Product manager interview questions, with strong answers
32 questions from Compass's mock interviews, the way an interviewer asks them. Each has a strong answer, a common miss, and why one lands and the other doesn't.
Product design
You're a PM at a photo-sharing app in 2016. A rival's disappearing "stories" format is growing fast with your users. What do you do?
A strong answerFirst, the goal: keep everyday sharing on our app, or win new users? I'll assume keeping casual sharing, inside the app we have.
Goal first, assumption stated out loud. That's the senior opening.
A common missShip the same format fast: a 24-hour feed above the main feed, built by one team and live in about six weeks.
A solution in the first sentence. Interviewers mark that down.
Who exactly are you building for?
A strong answerExisting users who rarely post because the feed feels permanent and judged, especially younger people also on the rival app.
A segment defined by a behavior and a pain, not a demographic.
A common missEveryone on the app. Stories work for every kind of user, so narrowing it down would only leave value on the table.
"Everyone" is not a segment. Pick one and say why.
What's their problem, in their words?
A strong answer"Posting feels like a performance." There's no low-pressure place for in-the-moment updates to people who follow me.
Their words, and it explains why they post less. Sharp.
A common missThey don't have enough filters and editing tools to make their posts look as polished as the ones they scroll past.
That pushes them toward polish, the opposite of their pain.
Give me a few options.
A strong answerThree that differ in kind: casual filters in the feed, a separate ephemeral app, or a 24-hour format above the feed.
Safe to bold, and genuinely different bets.
A common missJust the stories tray above the feed. It's clearly the right answer, so there's no need to weigh anything else.
One idea isn't a choice. Show the options you rejected.
Which one do you pick, and why?
A strong answerThe tray above the feed. It reuses our follower graph, which a newer rival can't match, and keeps sharing in our app.
Criteria named, and it plays to a real strength.
A common missWhichever option engineering can build fastest, since getting something out before the rival grows further matters most.
Speed is one criterion, not the reason. What wins for users?
How will you know it worked?
A strong answerSuccess is daily story posters and viewers. Guardrails: feed posting and time in the feed shouldn't collapse.
One success measure and honest guardrails.
A common missApp downloads going up in the weeks after launch, since a hit feature should bring a wave of new people in.
Downloads don't show anyone sharing more casually.
What worries you about this?
A strong answerBeing seen as a copy, cannibalizing the feed, and crowding the top of the app. I'd watch the feed guardrail weekly.
Real risks, each with a mitigation.
A common missNothing major. Users are clearly asking for this format, so they'll love it, and the launch should go smoothly.
There's always a risk. Interviewers want to hear you name it.
Metrics and goals
You run growth on a mature language-learning app. Daily users grew fast for years; now growth is in single digits. Where do you start?
A strong answerAgree on the goal first: daily active users or paying users? I'll assume daily actives, then find where users leak.
Goal first, and a hint of a model. Good start.
A common missLaunch a big marketing push to bring in new users. More people at the top of the funnel fixes slowing growth.
Spending before diagnosing. Interviewers want the why.
How would you break daily active users down?
A strong answerBy user state: new, current, at-risk, dormant, resurrected. Daily actives are flows between states, so each is a lever.
States and flows turn one number into levers. That's the senior move.
A common missBy feature: lessons, stories, leaderboards. Whichever feature people use least is the one that needs the work.
Feature usage says what people do, not who's slipping away.
Which flow do you focus the team on?
A strong answerModel it: nudge each rate a little and see which moves daily actives most. Keeping current users likely compounds most.
A simulation instead of a hunch. That's how you pick.
A common missAll of the flows equally, with a small squad on each, so that no lever gets neglected while we figure it out.
Focus is the point. Pick one and say why.
How will you know it's working?
A strong answerCurrent-user retention as the focus metric, daily actives as the outcome, and the share on long streaks as a quality check.
A focus metric, an outcome, and a quality signal. Clean.
A common missThe app-store rating going up in the months after the changes ship, since happier users leave better reviews.
Ratings lag, and they don't measure retention.
What worries you about this plan?
A strong answerStreak mechanics that feel manipulative, and mistaking correlation for cause. I'd test changes in experiments first.
Ethics and causality, with a plan. Exactly.
A common missNothing much. Retention work is low risk compared with new features, so there's little that can go wrong here.
Every lever has a cost. Name it.
Strategy
Millions of people watch your streaming service on someone else's password. Leadership wants them paying. Where do you start?
A strong answerDefine the goal: new paying members, not punishing sharers. And the scope: worldwide, but we can stage it by market.
Goal and scope, and the goodwill trade-off already in view.
A common missBlock shared logins outright. People who really want to keep watching will sign up for their own accounts.
A blunt solution before anyone's been considered. Slow down.
Who's affected by this change?
A strong answerThree groups: owners who share, borrowers watching free, and travelers legitimately on their own account away from home.
Three groups with different needs, including the edge case.
A common missOnly the borrowers. Account owners won't care what happens to their friends' access, so they're not affected.
Owners care a lot. They're the ones who might cancel.
What are your options?
A strong answerA hard block, a paid extra-member add-on, moving a profile to a new account, or a cheaper ad plan for borrowers.
Options that differ in kind, from gentle to firm.
A common missEmail every owner asking them politely to stop sharing their password, and trust that most of them will listen.
Hope isn't a plan. What would actually change behavior?
Which do you pick, and how do you roll it out?
A strong answerThe add-on plus profile transfer, so borrowers keep their history, tested in a few markets before a wider launch.
A combined answer, staged to learn before betting big.
A common missThe hard block everywhere on the same day, so that nobody can argue they were treated differently from anyone else.
One big switch maximizes backlash and teaches you nothing.
How do you know it worked, and what could go wrong?
A strong answerNew paid members and revenue, with cancellations and support contacts as guardrails. Risk: blocking real travelers.
Success, guardrails, and the false-positive risk. Strong.
A common missFewer shared logins. That's the whole point of the change, so that one number tells us everything we need.
Fewer logins could just mean people left. Measure the outcome.
Prioritization
Your video-call app went from about 10 million to over 200 million daily meeting participants in a few months. Security problems are in the news. What's the goal?
A strong answerKeep the trust of all these new users while the service stays up. Trust is the goal right now; features can wait.
A clear goal that names what you'll trade away.
A common missShip the most requested features quickly, so all the new users have reasons to stay once things calm down.
More features while trust is breaking makes it worse.
Who are these new users?
A strong answerSchools and consumers, with very different security expectations from the company IT teams we were built for.
New users with new needs. That explains the problems.
A common missThe same users as before, just many more of them. Nothing about the product needs to change for them.
Schools and families aren't IT teams. Their needs differ.
What are your options?
A strong answerPatch issues one by one, set up a security task force, or freeze features and put every engineer on trust.
Three real options, from small to drastic.
A common missRun an advertising campaign telling people the product is safe, so the headlines stop scaring them away.
Saying it's safe doesn't make it safe.
Which do you pick, and what does it cost you?
A strong answerThe freeze: a credible signal and focused effort. The cost is pausing the roadmap during a huge growth moment.
The pick, the reason, and the price paid. That's prioritization.
A common missKeep the roadmap and fix the issues as they come up in the press, one headline at a time as needed.
Reacting to headlines leaves you a step behind every time.
How do you know the 90 days worked?
A strong answerIssues fixed and reviews done, plus trust signals like sentiment, with uptime and growth as guardrails.
Output, outcome and guardrails, all named.
A common missFewer news articles about the problems, since the coverage is what's scaring people away in the first place.
Press can quiet down while the problems remain.
Estimation
You're adding an AI "explain my answer" button to a learning app. Estimate its monthly model cost. How do you start?
A strong answerWrite the formula first: users who tap it, times taps per user, times cost per call. Then fill in each part.
Formula first, so every assumption can be challenged.
A common missAsk the finance team for last month's AI bill, and add a little on top to cover the new feature's usage.
Someone else's number isn't an estimate. Build your own.
How many people will use it?
A strong answerMonthly actives, times the share who see it (paid tier only?), times the share who tap it. Each assumption said aloud.
A funnel with stated assumptions. Exactly right.
A common missIt depends on too many things. I'd wait for real launch data before putting any number on it at all.
Estimation means committing to assumptions. Make one.
What does one call cost?
A strong answerTokens in plus tokens out, times the model's price per token: the prompt and question in, a short explanation out.
A unit cost built from its parts. Strong.
A common missClose to nothing. Model prices keep falling every few months, so the cost per call barely matters anymore.
Small times millions is not small. Price it.
Your total comes out at $40 million a month. What now?
A strong answerThat's out of line for a feature this size, so I'd find the assumption that's off, likely taps per user, and fix it.
You checked it against reality and found the culprit.
A common missGo with it. The formula was right, so the number must be right too, even if it looks surprisingly large.
A right formula with a wrong input gives a wrong answer.
How would you bring the cost down?
A strong answerA smaller model for easy questions, caching common explanations, and a paid tier that matches cost to value.
Engineering and product levers together. That's the full answer.
A common missTurn the feature off if the bill gets too high, and bring it back later once model prices have dropped.
Giving up isn't a lever. Find ways to keep it affordable.
AI product sense
Design an AI assistant for a fintech's customer-service chat. What's the goal?
A strong answerFaster, always-on help at lower cost, for well-bounded jobs like refunds, returns and payments. Disputes may need people.
Goal plus scope, and the edge where humans stay.
A common missReplace the support team entirely within a year, since the models are now good enough to do their whole job.
That's a headline, not a product goal. Who does it serve?
What does it handle, and what does it hand off?
A strong answerIt answers order, return and payment questions from our own policy; money decisions and upset customers go to a human.
Clear boundaries, grounded in policy. That's the design.
A common missIt handles refunds from end to end and decides them on the spot, so that nobody ever has to wait for an agent.
Money decisions by a model invite costly mistakes.
How do you know it's good before launch?
A strong answerA golden set of real chats with good answers, graded against a rubric, then a small monitored slice of traffic.
Measured quality, then a careful rollout. Textbook.
A common missLaunch it to everyone and watch the complaints, since real users are always the best test of any feature.
Customers shouldn't be your test set for money questions.
What do you measure once it's live?
A strong answerShare of chats resolved, resolution time, repeat contacts, and satisfaction compared with human agents.
Outcomes for customers, with a fair comparison.
A common missCost savings. If the support budget falls after launch, then the assistant is clearly doing its job well.
Costs can fall while customers get worse help.
What worries you about it?
A strong answerWrong financial answers and a hard handoff. I'd ground it in policy, log failures, and keep a human one tap away.
Real risks with layered guardrails. Strong finish.
A common missNot much. Today's models are accurate enough for support work, so the risk of serious mistakes is quite low.
Accurate enough isn't always right, and here mistakes cost money.
Reading answers helps. Saying them under a little pressure helps more: Compass runs these as mock interviews and scores each turn.
Practice freeMore interview guides
- Technical program manager interview questions, with strong answersTPM interview questions on planning, dependencies, risk and launches, each with a strong answer, a common miss, and why one works.
- Sales and account executive interview questions, with strong answersAccount executive interview questions and role-plays on discovery, pipeline and closing, each with a strong answer, a common miss, and why.
- Case interview questions, with strong answersConsulting case interview questions on structure, profitability and market sizing, each with a strong answer, a common miss, and why one works.
- Investment banking interview questions, with strong answersInvestment banking technical questions on accounting, valuation and deals, each with a strong answer, a common miss, and why one works.