🧠Customer Interviews — Advanced
Past the basics: the second-order moves that separate good interviewers from great ones.
After your first 50 interviews, the marginal insight per interview drops sharply. The advanced moves — segment-based recruiting, hypothesis-driven prompts, observation over self-report — restore the signal.
Advanced customer interviews are hypothesis-driven, segment-aware, and lean on observation rather than self-report. The PM goes in with specific assumptions to test, recruits the right cohort to test them, and watches behavior more than they ask about it.
Six advanced moves
1. Hypothesis-driven prompts. Don't go in with 'tell me about your day.' Go in with 'I believe X is true about this segment; I'm here to invalidate it.'
2. Segment-precise recruiting. General recruiting gets you general insights. To learn about a specific segment, recruit only that segment — and recruit failures (lapsed users) as well as successes (active users).
3. Observation > self-report. Wherever possible, watch the user do the work. Their actual behavior is gold; their description of their behavior is silver.
4. The 'show me' move. When the user describes a workflow, ask them to share their screen and walk through it live. You'll see the workarounds and friction they didn't mention.
5. The five whys. For any expressed preference, drill into why 5 times. By the fifth why, you're at the root motivation. The first answer is usually superficial.
6. Pattern-matching across segments. Talk to 3 segments in parallel. The patterns that emerge across all 3 are the real ones. Patterns in only 1 are segment-specific.
What to do with the data
- Tag each interview by hypothesis confirmed/disconfirmed.
- Maintain a 'quotes database' — verbatim quotes by theme. Pull from this when writing PRDs.
- Run synthesis sessions with the trio. Don't synthesize alone — others see patterns you'd miss.
- Update the OST. Every interview should either confirm an existing opportunity or surface a new one.
How to spot a low-signal interview
- The user spends >50% of the time describing what they want you to build. (You've turned them into a focus group.)
- They never describe a recent, specific instance.
- They use only positive or only negative words. (Likely social-desirability bias.)
- They keep agreeing with your leading questions.
If three of those happen, the interview's value is low. Don't include it in synthesis.
Compensation and recruiting at scale
Once you're running 10+ interviews per week, recruiting becomes a job. Tools: UserInterviews, Respondent, Ethnio. Compensation: $50-150 per 30-min interview, depending on seniority. Pay senior buyers more — they're harder to recruit and their time is genuinely valuable.
For B2B: cold LinkedIn outreach with a clear ask ('20 min, no sales pitch, $100 gift card') converts at 5-15%.
Real-world examples
Sprig pioneered in-product micro-surveys that surface qualitative signal at scale. The pattern: trigger short open-ended questions at moments of friction, then deep-dive interview the most insightful respondents. The hybrid scales discovery without giving up depth.
Go deeper — recommended reading
Interview questions (1)
Q1You've done 5 customer interviews and have 5 different conclusions. What do you do?executionsenior▼
That's actually a normal early state — 5 interviews almost never converge.
Three moves:
- Check your recruiting. Did you recruit from 5 different segments? If yes, divergence is expected; the insight is in the segment-level patterns, not the aggregate. Slice by segment.
- Run 5-10 more interviews in the most relevant segment. Patterns typically emerge after 8-12 interviews within a segment.
- Re-examine your prompts. If interviews are diverging because you let conversations roam, tighten the prompts. Run the next batch with structured hypotheses.
If after 15+ interviews in one segment the conclusions still diverge, the segment itself isn't homogeneous — you might need to define a sub-segment to study.
The wrong move is averaging the 5 different conclusions into a generic insight. That generates a feature nobody actually wants.