Autocoder

Product Analyst Interview Questions and Answers

Last updated 2026-11-19

Product analyst interviews check whether you can define the right metrics for a product, find out why a metric moved and judge whether a feature worked. They combine product sense (what should we measure), SQL (retention, funnels, active users) and statistics (experiments). Case questions are typically open-ended, so structure matters as much as the numbers.

The answers below include the SQL and the arithmetic for the standard metrics. The experiment side has its own page, A/B testing interview questions, and the general analyst preparation is in how to crack the data analyst interview.

Typical roundsSQL test, product case, metrics and experiments, behavioural
Core skillsSQL, metrics design, experimentation, product sense
Questions here14 worked questions

Metrics vocabulary

A framework for metric-drop cases

First verify the data (tracking change, delayed pipeline). Then segment: by platform, geography, acquisition channel, new versus existing users and product area. Decompose the metric into drivers, such as active users times sessions per user times conversion per session. Check releases, campaigns, seasonality and external events. State the most likely cause with your confidence, and the next analysis that would confirm it.

Product Analyst Interview Questions and Answers: questions with worked answers

  1. 1

    How would you choose a north star metric for a food delivery app?

    medium

    The metric should reflect value delivered to customers and predict long-term revenue. For food delivery, weekly successful orders per active customer, or orders delivered on time, captures that customers got a meal as promised. Order count alone can be gamed by discounts, so pair it with guardrails: delivery time, order cancellation rate, customer ratings, contribution margin and repeat rate. Avoid vanity metrics such as app downloads, which do not show whether anyone is getting value.

  2. 2

    DAU is 200,000 and MAU is 1,000,000. What is stickiness, and how do you interpret it?

    easy

    Stickiness = DAU / MAU = 200,000 / 1,000,000 = 20 percent, so the average monthly user is active about 6 days a month (20 percent of 30). Whether that is good depends on the product: a messaging app may target above 50 percent, while a travel booking app is naturally much lower. Track the trend and segment by cohort, since a rise in new low-intent users can reduce stickiness even if existing users are unchanged.

  3. 3

    Write SQL for daily, weekly and monthly active users.

    easy

    Count distinct users per period from an events table. Define "active" clearly, for example any qualifying event, because including passive events like push opens inflates the number. For rolling 7-day actives, join each date to the previous six days. The query below gives DAU and a rolling WAU using a self-join on a user-date table.

    sql
    WITH ud AS (SELECT DISTINCT user_id, event_date FROM events)
    SELECT d.event_date,
           COUNT(DISTINCT d.user_id) AS dau,
           (SELECT COUNT(DISTINCT w.user_id) FROM ud w
             WHERE w.event_date BETWEEN d.event_date - 6 AND d.event_date) AS wau_rolling
    FROM ud d
    GROUP BY d.event_date
    ORDER BY d.event_date;
  4. 4

    Compute day-1 retention by signup cohort in SQL.

    medium

    Find each user's first active day, which defines the cohort, then left-join to events on the next day. Retention for a cohort is the number of users who returned on day 1 divided by the cohort size. Use a left join so that cohorts with no returners still show zero. The date arithmetic is PostgreSQL style. Generalise to day N by changing the offset, and be careful to exclude cohorts too recent to have a day-N observation.

    sql
    WITH first_day AS (
      SELECT user_id, MIN(event_date) AS d0 FROM events GROUP BY user_id
    )
    SELECT f.d0 AS cohort_date,
           COUNT(DISTINCT f.user_id) AS cohort_size,
           COUNT(DISTINCT e.user_id) AS returned_d1,
           ROUND(1.0 * COUNT(DISTINCT e.user_id) / COUNT(DISTINCT f.user_id), 3) AS d1_retention
    FROM first_day f
    LEFT JOIN events e ON e.user_id = f.user_id AND e.event_date = f.d0 + 1
    GROUP BY f.d0
    ORDER BY f.d0;
  5. 5

    Build a funnel in SQL: view item, add to cart, purchase.

    medium

    Use conditional distinct counts per step and compute step-to-step conversion. The simple version counts users who did each event at least once, which can include users who bought without a recorded view; for a strict funnel, join on timestamps so each step happens after the previous one. With 10,000 viewers, 1,500 carts and 600 purchases, view to cart is 15 percent, cart to purchase is 40 percent and overall is 6 percent.

    sql
    SELECT COUNT(DISTINCT CASE WHEN event_name = 'view_item'   THEN user_id END) AS viewed,
           COUNT(DISTINCT CASE WHEN event_name = 'add_to_cart' THEN user_id END) AS carted,
           COUNT(DISTINCT CASE WHEN event_name = 'purchase'    THEN user_id END) AS purchased
    FROM events
    WHERE event_date BETWEEN DATE '2026-11-01' AND DATE '2026-11-07';
  6. 6

    Monthly churn is 5 percent. What fraction of customers remain after a year, and what is the expected lifetime?

    medium

    If churn is a constant 5 percent a month, the retained fraction after 12 months is 0.95^12 = 0.54, so 54 percent remain and 46 percent churned, not 60 percent as the naive 12 x 5 would suggest. The expected customer lifetime is 1 / churn = 1 / 0.05 = 20 months. The constant-churn assumption is rarely true, since early churn is higher and then flattens, so use cohort retention curves for decisions.

  7. 7

    Compute LTV and LTV to CAC: ARPU is 200 per month, gross margin 60 percent, monthly churn 5 percent, CAC 800.

    medium

    A simple LTV = ARPU x gross margin / churn = 200 x 0.6 / 0.05 = 2,400. LTV to CAC = 2,400 / 800 = 3, which is often cited as a healthy minimum. The payback period is CAC divided by monthly gross profit per customer, 800 / 120 = 6.7 months. These formulas assume constant churn and no discounting, so test the sensitivity: if churn is 8 percent LTV falls to 1,500 and the ratio to 1.9.

  8. 8

    Daily active users dropped 10 percent over a week. How do you investigate?

    hard

    Verify the metric: pipeline delay, a logging change or a bot filter. Check whether the drop is a level change or a trend, and compare with the same weekday last week and last year. Then segment by platform and app version, country, acquisition channel and new versus returning users: a drop confined to one Android version points to a release bug, while new-user-only points to acquisition. Decompose into new users, retained users and resurrected users. Look at releases, outages, campaigns and external events. Conclude with a ranked hypothesis list and the single query that would test the top one.

  9. 9

    How would you evaluate a feature launch when you cannot run an A/B test?

    hard

    Use a quasi-experiment. Difference-in-differences compares the change in a metric for users or regions that got the feature with the change for similar ones that did not, assuming they would have trended in parallel without the feature; check parallel trends in the pre-period. Alternatives are matching or propensity-score methods, regression discontinuity if rollout depended on a threshold, and synthetic control for region-level launches. Be open about the assumptions, because without randomisation, selection effects (keen users adopt first) can bias the result.

  10. 10

    Simpson's paradox in product data: how do you avoid being misled?

    hard

    An aggregate trend can reverse inside every segment when the segment mix changes. Example: the new design converts 40 percent on desktop and 21.7 percent on mobile, beating the old design at 30 percent and 5 percent, yet overall it is 24.3 percent versus 26.4 percent, because the new group had mostly mobile traffic. Always check segment-level results for key dimensions such as platform, region and new versus returning users, and make sure experiment arms have similar mix by design.

  11. 11

    How do you decide which metrics to track for a new feature?

    medium

    Write the feature's intended behaviour change: who should do what differently, and why does it matter. Then pick one primary success metric tied to that change (for a new search filter: share of searches leading to a click), supporting diagnostic metrics (filter usage rate, time to first click), and guardrails (search latency, overall conversion). Define each metric precisely with denominator and time window, and instrument the events before launch so you can read results on day one.

  12. 12

    How do you define an active user, and why does the definition matter?

    medium

    An active user should be someone who performed an action that reflects real value, not just opening the app or receiving a notification. For a banking app that might be a login plus a transaction or balance check, for a video app watching at least a minute. A loose definition inflates the number and hides the decline in real engagement, while a strict one may exclude valid light use. Keep the definition stable over time, document it, and report both a broad and a core active metric when they tell different stories.

  13. 13

    Estimate the number of orders a food delivery service handles per day in a city of 5 million people. Show the structure.

    medium

    State assumptions and compute. Suppose 30 percent of the population are potential users: 1.5 million. Say 20 percent of those order in a typical week, 300,000 users, ordering about 1.5 times a week, giving 450,000 orders a week, or about 64,000 a day. Adjust for weekends being higher, and then sanity check against a different route: restaurants times orders per restaurant. The point is a transparent chain of assumptions that the interviewer can challenge, not a precise number. Say which assumption you are least sure of.

Frequently asked questions

How is a product analyst different from a data analyst?

A product analyst works closely with a product team on user behaviour, experiments and feature impact, so product sense and metric design matter more. The tooling overlaps heavily.

How much SQL is expected?

A lot: window functions, cohorts, funnels, retention and deduplication. Practise these on event-style tables.

Will I get a product case?

Very likely: improve a feature, diagnose a metric drop, or define success for a launch. Use a clear structure and tie each idea to a metric.

Do I need statistics?

Yes, mainly experiment design, confidence intervals and common pitfalls. The A/B testing and statistics pages cover them.

Related guides

More in Role-wise Interview Questions by Job and Tech Stack

See all in this section