Data Analyst Interview Questions and How to Answer Them
A data analyst interview isn't really a SQL quiz. It tests if you can take a vague business question, find the right data, catch what's wrong with it, and explain the answer to someone who doesn't care how you got there. Here are the questions you'll most likely face, why they get asked, and what separates a hire from a polite rejection.
What happens in each round
- 1
Recruiter screen
What happens
If your tools match the stack (SQL dialect, Excel, Tableau or Power BI, maybe Python), if you've worked with real business data, and if you can describe a past project in plain language in under two minutes.
- 2
Technical screen
What happens
Live SQL or a take-home. They're watching for correct joins, sensible handling of nulls and duplicates, and if you sanity-check results before calling them done.
- 3
Hiring manager
What happens
How you scope a fuzzy request, how you handle a stakeholder who wants a specific answer, and if you think about the decision behind the number, not just the number.
- 4
Panel or case
What happens
A business case or dataset walkthrough. You'll pick metrics, build a quick analysis, and present a recommendation. They care about structure and clarity more than a perfect chart.
Questions you're likely to get
1.Walk me through the difference between an INNER JOIN and a LEFT JOIN, and tell me about a time the wrong one burned you.
Why they ask
Join mistakes are the most common source of wrong numbers in a dashboard. They want to know you've felt that pain and learned from it.
How to answer
- Define both in one sentence each: inner keeps matches only, left keeps every row from the left table and fills gaps with nulls.
- Give a real example, like a customer list joined to orders where an inner join quietly dropped customers who never bought.
- Explain how you caught it, such as comparing row counts before and after the join.
- Mention the fan-out problem, where a one-to-many join duplicates rows and inflates sums.
2.How would you find the second-highest order value for each customer?
Why they ask
It tests window functions, which separate analysts who can handle real reporting from those who stop at GROUP BY.
How to answer
- Use ROW_NUMBER, RANK or DENSE_RANK partitioned by customer and ordered by order value descending.
- Say out loud which ranking function you'd pick and why ties matter.
- Wrap it in a CTE and filter where the rank equals two.
- Call out customers with only one order and how they'd show up (or not).
3.You pull a weekly revenue number and it doesn't match what finance reports. What do you do?
Why they ask
Mismatched numbers happen every week in this job. They want to see a calm, methodical process instead of guessing.
How to answer
- Start by asking finance for their definition: gross or net, refunds included, which timezone, booked or recognized.
- Compare at a lower grain, like day by day or order by order, to find where the gap starts.
- Check filters, test accounts, currency conversion and late-arriving records.
- Document the agreed definition so the argument doesn't come back next month.
4.A product manager asks you, 'Why did signups drop last week?' How do you approach it?
Why they ask
This is the core of the job: turning a vague question into a structured investigation.
How to answer
- First confirm the drop is real, not a tracking bug, a broken event or a holiday week.
- Segment it: by channel, device, country, landing page, new vs. returning traffic.
- Check for known changes like a release, a pricing test or a paused ad campaign.
- Report what you found, what you ruled out and what you still don't know.
5.How do you decide which chart to use for a given piece of data?
Why they ask
A lot of analyst work gets judged by people who only see the chart. Bad visuals kill good analysis.
How to answer
- Start from the question: trend over time gets a line, comparison across categories gets a bar, part of a whole is usually still a bar.
- Explain why you avoid pie charts with many slices and dual-axis charts that mislead.
- Mention sorting bars, labeling directly and cutting gridlines.
- Say you put the takeaway in the chart title, not just the metric name.
6.Tell me how you'd clean a dataset you've just been handed from a system you don't know.
Why they ask
Most of the job is cleaning. They want to see habits, not heroics.
How to answer
- Profile it first: row counts, distinct values, null counts, min and max for each column.
- Find the primary key and confirm it's actually unique.
- Look for duplicates, impossible values, mixed formats and odd default values like zero dates.
- Talk to whoever owns the source system before making assumptions, and log every change you make.
7.How would you design an A/B test for a new checkout button, and how would you know if it worked?
Why they ask
Many analyst roles support experiments. They want basic statistical judgment, not a lecture.
How to answer
- Pick one primary metric up front, like checkout completion, plus a guardrail like refund rate.
- Explain random assignment and why you set sample size and test length before starting.
- Warn against peeking at results daily and stopping when it looks good.
- Say you'd check for novelty effects and segment differences before recommending a rollout.
8.Which metrics would you put on a dashboard for a subscription business, and why?
Why they ask
They want to see you connect metrics to decisions, not list every number you can compute.
How to answer
- Name a few that matter: new subscribers, churn, retention by cohort, revenue per user.
- Explain who looks at each one and what they'd do if it moved.
- Separate headline metrics from diagnostic ones people drill into.
- Mention that you'd agree on definitions before building anything.
9.Tell me about an analysis you did that changed a decision.
Why they ask
Output that nobody acts on is wasted work. They want proof you've had real influence.
How to answer
- Set up the decision on the table and who owned it.
- Explain what you looked at and one thing that surprised you.
- Describe how you presented it, including what you left out.
- Share what actually happened afterward, even if it was only partly what you recommended.
10.A senior stakeholder wants the data to show their project succeeded, and it didn't. How do you handle it?
Why they ask
Pressure to shade numbers is real. They're testing your integrity and your tact at the same time.
How to answer
- Say plainly you won't change the finding, but you'll check your work twice first.
- Meet with them one to one before any big meeting so they aren't blindsided.
- Offer what did work, or what the data suggests they try next.
- Loop in your manager if the pressure keeps coming.
11.You've got three urgent requests from three teams and a day to deliver. How do you prioritize?
Why they ask
Analysts often serve many masters. They want to see you manage demand, not just absorb it.
How to answer
- Ask each requester what decision it feeds and when that decision happens.
- Offer faster, rougher versions where a rough answer is enough.
- Be open about the tradeoff with your manager instead of silently working late.
- Mention a shared request queue or intake form if you've used one.
12.How do you explain a finding to someone who isn't technical?
Why they ask
Most of your audience won't read SQL. Communication is half the role.
How to answer
- Lead with the answer and what it means for them, then the evidence.
- Use one clear chart instead of five.
- Replace jargon with plain words, like 'people who came back' instead of 'retained cohort'.
- State your confidence and the main caveat in one sentence.
Mistakes that sink good candidates
Writing SQL in the live screen without talking through your approach or checking the output
Presenting a take-home result with no caveats, as if the data were perfect
Talking only about tools and never about the business decision your work supported
Blaming bad data or difficult stakeholders for every project that went sideways
Need more Data Analyst interviews to prep for?
HeroApply applies to Data Analyst jobs that match you, every day. 2,633 jobs are open today.