Discovery interviews test hypotheses about problems, workflows, and decisions by listening to past behaviour in real context. The compact standard is Rob Fitzpatrick’s The Mom Test: ask about their life, not your idea; prefer the past over the hypothetical; talk less, listen more. You are not validating a pitch deck. You are reconstructing episodes—what happened, what they tried, what hurt, what they paid (time, money, attention) to make the pain stop.

A good interview has a simple arc: open on a concrete episode → walk the sequence (trigger, tools, people, failures) → harvest emotion and workarounds → only then, if ever, mention your idea. Anything that reverses that order contaminates the data.

Why it works

People are generous with predictions (“I would definitely use that”) and honest about the past (“last quarter I exported to Sheets and emailed it”). Compliments feel like validation and are almost worthless: friends, moms, and polite strangers will praise your idea to end the conversation kindly. Pitching mid-interview trains them to flatter you and trains you to hear what you hoped to hear.

Evidence from concrete episodes is what personas, need states, and voice-of-customer (VoC) are built from—and what agents can later load without inventing customers. Hypothesis-driven discovery (Steve Blank’s customer development) means you walk in with assumptions to kill, not a blank slate and not a feature vote. Focus groups and “would you use…” surveys fail here for the same reason: they elicit opinions about the future instead of behaviour from the past.

Going deeper

Craft that separates evidence from noise:

1. Open on an episode, not a category. “Walk me through the last time this happened” beats “How do you feel about expense tracking?” The last real instance recovers trigger, circumstance, tools, and emotion; the abstract question recovers a persona they wish they were.
2. Stay in the past tense. “What did you do next?” and “When did that break?” outperform “Would you want…?” and “How important is…?” Importance without a paid or endured cost is theatre.
3. Mine the four fields every time. Capture trigger (why that day), emotion words (verbatim), workaround (what they built or endured), and switching stakes (what a move would put at risk). Those four fields feed the other discovery pages directly.
4. Defer the pitch. If they ask what you are building, park it: “I’ll show you at the end—I don’t want to bias the story.” Showing early turns the rest of the call into a demo review.
5. Match the person to the question. Recent buyers and switchers carry buy-journey truth; practitioners who do the job weekly carry workflow truth. Interviewing only founders’ friends, colleagues, or people outside the Ideal Customer Profile (ICP) and calling it discovery is recruiting theatre.
6. Watch when you can. Contextual inquiry—seeing the work in its environment—beats recollection for routines people have stopped noticing. Pair episode interviews with observation when the job is habitual and invisible.

For builders and agents

Treat interviews as a pipeline into durable docs, not as vibes stored in chat. With consent, record or take near-verbatim notes; tag by segment and need state so synthesis can cluster. Keep raw session notes ephemeral (for example .productfeeling/sessions/); graduate only validated claims into docs/ via Synthesis.

Agents must not invent customers. Before /productfeeling persona or jobs, load what discovery already wrote—or run interviews and synthesis first. When an agent helps prep a call, have it generate past-tense prompts and a capture card, not a pitch script. When it summarises a transcript, force the four fields (trigger, emotion, workaround, stakes) and flag every compliment and future-tense promise as non-evidence.

Red flags that the session was a pitch, not discovery: you talked more than they did; the notes are mostly about your product; the only “proof” is “they said they’d buy.”

When to use it

• Before locking ICP, north star, or a major surface
• When feedback is loud but unsegmented
• When you only have opinions (“users want…”) and no episode
• Continuously—discovery is not a one-time phase

Do

• Open with a real episode: “Walk me through the last time this happened”
• Ask about what they did, tools used, who else was involved, what failed
• Capture trigger, emotion words, workaround, and switching stakes every time
• Stay in their world; if they ask what you are building, defer until after the story
• Prefer recent buyers/switchers for buy-journey truth; prefer practitioners for workflow truth
• Record (with consent) or take near-verbatim notes; tag by segment and need state

Don’t

• Pitch the product mid-interview or ask “Would you use a tool that…?”
• Lead with your category jargon or feature list
• Treat compliments as validation
• Interview only friends, colleagues, or people outside the ICP and call it discovery
• Confuse a sales demo with a discovery interview

Founder Tip

If you leave talking more than they did, you ran a pitch—not an interview.

Make It Yours

1. Hypothesis list — three assumptions you will try to kill this week.
2. Episode script — five prompts that stay in the past (trigger → steps → fail → emotion → next time).
3. Capture card — fields: segment, trigger, emotion words, workaround, stakes, awareness cue.
4. Stop rule — what evidence would make you change ICP or kill a proto-persona.

Insights & Metrics

1. Episode yield — Interviews with a full capture card ÷ Interviews run.
2. Hypothesis kill rate — Assumptions falsified or revised ÷ Assumptions tested (healthy discovery kills guesses).
3. ICP hit rate — Interviews that match ICP criteria ÷ Total interviews (low hit rate = recruiting problem).

Behind the Data

• Are you still collecting compliments instead of episodes?
• Which hypothesis has survived three interviews without evidence?
• Are session notes stuck in chat while nothing durable landed in docs/?

Related concepts

• Feeling North Star, Jobs-to-be-Done, User Agency
• Feeds: all other Customer Discovery pages; Synthesis

Further reading

• The Mom Test (Rob Fitzpatrick) — Evidence over politeness.
• User Interviews 101 (Nielsen Norman Group) — Structure, probing, bias.
• Customer Development is Not a Focus Group (Steve Blank) — Hypotheses, not feature voting.
• Contextual Inquiry (NN/g) — When watching beats asking.

Agent skill

• Primary command: /productfeeling init — capture what is already known and where durable findings will live; do not invent customers in chat
• Related commands: /productfeeling persona, /productfeeling jobs, /productfeeling sequence (Customer discovery)
• When the agent should load this page: “interview”, “Mom Test”, “discovery call”, “talk to users”, “validate assumption”
• Companion handoff: DocSlime — only validated findings graduate to docs/; raw session notes may stay in .productfeeling/sessions/. Never an external discovery skill. Impeccable out of scope during interviews. RedTeam optional on stubborn assumptions after synthesis.
• Feeling north star this practice serves: grounded clarity from evidence, not vibes
• Anti-goals: pitching in interviews, compliment-as-validation, non-ICP recruiting theatre
• Reference path: skill/reference/init.md
• Durable DocSlime targets: validated findings only → docs/strategy/ and docs/experience/ via Synthesis; sessions → .productfeeling/sessions/ until graduated