Data Analyst Interview Questions, by Round

Candidates often prepare only SQL and get surprised by the case round, which is where offers are frequently decided. This section maps the whole loop so your preparation matches it. Analyst interviews reuse five round types, and each scores a different thing.

• SQL. Live query writing or a take-home. The most common technical filter.
• Statistics and probability. Conceptual questions about what a result means, not proofs.
• Business case and metrics. A vague business question you turn into a measurable one.
• Spreadsheets and visualization. Excel or Google Sheets functions, and chart choice for an audience.
• Behavioral. Stakeholder conflict, being asked for a number you cannot yet support, and how you handle being wrong.

The dividing line between candidates who pass and candidates who do not is usually communication rather than technique. An analyst who writes a correct query in silence scores below one who says what they are checking and why.

SQL

• Find the second highest salary in a table.
• Write a query returning each customer's most recent order.
• What is the difference between an inner join and a left join, and when does the choice change your row count?
• Count users who placed an order in January but not in February.
• Calculate a running total of daily revenue.
• Rank products by revenue within each category.
• Find duplicate rows and write the query that removes them.
• Compute month over month growth as a percentage.
• What does HAVING do that WHERE cannot?
• How do NULL values affect a COUNT, a join, and an average?

Window functions and the NULL question separate candidates more than anything else on that list. Saying "a left join here will keep customers with no orders, and their totals become NULL rather than zero" out loud is exactly the signal interviewers are listening for.

Statistics and probability

• Explain p-value to someone who has not studied statistics.
• What is the difference between correlation and causation in a case you have seen?
• When would you use median instead of mean?
• What is a confidence interval telling you?
• How would you check whether an A/B test result is trustworthy?
• What is selection bias, and how might it appear in our data?

These rounds test whether you can be trusted with an interpretation. Precise plain language beats textbook phrasing, and admitting the limit of what a result supports is scored positively.

The case round hands you an underspecified business question. The work is converting it into something you could actually query.

• Signups are up but active users are flat. How do you investigate?
• How would you measure whether a new feature is successful?
• Revenue fell 10 percent last month. Walk me through your diagnosis.
• Which metric would you put on the executive dashboard, and what would you leave off?
• How would you tell whether a marketing channel is actually working?
• A stakeholder wants a number by end of day and the data is incomplete. What do you deliver?
• How would you define an active user for this product?

Spreadsheets and visualization

• Which function would you use to combine two tables on a shared key, and what breaks when keys do not match?
• Build a pivot that shows revenue by region and month. What do you check first?
• Which chart would you use for a distribution, and which for a trend?
• You have eight categories and one dominates. How do you present that honestly?

Behavioral

• Tell me about an analysis that changed a decision.
• Describe a time your number turned out to be wrong. What happened next?
• How did you handle a stakeholder who wanted a different answer than the data gave?

A junior analyst interviewing at a subscription company was asked why revenue fell while signups rose. Instead of guessing, she listed what she would segment by, plan tier, region, and new versus renewing, and said she would check whether a pricing change had shipped that month. The interviewer told her the check for a shipped change was the part most candidates skipped.

Reading questions builds recognition. Interviews test production under time pressure, and those are different skills.

• Write SQL by hand, out loud. Live rounds often use a plain editor with no autocomplete and no way to run the query. Practicing in that mode surfaces the syntax you only half know.
• Narrate while you query. Say what you are joining and what you expect the row count to do. Interviewers score reasoning they can hear.
• Practice the stakeholder translation. After each case, explain the result in three sentences with no jargon. That explanation is a graded round at most companies.
• Rehearse against follow-ups. The second question, "what if the data is incomplete", is where cases are actually won.

Running these prompts against an interviewer that pushes back is the closest simulation available; the mock interview tool will run analyst-style rounds with follow-ups. More role and topic sets live in the interview questions hub.

One boundary: take-home SQL tests and proctored assessment platforms are outside what a live assistant covers, and screen-shared query rounds show whatever is on your display. In a live video conversation, SubcueAI runs as a local desktop app or a Chromium browser Side Panel with no meeting bot joining the call. The detectability hub has the full picture.