10 role-specific questions
Data Analyst Interview Questions & Answers
Data analyst interviews test SQL, your approach to messy data and changing metrics, experimentation basics, visualisation, and — above all — turning numbers into decisions people act on.
Free, no sign-up. 65 questions in this role's practice pool, including behavioural, HR and situational questions.
Data analysis interview questions
SQL for analysis, metrics, data cleaning, visualisation, A/B tests and communicating insight.
1.
Write (or describe) a SQL query to find the top 5 customers by total spend last year.
BeginnerTechnicalPractise this question and get feedbackWhat the interviewer is evaluating
Core analytical SQL, and whether you check the data before trusting the answer.
Key points a strong answer covers
- Filter the period. Filter orders to last year.
- Aggregate. SUM the spend and GROUP BY customer.
- Order & limit. ORDER BY total DESC and LIMIT 5.
- Data quality. Consider refunds, cancelled orders, currencies and ties.
Suggested answer structure
- Define it in one sentence.
- Explain how it works underneath.
- Give a practical example from code or a project.
- Close with a trade-off, pitfall or when you'd use it.
Common mistakes
- Forgetting to exclude cancelled or refunded orders.
- Using a non-sargable date filter.
Likely follow-up questions
- How would you handle ties for fifth place?
- How would you show each customer's share of total revenue?
Example strong answer
I'd join or select from the orders table, filter to completed orders in the date range, group by customer and sum:
SELECT customer_id, SUM(amount) AS total_spend FROM orders WHERE status = 'completed' AND order_date >= '2025-01-01' AND order_date < '2026-01-01' GROUP BY customer_id ORDER BY total_spend DESC LIMIT 5;— then join to customers for names. Before trusting it I'd check a few things: whether refunds are separate records I need to subtract, whether amounts are in multiple currencies, and duplicates. A date range rather thanYEAR(order_date) = 2025lets the database use an index. For ties at fifth place I'd useDENSE_RANK()and filter on rank instead of LIMIT.2.
How do you clean and validate a messy dataset before analysis?
BeginnerRole knowledgePractise this question and get feedbackWhat the interviewer is evaluating
Rigour and data quality awareness.
Key points a strong answer covers
- Profile first. Profile the data: distributions, types, ranges.
- Missing values. Decide how to handle missing values.
- Duplicates & consistency. Remove duplicates and standardise formats.
- Outliers. Investigate outliers rather than deleting blindly.
- Document & reconcile. Document changes and reconcile totals with the source.
Suggested answer structure
- State your approach in one sentence.
- Walk through the steps or criteria you use.
- Give a real example from your work.
- Say how you measure whether it worked.
Common mistakes
- Deleting outliers without investigating.
- Not documenting cleaning steps.
Likely follow-up questions
- When is it right to remove an outlier?
- How would you make the cleaning reproducible?
Example strong answer
I start by profiling: row counts, column types, ranges, distributions, and how many values are missing or unique, so I understand what I have. Then I fix issues deliberately: standardise formats like dates, country names and casing; remove true duplicates after defining what a duplicate means; and decide on missing values case by case — exclude, fill or flag — depending on why they're missing. Outliers I investigate before touching: a £50,000 order may be a data-entry error or a real enterprise customer. I reconcile totals against the source system, and do all of it in code — SQL or Python — rather than by hand, so it's documented and reproducible.
3.
A key metric (for example daily sales) dropped 20% yesterday. How do you investigate?
IntermediateProblem solvingPractise this question and get feedbackWhat the interviewer is evaluating
Structured analytical thinking under pressure.
Key points a strong answer covers
- Check the data first. First check whether it's a real drop or a data/tracking issue.
- Context & seasonality. Compare with the same day last week/year; check holidays.
- Segment. Break down by channel, region, device, product.
- Internal & external events. Check releases, campaigns, outages and external events.
- Communicate. Share findings and next steps quickly.
Suggested answer structure
- Restate the problem and your assumptions.
- Break it into steps and reason out loud.
- Check your answer against an edge case.
- Summarise the result and what you'd verify next.
Common mistakes
- Assuming it's real without checking the data pipeline.
- Looking only at the total.
Likely follow-up questions
- How would you tell a tracking bug from a real drop?
- What if no single segment explains it?
Example strong answer
First I'd check it's real: did the data pipeline run fully, did tracking change in a release, are other data sources — the payment provider, the database — showing the same drop? Then context: compare with the same weekday last week and last year, and check for holidays or the end of a promotion. Next I segment: by channel, country, device, product category and new versus returning customers, to see whether the drop is everywhere or concentrated — say, only iOS app users, which points to a release. I'd check internal events like deployments, price changes and paused campaigns, and external ones like outages or a competitor sale. I'd share a quick update with what's ruled out, then the root cause.
4.
How would you design and analyse an A/B test?
IntermediateRole knowledgePractise this question and get feedbackWhat the interviewer is evaluating
Experimental rigour: planning, randomising and reading a test without fooling yourself.
Key points a strong answer covers
- Hypothesis & metric. A clear hypothesis, one primary metric and guardrail metrics.
- Sample size & duration. Calculate sample size up front and run full weeks.
- Randomisation. Random assignment to control and variant.
- Statistical significance. Check statistical significance and confidence intervals.
- Pitfalls. Avoid peeking, multiple comparisons; check sample ratio mismatch.
Suggested answer structure
- State your approach in one sentence.
- Walk through the steps or criteria you use.
- Give a real example from your work.
- Say how you measure whether it worked.
Common mistakes
- Stopping the test as soon as it looks significant.
- Testing many metrics and reporting whichever wins.
Likely follow-up questions
- What is sample ratio mismatch?
- What would you do with a statistically significant but tiny effect?
Example strong answer
I'd start with a hypothesis and one primary metric — 'a shorter checkout will increase purchase conversion' — plus guardrail metrics like average order value and refunds. Before launching I'd calculate the sample size needed for the smallest effect worth detecting at the usual 5% significance and 80% power, which sets the duration — rounded to full weeks to cover weekday patterns. Users are randomly assigned and stay in their group. During the test I don't peek and stop early when it looks good, since that inflates false positives. At the end I check the split matches the planned ratio, then compare conversion with a confidence interval and p-value, look at guardrails, and judge whether the effect is practically meaningful, not just significant.
5.
How do you choose the right chart, and what makes a good dashboard?
BeginnerRole knowledgePractise this question and get feedbackWhat the interviewer is evaluating
Communication of data.
Key points a strong answer covers
- Chart for the question. Line for trends, bar for comparisons, histogram for distributions, scatter for relationships.
- Avoid misleading charts. Avoid 3D, truncated axes and too many pie slices.
- Dashboard purpose & audience. Built for a specific audience and decisions.
- Hierarchy & simplicity. Few key metrics with context (targets, prior period) at the top.
Suggested answer structure
- State your approach in one sentence.
- Walk through the steps or criteria you use.
- Give a real example from your work.
- Say how you measure whether it worked.
Common mistakes
- Cramming 30 charts into one dashboard.
- Charts without context like targets or previous periods.
Likely follow-up questions
- When is a table better than a chart?
- How do you know if a dashboard is actually used?
Example strong answer
I start from the question. Trends over time get a line chart; comparing categories, a bar chart sorted by value; distributions, a histogram or box plot; relationships, a scatter plot; and part-to-whole only with very few parts — usually a stacked bar beats a pie. I avoid 3D, dual axes and truncated axes that exaggerate differences, and I label directly instead of relying on legends. A good dashboard is built for a specific audience and the decisions they make: a handful of key metrics at the top with context — target, previous period, trend — then drill-downs below. If a number can't change anyone's decision, it doesn't belong there. I check usage logs to see if dashboards are actually used.
6.
How do you present findings to non-technical stakeholders?
IntermediateRole knowledgePractise this question and get feedbackWhat the interviewer is evaluating
Turning analysis into decisions.
Key points a strong answer covers
- Lead with the answer. Lead with the answer and recommendation.
- Simple visuals & language. Plain language and one clear chart per point.
- Business impact. Translate into business impact.
- Uncertainty & caveats. Be honest about uncertainty and limitations.
- Next steps. End with clear next steps.
Suggested answer structure
- State your approach in one sentence.
- Walk through the steps or criteria you use.
- Give a real example from your work.
- Say how you measure whether it worked.
Common mistakes
- Walking through the methodology before the conclusion.
- Hiding uncertainty.
Likely follow-up questions
- How do you handle a stakeholder who rejects the data?
- How do you explain statistical significance simply?
Example strong answer
I lead with the answer and what I recommend — 'Customers who use the mobile app churn half as often, so I recommend we prioritise app onboarding' — then show the two or three pieces of evidence that support it, each with one simple chart and plain language. I translate findings into impact they care about: revenue, cost, customer numbers. I'm upfront about confidence and limitations — 'this shows correlation; the app users may simply be more engaged already' — because credibility matters more than a neat story. Methodology goes in an appendix for anyone who wants it. And I finish with clear next steps and owners.
7.
Which tools do you use for analysis — Excel, SQL, Python, BI tools — and when?
BeginnerTechnicalPractise this question and get feedbackWhat the interviewer is evaluating
Tool judgement: picking the right tool for the size and repeatability of the job.
Key points a strong answer covers
- SQL for extraction. SQL to extract and aggregate from databases.
- Excel / Sheets. Excel/Sheets for quick, small, ad-hoc analysis.
- Python / R. Python (pandas) or R for large, repeatable or statistical analysis.
- BI tools. Power BI/Tableau/Looker for shared dashboards.
Suggested answer structure
- Define it in one sentence.
- Explain how it works underneath.
- Give a practical example from code or a project.
- Close with a trade-off, pitfall or when you'd use it.
Common mistakes
- Claiming one tool for everything.
- Huge, fragile spreadsheets for recurring reports.
Likely follow-up questions
- When would you move a recurring Excel report into a BI tool?
- Which pandas operations do you use most?
Example strong answer
I use the tool that fits the job. SQL is my starting point for almost everything because the data lives in a database or warehouse, and aggregating there is efficient. Excel or Google Sheets are great for quick, small analyses, pivot tables and sharing something a stakeholder will tweak. When the data is large, the analysis will be repeated, or I need statistics — cohort analysis, regression, cleaning messy files — I use Python with pandas in a notebook so it's reproducible. Anything people need to monitor regularly goes into a BI tool like Power BI, Tableau or Looker, built on a clean data model, so nobody re-runs a fragile spreadsheet every Monday.
8.
How would you choose KPIs for a new subscription product?
IntermediateRole knowledgePractise this question and get feedbackWhat the interviewer is evaluating
Business acumen: choosing a few metrics that reflect value and drive decisions.
Key points a strong answer covers
- Tie to goals. Tie KPIs to business goals; pick a north-star metric.
- Acquisition & activation. Acquisition, activation and trial conversion.
- Retention & churn. Retention and churn by cohort.
- Revenue. MRR, ARPU, LTV, CAC payback.
- Leading vs lagging / few. A few actionable metrics, leading and lagging, not vanity metrics.
Suggested answer structure
- State your approach in one sentence.
- Walk through the steps or criteria you use.
- Give a real example from your work.
- Say how you measure whether it worked.
Common mistakes
- Choosing vanity metrics like total sign-ups.
- Too many KPIs.
Likely follow-up questions
- What's the difference between a leading and a lagging indicator?
- How would you calculate customer lifetime value?
Example strong answer
I'd start from the business goals and the customer journey, then pick a small set across it. A north-star metric reflecting real value — say, weekly active subscribers who used the core feature. Acquisition: sign-ups and cost per acquisition by channel. Activation: the share of new users who reach the 'aha' moment, which is a leading indicator of retention. Conversion from trial to paid. Retention and churn, always looked at by cohort, since averages hide trends. Revenue: MRR, ARPU, lifetime value and CAC payback. I'd avoid vanity metrics like total cumulative sign-ups, keep each KPI clearly defined in one place, and make sure someone owns each.
9.
What's the difference between correlation and causation? Give an example.
IntermediateRole knowledgePractise this question and get feedbackWhat the interviewer is evaluating
Statistical reasoning.
Key points a strong answer covers
- Definition. Correlation means two variables move together.
- Causation. Causation means one actually changes the other.
- Confounders. Confounders or reverse causation can explain correlations.
- Example. A concrete example.
- Establish causation. Experiments (randomised tests) establish causation.
Suggested answer structure
- State your approach in one sentence.
- Walk through the steps or criteria you use.
- Give a real example from your work.
- Say how you measure whether it worked.
Common mistakes
- Recommending a change based purely on correlation.
- No concrete example.
Likely follow-up questions
- How could you test causation without an A/B test?
- What is Simpson's paradox?
Example strong answer
Correlation means two things tend to move together; causation means one actually makes the other change. A classic example: ice cream sales and drowning incidents both rise in summer — hot weather drives both, so it's a confounder. At work it's often subtler: users who enable notifications retain better, but it may be that already-engaged users turn on notifications, not that notifications create engagement — reverse causation. So before recommending 'push everyone to enable notifications', I'd test it: randomly prompt half of new users and compare retention. When experiments aren't possible, methods like difference-in-differences or matching help, but with more caveats.
10.
A manager asks you for 'all the data on customers'. How do you respond?
IntermediateSituationalPractise this question and get feedbackWhat the interviewer is evaluating
Scoping requests and understanding the real question.
Key points a strong answer covers
- Ask the underlying question. Ask what decision or question they're trying to answer.
- Scope. Agree a specific scope, period and definitions.
- Privacy. Consider privacy and access rules for personal data.
- Deliver iteratively. Deliver a first cut quickly and refine.
Suggested answer structure
- Clarify the situation and what's at stake.
- Say what you'd do first, and why.
- Explain who you'd involve and how you'd communicate.
- Describe how you'd check the outcome and prevent a repeat.
Common mistakes
- Exporting everything into a giant spreadsheet.
- Ignoring data protection.
Likely follow-up questions
- What if they insist on the full export?
- How do you prioritise many such requests?
Example strong answer
I'd ask what they want to decide or understand, because 'all the data' usually means a specific question — for example, 'why are our renewals falling?' Then I'd agree the scope: which customers, which period, which metrics and how they're defined. I'd also check privacy: a full export of personal data probably isn't needed and may not be allowed under GDPR or our policies, so aggregated or anonymised data is usually better. Then I'd deliver a quick first version — a summary of renewal rates by segment, say — and check it answers their question before investing more time.
Ready to answer them out loud?
Run a timed data analyst mock interview, type or speak your answers, and get a score breakdown with what to improve.