Each question below includes what the interviewer is actually testing. That matters more than the answer itself — once you know what is being probed, you can answer from your own experience instead of reciting someone else’s script, which is obvious to everyone in the room.
Whether your work has consequences. Analysts who only produce dashboards nobody acts on are common and expensive.
Structure it as: the decision that was about to be made, what you found, what happened instead, and the resulting number. If the decision did not change, say so and explain what you learned about influence.
Diagnostic instinct. They want to see you rule out instrumentation before theorising about user behaviour.
Check the tracking first — most sudden cliffs are broken pipelines or a bad release, not real. Then segment by platform, geo, version, and cohort to find whether the drop is uniform or concentrated. Uniform means measurement; concentrated means real.
Whether you get defensive, and whether your work is reproducible.
Reconcile definitions before defending the number — most disputes are two correct calculations of two different metrics. Show the query, agree on the definition in writing, then re-run.
None of this matters if your resume is filtered out first. The most common cause is not a weak background — it is a document that screening software cannot read correctly, or one that lists duties where it should show results.
Check your resume against data analyst keywords or run a full scan. Both are free and take about three seconds.