Data Analyst Interview Questions (2026) and How to Answer

Behavioural questions (1–10)

How you've handled real situations. Answer each with one true story, in STAR order.

1. 1.Tell me about an analysis that led to a business change.

   What they're really asking: Impact beyond reporting: did someone act on your work?

   Answer outline (STAR)

   • SituationA business question from sales, marketing or operations.
   • TaskFind an answer they could act on.
   • ActionPull and clean the data, test the obvious explanations, present one clear recommendation.
   • ResultThe decision taken, and the metric that moved.

   Full sample answer (fictional example)

   Customer support noticed that many tickets came from new customers. I pulled six months of ticket data, joined it to account start dates, and found that 38 percent of first-month tickets were about one setup step. I wrote a one-page summary with a chart and a recommendation: add an in-app guide to that step. Product built it in three weeks. Tickets about that step fell 60 percent, and first-month tickets overall fell 14 percent. Support estimated it saved about 120 agent-hours a month.

2. 2.Tell me about a time a stakeholder challenged one of your numbers.

   What they're really asking: Can you defend your work with evidence, or admit an error quickly?

   Answer outline (STAR)

   • SituationA senior person who believed your figure was wrong.
   • TaskSettle it with facts and keep the relationship.
   • ActionWalk through the definitions and filters together; check against a second source.
   • ResultWho was right, and the metric definition you then documented.

   Full sample answer (fictional example)

   A regional sales director said my conversion number for his team was too low. Rather than defend it, I asked him to walk me through his figure. His number counted leads that came back after 90 days, while my definition cut off at 30. Both were reasonable. I showed both numbers on one page with the rules beside them, and checked against the CRM export as a second source. We agreed to report both and name them differently. I added the definitions to our metrics page, and he now uses the written version in his meetings.

3. 3.Describe a dashboard you built that people actually used.

   What they're really asking: Do you design for decisions, not decoration?

   Answer outline (STAR)

   • SituationA team making decisions from scattered spreadsheets.
   • TaskGive them one trusted view.
   • ActionAsk users which decisions they make, pick a few metrics, build, then iterate on usage.
   • ResultWeekly users, and the reports or meetings it replaced.

   Full sample answer (fictional example)

   Sales managers kept their pipeline in four different spreadsheets. I interviewed six managers and asked which decisions they made each week; most said where to spend coaching time and which deals were at risk. I built one dashboard with five metrics: pipeline by stage, deals stalled over 14 days, win rate by rep, forecast against target and new leads. After launch I checked usage weekly and removed two charts nobody opened. Within two months all six managers used it each Monday, and the spreadsheet meeting was dropped.

4. 4.Tell me about cleaning a messy dataset.

   What they're really asking: Do you check data before trusting it, and document your choices?

   Answer outline (STAR)

   • SituationDuplicates, missing values or IDs that didn't match across sources.
   • TaskGet to data you could trust.
   • ActionProfile the data, set a rule for each problem, document every assumption.
   • ResultThe analysis it made possible, and the fix requested at the source.

   Full sample answer (fictional example)

   I was given a customer file merged from three systems. The same customer appeared under different IDs, emails were in mixed case, and about 9 percent of country values were blank. I profiled each column, then set a rule for each problem and wrote it down: lowercase emails, match customers on email plus postcode, and fill blank countries from the postcode when possible, otherwise label them unknown. I recorded how many rows each rule changed. The analysis ran on 41,000 clean customers, and I asked the source system's owner to fix blank countries at entry.

5. 5.Tell me about delivering an analysis against a very tight deadline.

   What they're really asking: Can you be fast and still honest about what the numbers can't tell you?

   Answer outline (STAR)

   • SituationA leadership request due the next morning.
   • TaskDeliver something reliable in time.
   • ActionAgree the core question, reuse existing data, state the caveats clearly.
   • ResultWhat was decided, and the fuller follow-up you delivered later.

6. 6.Tell me about automating a recurring report.

   What they're really asking: Do you remove repeat work so you can do more valuable analysis?

   Answer outline (STAR)

   • SituationA weekly report that took half a day of copying and pasting.
   • TaskMake it refresh itself without losing accuracy.
   • ActionWrite the query once, schedule the refresh, add checks and a note on definitions.
   • ResultHours saved, and what you did with the time.

7. 7.Tell me about presenting findings to senior leaders.

   What they're really asking: Can you shape a message for people with five minutes and no patience for method?

   Answer outline (STAR)

   • SituationA leadership meeting where you had one slide's worth of time.
   • TaskGet a decision, not just attention.
   • ActionLead with the answer and the ask, show one chart, hold detail for questions.
   • ResultThe decision made, and what you'd trim next time.

8. 8.Tell me about learning a new tool quickly.

   What they're really asking: How fast can you pick up what the job needs?

   Answer outline (STAR)

   • SituationA move to a new BI platform, or a need for Python.
   • TaskBe productive within weeks.
   • ActionRebuild one existing report in the tool, use the docs and a mentor, share what you learn.
   • ResultWhat you delivered, and how you helped others.

9. 9.Tell me about a mistake you found in your own analysis.

   What they're really asking: Do you catch your errors, own them and make checks routine?

   Answer outline (STAR)

   • SituationA figure that didn't reconcile, or a number a colleague questioned.
   • TaskCorrect it quickly and rebuild trust.
   • ActionRecheck the logic, find the cause, tell everyone who saw it, add a check.
   • ResultThe corrected result, and the check that prevents a repeat.

10. 10.Tell me about getting data from a team that was slow to help.

    What they're really asking: Can you work with engineers respectfully to get what you need?

    Answer outline (STAR)

    • SituationA dataset you needed that sat in another team's system.
    • TaskGet access without waiting for ever.
    • ActionExplain the business case, make the request small and specific, offer to test or document.
    • ResultWhat you received, and how the relationship changed.

Technical and role-skill questions (11–22)

Can you do the work itself? Think out loud and name the trade-offs.

1. 11.What's the difference between WHERE and HAVING, and between an INNER and a LEFT JOIN?

   What they're really asking: SQL basics that screen candidates out quickly.

   Answer outline

   • WHERE filters rows before grouping; HAVING filters groups after aggregation.
   • An INNER JOIN keeps matching rows only; a LEFT JOIN keeps every row from the left table.
   • A WHERE condition on the right table quietly turns a LEFT JOIN into an inner one.

   Full sample answer (fictional example)

   I think of it this way. WHERE filters individual rows before they are grouped, so it can't use aggregates; HAVING filters after grouping, so it can, for example HAVING COUNT(*) greater than five. For joins, an INNER JOIN returns only rows that match in both tables, while a LEFT JOIN keeps every row from the left table and fills missing right-side columns with nulls. A trap I watch for is putting a filter on the right table in the WHERE clause of a LEFT JOIN: it removes the null rows and quietly turns it into an inner join. The condition belongs in the ON clause.

2. 12.Write a query that finds the top three products by revenue in each category.

   What they're really asking: Window functions, asked in most analyst SQL tests.

   Answer outline

   • Aggregate revenue by category and product.
   • Rank with RANK() or DENSE_RANK() partitioned by category, ordered by revenue descending.
   • Keep ranks 1 to 3, and say how you'd handle ties.

3. 13.Weekly sign-ups dropped 20%. How do you investigate?

   What they're really asking: Do you check the data itself before chasing a business explanation?

   Answer outline

   • Rule out data problems first: tracking changes, pipeline delays, changed definitions.
   • Segment by channel, device, country, and new versus returning to find where it fell.
   • Then external causes: seasonality, a campaign ending, a release, a competitor.

   Full sample answer (fictional example)

   I'd start by ruling out data problems: did tracking change, did a pipeline run late, or did a definition change? If the data is sound, I'd break the drop down by channel, device, country and new versus returning visitors to find where it fell. A drop concentrated in one segment points to a cause, such as a broken form on mobile or a paused ad campaign. Then I'd check external causes: seasonality, holidays, a release, a competitor's promotion. I'd report what I found, how confident I am and what I'd check next.

4. 14.How would you choose the KPIs for a new product feature?

   What they're really asking: Can you measure success without drowning the team in metrics?

   Answer outline

   • Start from the feature's goal and pick one primary metric tied to it.
   • Add adoption, engagement and guardrail metrics.
   • Define each one precisely: numerator, denominator and time window.

5. 15.How do you choose the right chart for a finding?

   What they're really asking: Can you make the point of a chart obvious at a glance?

   Answer outline

   • Lines for trends over time, bars for comparing categories.
   • Avoid pie charts with many slices and misleading dual axes.
   • One message per chart, stated in its title.

6. 16.Mean or median: when would you use each?

   What they're really asking: Do you know when an average misleads?

   Answer outline

   • The median suits skewed data such as order values or salaries.
   • The mean is sensitive to outliers, but it adds up across groups.
   • Report both when the gap between them tells a story.

   Full sample answer (fictional example)

   I use the median when the data is skewed, such as order values or salaries, where a few large values pull the mean up and no longer describe a typical case. The mean is more sensitive to outliers, but it has one big advantage: it adds up, so mean revenue per customer times customers gives total revenue. I often report both. When the mean is far above the median, that gap tells a story, for example that a small group of big customers drives the total, which is worth mentioning.

7. 17.How do you build an Excel report that other people can maintain?

   What they're really asking: Spreadsheet skill, still tested in most analyst roles.

   Answer outline

   • Keep raw data, calculations and output on separate sheets.
   • Use tables, XLOOKUP or INDEX/MATCH and PivotTables instead of hard-coded values.
   • Document the assumptions, and refresh from a query where you can.

8. 18.How would you calculate month-over-month growth in SQL?

   What they're really asking: Can you use window functions for a standard business metric?

   Answer outline

   • Aggregate to one row per month, then use LAG to bring in the previous month's value.
   • Growth is this month minus last month, divided by last month; guard against dividing by zero.
   • Check partial months and time zones, and leave out the first month.

9. 19.What is a cohort analysis, and when would you use it?

   What they're really asking: Do you know how to see behaviour over time without mixing groups?

   Answer outline

   • Group users by when they started, and track each group's behaviour since sign-up.
   • It separates a real change in retention from the effect of new users joining.
   • Use it for retention, revenue per cohort, and whether recent cohorts behave better.

10. 20.How do you decide whether a difference between two numbers is meaningful?

    What they're really asking: Do you avoid reading noise as a trend?

    Answer outline

    • Look at sample size and how much the metric normally varies week to week.
    • Use a basic test or a confidence interval, and report the size of the gap, not just its direction.
    • Ask whether the difference is big enough to change a decision.

11. 21.How do you keep a dashboard fast when it reads a very large table?

    What they're really asking: Do you build for the people waiting for it to load?

    Answer outline

    • Pre-aggregate in the warehouse, and connect the dashboard to summary tables.
    • Limit filters and visuals per page, and use extracts or caching where the tool allows.
    • Refresh at the pace the decision needs, not every minute.

12. 22.How would you find and remove duplicate rows in a table?

    What they're really asking: A very common cleaning task, and a test of care.

    Answer outline

    • Define what makes two rows duplicates before touching anything.
    • Find them with GROUP BY and HAVING COUNT over one, or ROW_NUMBER over the key.
    • Keep the right row by a rule, such as the latest, and fix the source of the duplicates.

Take-home or live exercise: tips

Analyst interviews often include a live SQL test, a take-home analysis of a small dataset, or a critique of a dashboard.

• Clarify the question and the definitions before you query, and read results back for common sense.
• In a take-home, put the answer and recommendation on the first page, with the method after it.
• In a dashboard critique, ask who uses it and which decision it supports before judging the design.

Skills employers test for

Technical rounds follow the job ad. These skills are named most often in the 493 newest data analyst job ads we analysed (out of 1,422 live postings), shown as the share of ads that mention each. Expect questions on the top few.

1. 1SQL75%
2. 2Python58%
3. 3Power BI45%
4. 4Tableau42%
5. 5Excel34%
6. 6Data quality33%
7. 7Statistics32%
8. 8Data visualization31%

Full list: Data Analyst resume keywords.

From live job ads · computed November 2, 2026 · refreshed daily

Situational questions (23–30)

Scenarios that test judgement: say what you'd do first, and why.

1. 23.Two dashboards show different numbers for the same metric. What do you do?

   What they're really asking: Can you reconcile conflicting numbers and stop it happening again?

   Answer outline

   • Compare definitions, filters, time zones and refresh times.
   • Establish the correct one, and fix or retire the other.
   • Write the definition down where everyone will find it.

   Full sample answer (fictional example)

   I'd compare the definitions behind each dashboard first: the filters, the time zone, whether refunds are included, and when each last refreshed. Usually the difference is one of those. Then I'd work out which version is right for the question people are asking and check it against a source of truth, such as the finance system. I'd fix one and retire the other, or, if both are valid, rename them clearly. Finally I'd write the definition where everyone can find it, so the question doesn't come back next quarter.

2. 24.A manager asks for data to support a decision they've already made. How do you respond?

   What they're really asking: Will you report what the data says, even when it's unwelcome?

   Answer outline

   • Do the analysis honestly, whatever it shows.
   • If it doesn't support the decision, say so privately first, with the evidence.
   • Offer a way to reduce the risk, such as a small test.

   Full sample answer (fictional example)

   I'd do the analysis honestly, whatever it shows, because my value is that people can rely on the numbers. If the data supports the decision, great. If it doesn't, I'd tell the manager privately first, with the evidence, rather than surprising them in a meeting. I'd also look for something constructive: perhaps the data supports a modified version, or the risk can be reduced with a small test or a phased rollout. Most managers prefer to hear the problem from me early than to discover it later from someone else.

3. 25.Five teams send you urgent requests on the same day. How do you prioritise?

   What they're really asking: Can you say no, or not yet, without upsetting people?

   Answer outline

   • Ask what decision each request informs, and by when.
   • Rank by business impact and deadline, and tell the others when to expect theirs.
   • Spot repeated requests that a self-serve dashboard could answer.

4. 26.Your analysis shows a campaign didn't work, and the marketing lead is proud of it. What do you do?

   What they're really asking: Can you deliver bad news in a way people can act on?

   Answer outline

   • Check your method first, including how conversions were attributed.
   • Frame it as learning: what worked, what didn't, what to test next.
   • Show the lead before any wider audience.

5. 27.A stakeholder asks for 'the number' with no definition. What do you do?

   What they're really asking: Can you clarify a request without making the person feel foolish?

   Answer outline

   • Ask what decision it informs, and what they will compare it with.
   • Offer the definitions you could use, with each one's result, and let them choose.
   • Write the chosen definition into the deliverable.

6. 28.You find an error in a report you sent last week. What do you do?

   What they're really asking: Honesty and speed: do people hear it from you first?

   Answer outline

   • Confirm the error and its size before you raise it.
   • Tell everyone who received the report straight away, with the corrected figure.
   • Explain the cause briefly, and add a check so it can't recur.

   Full sample answer (fictional example)

   I'd first confirm that it is really an error and measure its size, so I can say exactly what's wrong. Then I'd tell everyone who received the report straight away, before they act on it, with the corrected figure and a short explanation, without hiding it. I'd then find the cause: a wrong filter, a stale table or a join duplicating rows. Finally I'd add a check, such as a reconciliation against a known total, so the same problem is caught before sending. People trust analysts who report their own mistakes quickly.

7. 29.A new dataset arrives and nobody can explain its columns. How do you proceed?

   What they're really asking: Can you work carefully through documentation gaps?

   Answer outline

   • Profile each column: type, range, nulls, distinct values and examples.
   • Ask the source owner, and compare totals with a system you trust.
   • Write a data dictionary as you learn, and flag the fields you still doubt.

8. 30.The CEO asks a question the data can't answer. How do you respond?

   What they're really asking: Will you say so, and still be useful?

   Answer outline

   • Say plainly what the data does and doesn't cover.
   • Offer the nearest answer, with its caveats, and a way to get a better one.
   • Suggest a survey, a test or a new tracking event, and give a timeline.

Red flags to avoid in your answers

Each one is quick to fix before the day.

1. Reporting numbers without checking them.

   Describe the reconciliations and sense checks you run before anything is shared.

2. Answering the question asked, but never the one behind it.

   Ask what decision the number informs, and shape the analysis to that.

3. Burying the finding under method.

   Lead with the answer, then the evidence, then the caveats.

4. Saying 'the data speaks for itself'.

   Tell the story: what happened, why, and what should be done.

How to prepare for a data analyst interview in 7 days

An hour or two a day is enough; the order matters more than the length.

1. 1

   Day 1: Learn the business

   Read the company's site and results, and guess its five key metrics and how each is defined.

2. 2

   Day 2: SQL daily drill

   Do five timed queries on joins, aggregates, windows and dates against a sample database.

3. 3

   Day 3: Metric-drop script

   Rehearse your framework for a fallen metric aloud: data checks, segments, external causes, next steps.

4. 4

   Day 4: Excel and BI

   Rebuild a small report in Excel or your BI tool, from raw data to a clean dashboard.

5. 5

   Day 5: Story practice

   Write four STAR stories: an analysis that changed a decision, a challenged number, a deadline and a mistake.

6. 6

   Day 6: Portfolio piece

   Prepare a five-minute walkthrough of one project: question, data, finding and result.

7. 7

   Day 7: Mock session

   Do a mock with a friend: a SQL test, a metric case and two behavioural questions. Then rest.

Questions to ask the interviewer

Pick two or three; the answers tell you whether you want the job.

1. Which BI tools and data warehouse does the team use?
2. Who are the main stakeholders, and how do requests reach the analysts?
3. How much time goes to ad hoc requests versus deeper projects?
4. How are metric definitions agreed and documented?
5. Which analysis changed a decision here recently?

Preparation tips

• Practise SQL daily for a week beforehand: joins, aggregations, window functions and dates.
• Rehearse a metric-drop framework out loud: data issues, segments, external causes.
• Bring a portfolio piece: a dashboard or analysis with a clear question and answer.
• Learn the company's business model and guess its key metrics before you go in.

Prepare the general questions too: tell me about yourself, why do you want to work here? and why are you leaving? Afterwards, send a thank-you email; when the offer comes, negotiate the salary.