Interview Basecamp
Interview questions
Data analyst interviews test whether people can trust your number and whether you can tell them what to do with it. You will be asked to define a metric before you query, to explain a drop without blaming the pipeline by reflex, and to refuse a cut of the data that flatters a decision already made. The questions below are the ones that separate a person who pulls charts from a person who makes the chart decision-ready. Role prep for a data analyst builds a prep course of Learning Blocks on those competencies, from grain and cohorts to how you brief a stakeholder.
Show how you handled pressure to make the number say something it did not.
What interviewers look for
You kept the finding and changed the conversation, not the data. Interviewers want the claim you stood behind, how you checked you were not wrong, and how you helped them update the plan. Rewriting the chart until it fit, or going around them to embarrass them, are both failures.
What interviewers look for
A specific bug: a join, a filter, a timezone, a definition. They want the check that caught it, who you told, and the habit you changed so the next number is reproducible. A story with no concrete error is usually polished past usefulness.
What interviewers look for
You narrowed the question before you opened a query tool. Who is the decision-maker, what will they do differently, and by when. Interviewers like a metric tied to that action and a recommendation with a condition. A dashboard with twelve tiles and no ask does not answer this.
What interviewers look for
You can name the behavior the metric rewarded and the behavior you actually wanted. Strong answers propose a replacement or a paired guardrail and say who was upset. Interviewers are unconvinced by candidates who only complain that the metric is bad.
What interviewers look for
A certified definition, a documented filter, and a person who could rerun it. They listen for what you refused to automate until the definition was stable, and how you checked their first independent use. Heroics where every question still comes to you are the opposite of the point.
Definitions, grain, and the failure modes that make a confident chart wrong.
What interviewers look for
One row means one thing you can say out loud: one account-day, one order, one user. They want the population exclusions and the timezone or cutoff. Candidates who start with SELECT and discover the grain later tend to double-count.
What interviewers look for
You separate a tracking break from a real change before you brief anyone. Check volume of events, a recent release, a filter, and whether the drop is in one segment. Interviewers want the order: instrumentation, then mix, then behavior. Jumping to a product story first is a miss.
What interviewers look for
When mix or tenure would make the period comparison lie: onboarding, retention, or a changing acquisition source. They want an example where period-over-period looked healthy while a cohort got worse, or the reverse. If you cannot say what question each view answers, the choice is habit.
What interviewers look for
An explicit rule for each: nulls excluded or bucketed, a restatement window you document, and a primary key you dedupe on. Interviewers want the number to match if someone reruns it next week. Hand-waving that the warehouse is messy is not a method.
What interviewers look for
You talk about the decision, the size of the effect they need, and how much noise three days contains. You do not need a statistics lecture. They want a plain statement of what the test can and cannot support, and a recommendation to wait, ship, or instrument a better measure.
The ask arrives before the data is ready. Show how you answer without inventing certainty.
What interviewers look for
A status with a date, the leading signals you do have, and a refusal to call a winner. Interviewers like a short note that says what would count as success and when you will know. A decorative chart that implies a conclusion you do not have is worse than waiting.
What interviewers look for
You stop the debate about which chart is right and name the definition each one uses. They want one owner, one definition for the decision in the room, and a follow-up to reconcile the pipelines. Picking a side to look decisive, without knowing the grain, is reckless.
What interviewers look for
You say what the history can support and where the change breaks it. A range, the assumption behind it, and the input you need from the people who know the change. Interviewers do not want a single precise number produced to end the conversation.
What interviewers look for
You name the search: many cuts will look special by chance. Strong candidates ask for the slice that was suspected beforehand, or treat the search as exploration that needs a follow-up. Quietly delivering the flattering cut is an integrity failure, not a favor.
What interviewers look for
You can offer a primary number if you also say what it hides. Interviewers want the definition, the population, and one companion metric so a good headline cannot cover a broken segment. Inventing a blended score with no owner is how health scores rot.