Your PMF survey has a blind spot built into how it's sampled. It goes to current users, the people still using the product today. That's the right population for a score, but it's the wrong population if you want to know exactly where things break, because the users who found the biggest problem already voted with their feet. They churned, and the standard survey never reaches them again.
That's what an exit interview is for. Not a replacement for the Sean Ellis survey, a complement to it: a short conversation or open-ended form with someone who already left, aimed at the one moment that matters, the day they decided your product wasn't worth keeping.
Why this group is worth the extra effort
Read the three PMF survey segments again with churn in mind. "Not disappointed" users usually don't bother canceling, they just quietly stop opening the app. "Somewhat disappointed" users are the ones who actually churn on purpose, because they saw enough value to try, hit a specific wall, and left instead of working around it. That's precisely the segment your PMF score most wants to convert, and exit interviews are the only place you'll hear their reasoning in their own words, after the fact, when they have nothing left to lose by being honest.
The dropdown trap
Most cancellation flows ask "why are you leaving?" with a dropdown: too expensive, missing a feature, switched to a competitor, other. It feels efficient. It's mostly noise, because it forces a real, specific story into a pre-written category that was written before this particular user ever canceled.
Groove ran into this directly and published the fix: they replaced their closed-ended cancellation survey with a single open question, and response rates grew from 1.3% to 10.2%, a 785% increase (Groove, "How We Grew Our Customer Exit Survey Responses by 785%"). People will tell you the real story if you ask an open question instead of handing them a multiple-choice test about their own decision.
See who's actually at risk before they leave
PMFtracker segments your PMF survey into very, somewhat, and not disappointed automatically, so you know which users are the growth lever before they turn into an exit interview.
Measure your PMF score free → 14-day free trial · No credit cardWho to interview, and who to skip
Not every canceled account deserves a conversation. Skip signups who created an account and never really used the product; they can only tell you your onboarding failed, which is a real problem but a different one from a job the product used to do and stopped doing. Prioritize users who were genuinely engaged, using the product regularly for at least a few weeks, and then stopped. Their story is about a job that broke, not a job that never started.
Timing matters too. Reach out within a week or two of cancellation, while the specific day and the specific reason are still fresh. Wait a month and you'll get a vague, rationalized summary instead of the actual sequence of events.
The questions that avoid the feature-request trap
The Product Death Cycle applies just as hard to exit interviews as it does to feature requests from active users: it is not the customer's job to design your roadmap. Asking "what feature would have kept you?" gets you a guess, not evidence. Ask about what actually happened instead:
- "Walk me through the day you decided to cancel." Not their opinion of the product, the actual sequence of events. Specific days produce specific, checkable answers.
- "What were you trying to get done right before that?" This surfaces the job they originally hired you for, in their own words.
- "What did you switch to, or start doing instead?" Tells you whether you lost to a competitor, a manual workaround, or the job disappearing entirely, three completely different problems that need three different fixes.
- "Where in that story did we stop being useful?" This is the one question that gets closest to a feature answer, but it's still anchored to their real story instead of a wish list.
Notice what's missing: "what would you change?" It's tempting to ask, and it's the question most likely to produce a plausible-sounding, low-value answer.
Turning answers into three buckets
Once you've run a handful of these, sort the stories into three buckets, because each one points at a different fix:
- The job stopped being done well. Something you used to nail started failing them, a reliability issue, a workflow that got worse, a competitor doing the same job better. This is your roadmap, sourced from real events instead of feature requests.
- The job was never quite right. They were doing something adjacent to what you built for, and it mostly worked until it didn't. This is an ICP signal: you may be acquiring people just outside your real fit.
- The job disappeared. Their situation changed, budget cuts, a role change, the underlying need went away. There's nothing to fix here. Don't let one vivid story from this bucket derail a roadmap decision.
The mistake to avoid is treating every story equally. Five people independently describing the same broken moment is a pattern worth building for. One person's uniquely bad week is an anecdote, and anecdotes make bad roadmaps, the same trap as reading feature requests from users who never left.
Turn every open-ended answer into a pattern
PMFtracker analyzes the open-ended answers behind your PMF score so the pattern across strangers, not the loudest single voice, drives what you build next.
Start Tracking PMF → Set up in 5 minutes · No credit card requiredWhat this does for your PMF score
Exit interviews don't feed a percentage the way the Sean Ellis survey does. They explain the percentage you already have. If your score is stuck below 40%, the "somewhat disappointed" segment is telling you there's a fixable barrier, and the churned users who used to sit in that segment are the ones who can describe the barrier precisely, because they're the ones who hit it hardest. Run the survey to know your number. Run exit interviews to know exactly what to do about it.
