TeardownIQ

Interview Prep

Product manager interview questions, with the framework for each

These are the questions inside TeardownIQ's question bank — the same ones you can practice out loud with Marvel, our AI interview coach. Each one lists the framework interviewers expect and the beats a strong answer hits. Nothing here is invented for SEO: it is the product.

How to use this page

  1. Pick a category that matches the loop stage you are preparing for.
  2. Read the framework first — most weak answers fail on structure, not content.
  3. Say your answer out loud against a timer. Reading a question is not practising it.
  4. Then run it with Marvel, who probes follow-ups and scores you on structure, evidence and clarity.

Tell Me About Yourself questions

Walk me through your background and what brought you to product management.

Past → Present → FutureThree beats: where you started, what you do now, why this role next.

What a strong answer covers

  1. Past: 1-2 sentences on your origin story (engineering, design, founder, etc.).
  2. Present: current role, scope, one signature outcome with a metric.
  3. Future: why this company / this role is the logical next chapter.
  4. Land on a hook that invites the interviewer's first follow-up.

What's a product you've shipped that you're most proud of, and why?

Context → Decision → ImpactAnchor in a real shipped thing. Lead with the hardest call you made.

What a strong answer covers

  1. Context: the user problem in one line, the constraint that made it hard.
  2. Decision: the non-obvious call you made (and what you said no to).
  3. Impact: a clear metric or behaviour change post-launch.
  4. Reflection: what you'd do differently — shows self-awareness.

Why are you leaving your current role and why this company?

Push / PullBe honest about what you're moving toward, never trash your current employer.

What a strong answer covers

  1. Pull (lead with this): a specific reason this company / role fits — product, mission, stage, team.
  2. Push: an honest growth gap your current role can't fill (scope, domain, ambiguity).
  3. Bridge: how your last 12-24 months prepared you for the next chapter here.

Tell me about a time you changed your mind about a product decision.

Belief → Evidence → New BeliefStrong PMs update on data. Show the loop.

What a strong answer covers

  1. The original hypothesis and why you held it.
  2. The evidence (data, user research, A/B test) that contradicted it.
  3. How you translated the evidence into a new decision and communicated it.
  4. The outcome and what it taught you about your own blind spots.

What kind of PM are you — and what kind aren't you?

Strengths / Anti-StrengthsSelf-aware framing. Pick a real anti-strength, not a humblebrag.

What a strong answer covers

  1. 2-3 strengths anchored in concrete examples (not adjectives).
  2. 1 honest anti-strength (e.g. 'I default to shipping over polish — I pair with design leads to counterbalance').
  3. How you partner with eng/design/data given your shape.
  4. Why this shape fits this specific role.

Product Sense questions

Design a product to help remote teams feel more connected.

CIRCLESComprehend, Identify users, Report needs, Cut prioritise, List solutions, Evaluate, Summarise.

What a strong answer covers

  1. Clarify scope: which kind of remote teams, what 'connected' means in this context.
  2. Pick 1-2 user segments and articulate their top unmet need.
  3. Brainstorm 3-4 solutions, then prioritise on impact × feasibility.
  4. Pick one, sketch the core flow, and name the success metric.
  5. Risks and what you'd test first.

How would you improve YouTube for creators?

User Journey + Jobs-to-be-DoneWalk the creator journey end-to-end, find the highest-leverage friction.

What a strong answer covers

  1. Pick a creator segment (new, growing, established) — different jobs, different pain.
  2. Walk their journey: ideation → record → publish → grow → monetise.
  3. Identify the 1-2 stages with the steepest drop-off or biggest unmet job.
  4. Propose 2-3 features for that stage, prioritise one.
  5. Define success: leading metric (e.g. weekly active creators) and guardrail (e.g. quality).

Your favourite product is X. How would you make it 10x better?

First Principles + CounterfactualDon't iterate the surface. Question the underlying constraint.

What a strong answer covers

  1. State what makes the product loved today (the core job it nails).
  2. Identify the underlying constraint that caps it (technical, economic, behavioural).
  3. Propose a directional bet that removes the constraint — not a feature.
  4. Sketch one concrete v1 of that bet you could ship in 6 months.
  5. How you'd know it's working in 90 days.

Design an app for elderly users to manage their medication.

CIRCLES + Accessibility-firstConstraint-led design. Accessibility isn't a feature — it's the spec.

What a strong answer covers

  1. Clarify: which segment of elderly (independent, assisted living, with caregivers).
  2. Surface non-obvious needs: cognitive load, dexterity, vision, fear of mistakes.
  3. Sketch a core flow that minimises taps and reads aloud by default.
  4. Identify a caregiver loop — the second user.
  5. Success metric: adherence rate, not DAU.

If you were CEO of [a company you know well] for a day, what would you change?

Strategy DiamondArenas, vehicles, differentiators, staging, economic logic.

What a strong answer covers

  1. Where they play today vs. where the puck is going.
  2. Your one strategic shift (kill, double-down, or new bet).
  3. Why that shift — what changed in the market, tech, or user behaviour.
  4. First 90 days: what you'd resource, what you'd cut.
  5. What success looks like in 18 months.

Product Metrics questions

How would you measure the success of Instagram Stories?

Metric Tree (HEART or AARRR)North star → input metrics → guardrails. One number rules them.

What a strong answer covers

  1. North star: pick one (e.g. daily story views per active user).
  2. Inputs that drive it: creation rate, reach, completion rate.
  3. Guardrails: report quality, time spent (avoid dark-pattern wins).
  4. Counter-metrics: cannibalisation of feed posts.
  5. How you'd know the north star is the wrong choice.

DAU is dropping 5% week-over-week. How do you diagnose it?

Segment → Funnel → CohortDon't guess. Slice the data along three axes before forming a hypothesis.

What a strong answer covers

  1. Confirm the data: instrumentation change? seasonality? holiday? release date?
  2. Segment: by platform, geo, user tenure, acquisition channel.
  3. Funnel: where in the activation/retention path is the drop?
  4. Cohort: is it a new-user problem (acquisition/activation) or existing-user (retention)?
  5. Form 2-3 hypotheses and the test that would falsify each.

You launched a feature. Engagement is up but revenue is flat. What do you do?

Goal Tree + CounterfactualEngagement is a means, not an end. Trace the chain back to the business goal.

What a strong answer covers

  1. What was the original hypothesis — engagement → retention → revenue?
  2. Where did the chain break: retention flat? conversion flat? ARPU flat?
  3. Is the engagement healthy (deeper) or empty (more sessions, no value)?
  4. Decision tree: kill / iterate / accept as a leading indicator.
  5. What you'd ship next to test the broken link.

Pick a metric you'd remove from the company dashboard. Why?

Goodhart's Law lensMetrics that become targets stop being good metrics.

What a strong answer covers

  1. Pick a real metric that gets gamed (e.g. signups, NPS, CSAT).
  2. Show how teams optimise for it in ways that hurt the user or the business.
  3. Propose what to replace it with (an outcome metric, not an output).
  4. How you'd manage the political cost of removing it.

How would you set the price for a new B2B SaaS feature?

Value-based pricing + Willingness-to-payCost-plus is for commodities. Anchor on the value created.

What a strong answer covers

  1. Quantify the value the feature creates for the customer (time saved, revenue lifted).
  2. Identify the buying unit (per seat, per workspace, per usage).
  3. Run a Van Westendorp / WTP study or use comparable pricing.
  4. Choose a pricing model: tiered, usage, add-on. Justify why.
  5. Guardrails: discount policy, grandfather logic, expansion path.

Behavioural questions

Tell me about a time you had a serious conflict with engineering. How did you resolve it?

STARSituation, Task, Action, Result. Be specific about *your* action.

What a strong answer covers

  1. Situation: the project, the team, the source of the disagreement.
  2. Task: what you owned and what was at stake.
  3. Action: the specific moves you made (1:1, written doc, escalation, compromise).
  4. Result: outcome with a metric or shipped artifact, plus the relationship after.

Describe a time you had to make a decision with incomplete information.

STAR + Decision FrameworkShow the framework you used to decide, not just the gut call.

What a strong answer covers

  1. Situation: the deadline / pressure / stakes.
  2. Task: the call you owned.
  3. Action: the framework (expected value, reversibility, opportunity cost) you applied.
  4. Result: what happened, and what you'd do with hindsight.

Tell me about a project that failed. What did you learn?

STAR-L (with Lessons)Own the failure cleanly. Avoid blaming the team or external factors.

What a strong answer covers

  1. Situation + your hypothesis going in.
  2. What you tried and where it broke (instrumented metric, not vibes).
  3. The decision you made when it became clear it was failing (kill, pivot, double down).
  4. Lessons: 2 specific, transferable insights — and how you've applied them since.

Tell me about a time you influenced a senior leader without authority.

STAR + Stakeholder MapInfluence is a sequence of small commitments, not one big pitch.

What a strong answer covers

  1. Situation: the decision you needed and why the leader was the gatekeeper.
  2. Task: what you wanted them to say yes to.
  3. Action: the prep (data, allies, framing in their language) and the sequence.
  4. Result: the decision and how the working relationship changed after.

Tell me about a time you had to deprioritise something the team really cared about.

STAR + Trade-off FramingHow you say no is as important as what you say no to.

What a strong answer covers

  1. Situation: the competing priorities and the constraint (people, time, strategy).
  2. Task: the call you owned.
  3. Action: the criteria you used and how you communicated to the team.
  4. Result: what got shipped, how the team felt, and what you'd revisit.

Estimation questions

How many Uber rides happen in London on a typical weekday?

Top-down DecompositionPopulation → addressable users → frequency → adjustments. Show your assumptions.

What a strong answer covers

  1. Population of London (~9M) and assumptions about smartphone / Uber adoption.
  2. Addressable rider base × average rides per rider per week.
  3. Convert to weekday and apply weekday vs weekend skew.
  4. Sanity check against a known anchor (e.g. London taxi industry size).
  5. State the answer with a confidence range.

Estimate the annual revenue of a mid-sized airport coffee shop.

Bottom-up: Traffic × Conversion × AOV × DaysBuild from one transaction up to the year.

What a strong answer covers

  1. Footfall past the shop per day (terminal traffic × % passing this gate).
  2. Conversion rate (passersby → buyers).
  3. Average order value (coffee + pastry attach rate).
  4. Operating days × seasonality.
  5. Cross-check against staff cost / margin sanity (if revenue too low to support the lease, recheck).

How much storage does Gmail need to support all its users for one year?

Users × Volume × CompressionBe explicit about per-user assumptions and where compression / dedup applies.

What a strong answer covers

  1. Total Gmail users (~1.8B) and active fraction.
  2. Average emails received per user per day × average size (with attachments).
  3. Convert to per-user storage per year.
  4. Apply dedup, compression, and tiering (cold vs hot) to get net storage.
  5. Add overhead (replicas, backups) and state the final number with units.

AI PM Specific questions

How would you decide whether to build, fine-tune, or buy an LLM for a feature?

Build / Fine-tune / Buy Decision TreeDriven by differentiation, latency, cost, and data moat.

What a strong answer covers

  1. What's the user job and what level of quality is 'good enough'?
  2. Is the differentiator the model or the workflow around the model?
  3. Latency, cost per request, and privacy constraints.
  4. Data: do you have proprietary data that would make fine-tuning win?
  5. Recommend: foundation API → RAG → fine-tune → train, only as needed.

How do you measure quality for a generative AI feature?

Eval Stack: Offline + Online + HumanNo single metric. Layer automated evals, online behaviour, and human review.

What a strong answer covers

  1. Offline: golden set, rubric-based scoring, regression suite per release.
  2. Online: implicit signals (regen, edit, abandon), explicit (thumbs).
  3. Human-in-the-loop: weekly sample review with a quality rubric.
  4. Guardrails: hallucination rate, safety, latency budget.
  5. How you'd ship a model upgrade safely (shadow → canary → ramp).

Your AI feature is hallucinating. How do you fix it?

Diagnose → Mitigate → MeasureHallucination has many root causes. Fix the right one.

What a strong answer covers

  1. Diagnose: prompt issue, retrieval gap, model capability, or stale context?
  2. Mitigate (in order of cost): prompt fix → RAG → tool use → fine-tune → model swap.
  3. Add UX guardrails: citations, confidence display, easy correction.
  4. Track hallucination rate as a first-class metric.
  5. Decide acceptable rate by use case (medical vs marketing copy ≠ same bar).

How would you price an AI feature with high variable inference cost?

Unit Economics + Tiered PricingInference cost shows up in COGS. Pricing must protect the margin.

What a strong answer covers

  1. Model the variable cost per request and cost per power user.
  2. Choose pricing shape: usage-based, tiered with caps, or freemium with upsell.
  3. Identify abuse vectors and design fair-use limits.
  4. Migration: how you handle legacy customers when costs shift.
  5. Long-run: as model costs fall, do you cut price or expand value?

Walk me through how you'd ship an agentic workflow safely to enterprise customers.

Agent Safety StackAgents take actions. Errors are no longer just bad text — they're bad outcomes.

What a strong answer covers

  1. Scope the agent: tasks, tool surface, blast radius if it goes wrong.
  2. Permissions: read vs write tools, human-in-the-loop checkpoints.
  3. Observability: full action traces, replay, audit logs.
  4. Reversibility: undo, sandboxes, dry-run mode for destructive actions.
  5. Rollout: design partners → opt-in beta → GA, with clear kill switch.

Preparing for an AI PM loop?

AI PM interviews add a layer these questions only partly cover: model choice, retrieval design, evals, and defensibility. The fastest way to build that instinct is to reverse-engineer real AI products across those lenses.