Product Metrics Interview Questions: How to Define, Diagnose and Decide
Five question types, a five-step playbook for a falling metric, and a worked funnel example. Judgment, not maths.
Metrics interviews test judgment, not arithmetic. There are five question types: define metrics, diagnose a drop, evaluate a launch, choose a North Star, and resolve trade-offs. For a falling metric, follow five steps: is it real, where is it, when did it start, what changed, then test the cause. Always pair a primary metric with a guardrail.
- The five types of metrics question
- Vocabulary you should use precisely
- How to answer "define the metrics"
- The metric-dropped playbook
- Evaluating a launch and choosing a North Star
- When two metrics disagree
- Formulas worth knowing cold
- Fifteen questions to practise
- Metrics for common kinds of product
- Setting a target when you have no baseline
- Experiment thinking in an interview
- Second worked example: evaluating a launch
- Mistakes to avoid
Metrics questions are where many candidates freeze, because they sound like maths but are really about judgment. Nobody expects you to compute anything hard. They want to see that you know which number matters, how it can mislead you, and what you would do when it moves. The good news: there are only five question types, and each has a repeatable approach.
The five types of metrics question
Vocabulary you should use precisely
| Term | Meaning | Example |
|---|---|---|
| Primary metric | The one number the change is meant to move | Week-1 activation rate |
| Guardrail metric | A number that must not get worse | Support tickets per user |
| Leading indicator | Moves early and predicts outcomes | Completed first lesson |
| Lagging indicator | Confirms outcomes later | Day-30 retention |
| Input metric | Something you can directly influence | Notifications delivered |
| Output metric | The result you care about | Weekly active users |
How to answer "define the metrics"
Say the goal first. Translate it into user behaviour ("a successful search ends in a click and no immediate re-search"). List a few candidate metrics, then choose one primary and one guardrail and explain why. Finish with a target and a time frame. Candidates who list ten metrics and never choose lose points for the same reason as those who list ten features and never choose.
The metric-dropped playbook
"Daily active users fell 10% this week. What do you do?" is the most common metrics question. The trap is guessing causes. The skill is narrowing before guessing.
1. Is it real? Rule out broken tracking, a delayed data pipeline, or a change in how the metric is counted. Interviewers love hearing this first. 2. Where is it? Break the drop down by platform, country, user type and channel. A drop concentrated in one segment is a very different problem from a uniform one. 3. When did it start? A sudden cliff points to a release or outage; a slow slide points to a product or market issue. 4. What changed? Releases, campaigns, competitor moves, seasonality, holidays. 5. Test the cause. Form one or two hypotheses and say how you would confirm them.
A worked example, with an illustrative funnel
Suppose new-user signups fell, and the funnel looks like this after the drop:
The funnel points at OTP verification, not at acquisition or the signup screen. Now the hypotheses become specific: an SMS provider issue, a change to the OTP screen, or a new delay on one telecom operator. Check delivery rates by operator, look at whether the drop started at a release, and compare Android versus iOS. That is a far better answer than "maybe users lost interest".
Evaluating a launch and choosing a North Star
To evaluate a launch, compare against a control group whenever possible, look at the primary metric and the guardrails together, and check whether the effect holds across segments and over time (novelty effects fade). For a North Star, pick the number that captures the value users get, not just activity. For a learning app, "lessons completed that led to an improvement" beats "time in app", because time can go up when users are confused. I wrote about exactly this trap in the metric we almost built the wrong feature for.
When two metrics disagree
Formulas worth knowing cold
| Metric | How it is calculated |
|---|---|
| Conversion rate | Users who completed the action ÷ users who could have |
| Retention (day N) | Users active on day N ÷ users who started on day 0 |
| Churn rate | Users lost in a period ÷ users at the start of the period |
| ARPU | Revenue in a period ÷ number of users |
| CAC | Total acquisition spend ÷ new customers acquired |
| LTV (simple) | Average revenue per customer per period × expected lifetime |
Fifteen questions to practise
- What metrics would you track for a new checkout flow?
- Daily active users dropped 10% overnight. Walk me through it.
- How would you measure the success of a search feature?
- What is the North Star metric for a music streaming app?
- An A/B test raised conversion but lowered average order value. What do you do?
- How would you know a notification is helping, not annoying?
- What metrics would you use for a marketplace?
- How would you measure the health of a community feature?
- Retention is flat but revenue is rising. Is that good?
- How would you set a target for a metric you have never measured?
- What is wrong with using time-in-app as a success metric?
- How would you measure customer support quality?
- Which metric would you show the CEO for a subscription business?
- Signups doubled after a campaign, but activation fell. Why?
- How would you measure the impact of a redesign with no A/B test?
Metrics for common kinds of product
Interviewers often ask for metrics for a product type rather than a feature. Having a starting set in your head saves you from blanking. These are starting points, not laws; always tie them back to the goal.
| Product type | North Star candidate | Typical guardrail |
|---|---|---|
| Marketplace | Successful transactions per week | Cancellation rate, complaints |
| Subscription app | Weekly active subscribers using the core feature | Refunds, support contacts |
| E-commerce | Orders delivered as promised | Return rate, late deliveries |
| Social or community | Meaningful interactions per active user | Reports, blocks, spam rate |
| Learning app | Lessons completed that lead to improvement | Content quality ratings |
| Fintech | Users completing a payment or transfer monthly | Failure rate, fraud rate |
Setting a target when you have no baseline
"How would you set a target for something you have never measured?" is a favourite. Do not invent a number. Say you would first measure for two to four weeks to find the baseline, look at how similar features or products perform, and set an initial target as a range. Then commit to revisiting it once real data arrives. This shows you know the difference between a guess and an estimate.
Experiment thinking in an interview
Three details signal experience. Sample size and duration are decided in advance, so you do not stop the moment the numbers look good. Novelty effects can inflate early results, so run long enough for behaviour to settle, at least a full week to cover weekday and weekend patterns. And a result is not automatically a cause if the groups were not randomised; say so when they were not.
Second worked example: evaluating a launch
Prompt: "We launched a redesigned search page. How do you know if it worked?" (Illustrative numbers.)
Goal: help users find what they want faster. Primary metric: searches that end in a click within the first page of results. Guardrail: searches followed by another search within thirty seconds, which usually means the first one failed, and page load time.
Suppose the click-through rate rose from 58% to 61% in a 50/50 test, but re-search within thirty seconds also rose from 12% to 14%. The primary metric improved and the guardrail worsened slightly. Averaging them would hide the story. The better move is to look at segments: perhaps the redesign helped experienced users who know what they want and hurt new users who reformulate more. Then decide: ship to the segment that gained, iterate for the rest.
The lesson repeats across nearly every metrics question: a single number rarely answers it, a pair does, and segments explain the pair.
Mistakes to avoid
Naming ten metrics and choosing none. Proposing vanity metrics such as raw downloads or page views. Guessing causes for a drop before segmenting. Forgetting guardrails. Quoting a benchmark retention number without knowing the category. And treating a correlation as proof of cause without saying how you would test it.
Continue with the 50 question bank and product sense interviews.
Frequently asked questions
What are product metrics interview questions?
Questions testing whether you can pick, interpret and act on metrics: defining success measures, diagnosing a drop, evaluating a launch, choosing a North Star and resolving conflicts between metrics.
How do I answer 'a metric dropped, what do you do?'
Check the data is real, segment to find where it dropped, find when it started, list what changed, and then form and test hypotheses. Narrow first, guess second.
What is a guardrail metric?
A metric that must not get worse when you change something, used to catch unintended harm. Examples include support tickets, refund rate and page load time.
Do I need to be good at maths for PM metrics interviews?
Basic arithmetic and percentages are enough. Interviewers care about judgment: which metric matters, how it can mislead, and what you would do next.
Want this sharpened for your specific situation?
I take on a small number of career calls and resume reviews each month.
See how we could work together