Data Analyst Mock Interview Guide
How to prepare for data analyst interviews: SQL edge cases, choosing metrics, spotting bad data and explaining results to people who don't read dashboards.
HotSeat team
Updated November 2026 3 min read
Your SQL can be perfect and you can still lose a data analyst interview. Analysts get hired to help someone make a decision. So alongside the query, interviewers want to know whether you'd spot bad data, pick a sensible metric and explain the result to someone who has never opened a dashboard.
A classic question brings all of that together, so let's start there.
In this article
‘A key metric fell 20% this week. What do you check first?’
Resist explaining why customers are leaving. First, find out whether the drop is real. Did tracking change? Has the metric's definition shifted? Is the latest data just incomplete? An alarming drop can easily turn out to be a broken event or a late-running pipeline.
If it's real, find where it started. Which day? Which platform, region or customer group? A sudden drop in one group points somewhere very different from a slow decline everywhere. Only then start forming a theory.
You'll probably be asked when you'd alert leadership. Say you'd give them an early heads-up if the impact looked serious, clearly labelled as unconfirmed, rather than going quiet until you have the perfect answer.
The SQL that actually trips people up
Writing a basic join is the easy part. Practise these until they're boring:
- Duplicate or missing keys: Know what happens to your row count when a join key isn't unique, and how to check before you trust the result.
- Grain: Are you counting users, sessions or orders? Many wrong answers come from aggregating at the wrong level.
- Window functions: Ranks, running totals and comparing a row to the previous one.
- Dates: Time zones and incomplete periods. This week's numbers will always look low on a Tuesday.
- Nulls: Know how they behave in comparisons, counts and averages.
- Debugging: Practise reading someone else's query and working out why the number is wrong. It's a common exercise, and a very realistic one.
‘How would you define success for a new feature?’
Start with what the feature is meant to change about how people behave, then pick one main measure for that. Add a guardrail or two, so you'd notice if you were winning on one number by damaging another. Say what time window you'd look at and what you'd compare against.
The follow-up tends to be something like ‘engagement went up but retention went down’. There's no single right answer. They want to hear you ask which users changed, and go back to what the feature was for.
Explaining a result to a non-analyst
Imagine a product manager asks what you found. Both answers are invented:
“The coefficient was statistically significant with a p-value below 0.05, and the model had an adjusted R-squared of 0.62.”
The numbers may be right, but the listener still doesn't know what to do.
“Customers who finished setup on their first day were more likely to come back the next week, even after we accounted for company size. That doesn't prove setup caused it, so I'd test a shorter setup flow before we commit to a full product change.”
A finding, its limit and a next step, in language anyone in the meeting can follow.
Stories to have ready
Prepare one analysis that changed a decision, including who disagreed and how you know it made a difference. If your work has only ever ended in a slide deck, say what you'd hoped would happen and what got in the way.
Also prepare a time the data was wrong and you caught it, or better, a time you didn't catch it at first. Telling that story well shows you've learned to check your work.
Practise turning analysis into a business conversation
Try SQL reasoning, metric choices and follow-up questions in a data analyst mock interview.
Start a data analyst interviewSources and further reading
- PostgreSQL SQL tutorial (PostgreSQL Global Development Group)
- Data analyst interview preparation guide (Harvard FAS Mignone Center for Career Success)
- Structured interviews (U.S. Office of Personnel Management)