Discovery interview questions
Discovery questions a Product manager can be asked, from first job to senior. Open one to see the strongest answer and why it works.
A stakeholder asks for a chatbot on the settings page. First move?
The strongest answer: Name the job they are hiring the page to do, and whether a bot is the only way — I do not open Figma for a widget first
Why it works: Exactly. Opportunity before solution. Invented lifts are how you skip the job.
- Write the job before the widget.
- An ask ≠ the solution.
- Do not multiply the widget.
- No invented lift %.
You cannot name who succeeds if this ships. What is missing?
The strongest answer: A user and a moment — I stop and find one real person in that moment before I write stories
Why it works: Exactly. A person in a moment. Everyone, stock photos, and only-CEO are how you build a hallway.
- Name a person and a moment.
- Stock personas are décor.
- Everyone ≠ a user.
- Sponsor ≠ user.
What does “jobs to be done” actually mean, and what does a job statement look like?
The strongest answer: People hire a product to make progress in a situation — the job is that progress, written as the situation, the motivation and the outcome, with none of our features in the sentence
Why it works: Exactly — situation, motivation, outcome. “When I …, I want to …, so I can …” keeps the solution out of the sentence.
- A job is progress in a situation.
- Write situation, motivation, outcome.
- Keep our product out of it.
- Same person, different jobs.
How do you start a discovery interview?
The strongest answer: With their last time they tried the job, not “would you use our idea?” — stories beat compliments
Why it works: Exactly. Last time they tried. Demos and excitement scores collect politeness.
- Ask the last time they did the job.
- Do not lead with your idea.
- Demos collect praise.
- Excitement scores are not evidence.
“We need a dashboard.” What is the opportunity underneath?
The strongest answer: A decision someone cannot make with what they have — I write that, I do not start with twelve charts
Why it works: Exactly. A stuck decision. Wallpaper and invented insight lifts are décor.
- Opportunity = a stuck decision.
- Dashboards are one shape.
- Do not lead with twelve charts.
- No invented insight lift.
A competitor shipped a feature. Your move?
The strongest answer: Ask which of our users has that job and whether we must win it — I do not copy the screenshot as a spec
Why it works: Exactly. Their screenshot is their bet. Extra toggles and invented leapfrogs are panic.
- Check our users’ job.
- Do not copy a screenshot as spec.
- Extra toggles ≠ winning.
- No invented leapfrog %.
How do you keep a team from marrying the first solution?
The strongest answer: Write the opportunity, then at least two ways, and a reason you picked — I do not start the sprint in love
Why it works: Exactly. Two ways and a pick. First-love, ten-and-none, and loudest-voice skip the opportunity.
- Opportunity, then ≥2 ways, then a pick.
- Do not sprint in first love.
- Do not collect ten and choose none.
- Loud ≠ chosen.
The PRD is signed. Do you stop talking to users?
The strongest answer: No — I keep a thin stream so the bet can still die — signed is not sacred
Why it works: Exactly. Signed is a snapshot. Phases and vanity dashboards are how you ship a dead bet.
- Keep a thin discovery stream.
- Signed ≠ sacred.
- Do not phase-gate learning to death.
- Do not replace users with vanity.
When is a persona useful, and when is it a costume?
The strongest answer: Useful when it names a job and a constraint; a costume when it is a stock photo and a cute name we never meet
Why it works: Exactly. Job and constraint. Names, dogs, and “always/never” skip the work.
- Persona = job + constraint.
- Stock photos are costumes.
- A cute name ≠ research.
- Do not invent biographies.
A design partner loves you. What do you still refuse to write?
The strongest answer: “Customers will love this” — they are one account, and I will say so
Why it works: Exactly. One helped account. Logo walls, invented %, and hidden white-glove are how you fool yourself.
- Label a design partner as one account.
- Do not write “customers will.”
- No invented case-study %.
- Disclose white-glove.
A six-week build is scoped. One slide lists four assumptions: people want this, they will trust us with the data, we can charge a premium, and we can build it in six weeks.
Which do you test first, and how do you decide?
The strongest answer: The one that would kill the bet and that we know least about — often not the build risk, since engineering has done similar work. I buy the cheapest evidence on that one before the six weeks start
Why it works: Right — rank by damage if wrong against how little you know, then buy the cheapest evidence that could change the plan.
- Test what would kill the bet.
- Rank by damage times ignorance.
- Buy the cheapest evidence.
- Test before the build, not after.
Eight PMs do ad-hoc calls. What do you standardise?
The strongest answer: How we store consent, notes, and the decision they informed — I do not standardise every script
Why it works: Exactly. Memory and consent. One-script, one-bottleneck, and novels are how discovery dies.
- Standardise consent, notes, decision-link.
- Let scripts differ.
- Do not bottleneck every call.
- Do not write a shelf novel.
Notes live in twelve Drive folders. What do you fund?
The strongest answer: A place tagged by decision, with access rules — I do not fund a 80-page toolkit nobody uses
Why it works: Exactly. Findable by decision. Toolkits, bans, and memory are how you re-interview the same pain.
- Store notes by the decision they informed.
- Access rules.
- Do not fund a shelf toolkit.
- Do not ban notes to stay light.
How do you run discovery next to delivery without lying to both?
The strongest answer: A named split of time and a rule that discovery can kill a delivery slice — I do not pretend 100% of both
Why it works: Exactly. Named split and a kill right. 100%-both, pause-all, and standup-as-discovery are lies.
- Name the time split.
- Discovery may kill a slice.
- Do not claim 100% of both.
- Standup ≠ discovery.
PMs are rewarded only for shipping. What breaks?
The strongest answer: They will ship the first idea — I reward a killed bet and a well-asked question, not only a launch
Why it works: Exactly. Incentives write the science. Ship-only, mandatory launches, and leaderboards are how you get junk.
- Reward questions and honest kills.
- Not only launches.
- Do not require a launch every sprint.
- Do not run a launch leaderboard.