Question 1

Tell me about an analysis that actually changed a decision.

Accepted Answer

The question someone asked, the data you pulled and one limitation of it, the finding in one sentence with a number, who you presented it to, and the decision they made differently. If the decision saved or earned money, say how much. This is the single most important story to have ready. Common mistake: Describing an impressive analysis that was admired and then ignored. The question is about the decision, not the work.

Question 2

Explain the difference between an inner join and a left join, and when you would use each.

Accepted Answer

In plain words: an inner join keeps only rows that match on both sides; a left join keeps everything from the left table and fills missing matches with nulls. Then a real example — customers with and without orders — and the classic bug: an inner join silently dropping the customers who never purchased. Bonus: mention how a join on a non-unique key inflates row counts. Common mistake: Reciting the textbook definition without the example of when the wrong join produced a wrong number.

Question 3

A key metric dropped 20% last week. Walk me through how you would investigate.

Accepted Answer

A method: first confirm the data pipeline is not broken (late loads, a tracking change), then segment — by platform, region, channel, new versus returning — to find where the drop is concentrated, then check what changed at that time (a release, a campaign ending, a holiday). Say what you would tell the stakeholder at each stage, including "not sure yet" at hour one. Common mistake: Jumping to a business explanation before ruling out a tracking or pipeline problem. Half of all sudden drops are instrumentation.

Question 4

Tell me about a dashboard you built that nobody used. What did you learn?

Accepted Answer

Be honest: what you built, why you thought it was needed, why it was ignored (wrong question, too many charts, nobody owned the metric, no decision attached to it), and what you did differently next time — usually sitting with the user first and building one chart tied to a weekly decision. Common mistake: Claiming all your dashboards are used. Every analyst has one that was not; the panel wants the lesson.

Question 5

How would you explain a window function to someone who only knows basic SQL?

Accepted Answer

An analogy: a window function lets each row look at its neighbours without collapsing them the way GROUP BY does — a running total, a rank within a group, the previous row's value. Give one concrete use: ranking each customer's orders by date to find their first purchase, or a seven-day rolling average. Say which ones you use most (ROW_NUMBER, LAG, SUM OVER). Common mistake: Writing syntax in the air instead of explaining what problem it solves. The panel is testing whether you can teach, not whether you can type.

Question 6

Describe a data quality problem you found and what you did about it.

Accepted Answer

How you noticed (a number that did not reconcile, duplicates, nulls where there should not be), how you sized the impact, who owned the source, what the fix was, and the check you put in place so it would be caught automatically next time. Include the awkward part: telling a stakeholder a previous report was wrong. Common mistake: Presenting yourself as the person who quietly cleaned the data in Excel. The panel wants to hear that you fixed it at the source and told people.

Question 7

A stakeholder asks for a number the data cannot actually give them. What do you do?

Accepted Answer

Find out the decision behind the request, explain plainly what the data can and cannot show, and offer the closest honest proxy with its caveats stated up front. Give an example — "churn" when you only have last-login, or "attribution" from a single touchpoint. Show that you did not just produce a misleading number to be helpful. Common mistake: Producing the number anyway with a footnote nobody reads. This is how analysts lose credibility six months later.

Question 8

Explain how you would set up and read a simple A/B test.

Accepted Answer

Random assignment, one primary metric decided before launch, a sample size worked out from the minimum effect you care about, running for full weekly cycles, and checking the result with a significance test and a confidence interval rather than eyeballing. Mention the two classic errors: peeking early and stopping, and testing ten metrics and reporting the one that moved. Common mistake: Describing a test with no pre-registered metric or sample size. The panel hears "we looked until something was significant."

Question 9

Which tools do you use, and how do you choose between Excel, SQL and a BI tool for a task?

Accepted Answer

Name what you actually use (SQL dialect, Power BI or Tableau or Looker, Python or R if real, Excel) and give a rule: SQL for anything repeatable or over a few hundred thousand rows, Excel for one-off exploration and stakeholder hand-offs, BI for anything someone will look at weekly. Give one example of choosing the wrong tool and what it cost. Common mistake: Listing every tool you have touched. Panels probe the third one on the list and the answer gets thin.

Question 10

Tell me about a time you presented a finding that the audience did not want to hear.

Accepted Answer

What the finding was, why it was unwelcome (it contradicted a plan, a campaign, a senior opinion), how you prepared — checking the analysis twice, leading with the evidence, offering what to do next — and how the room reacted. Show that you stayed factual and did not soften the number to please people. Common mistake: A story where you diluted the finding to keep the peace. The panel is hiring you precisely to not do that.

Question 11

How do you make sure a report is correct before you send it?

Accepted Answer

A checklist: reconcile totals against a known source, sanity-check with a back-of-envelope estimate, look at the extremes and the nulls, re-run the query on a period you already know the answer for, and have someone else read the headline. Give an example of an error one of these steps caught. Common mistake: "I double-check my work." That is a claim, not a method.

Question 12

Describe a time you had to work with messy or incomplete data to meet a deadline.

Accepted Answer

What was missing or wrong, the assumptions you had to make, how you documented and flagged them in the deliverable, and how you communicated the confidence level to the stakeholder. Panels are checking that you can be useful under uncertainty without pretending the uncertainty is not there. Common mistake: Hiding the assumptions in the appendix. State them in the first slide or the first sentence of the email.

Question 13

How do you prioritise when three teams each want an analysis this week?

Accepted Answer

Ask each requester what decision the analysis feeds and when that decision is made; rank by the size and date of the decision, not the seniority of the asker; tell the other two clearly when they will get theirs. Give an example of pushing back on a senior request because a smaller one had a harder deadline. Common mistake: Answering "the most senior person first" or "first come, first served." Neither is prioritisation.

Question 14

Why data analysis, and why this company?

Accepted Answer

One genuine reason you like the work (finding the real cause of a number, changing a decision) and two specific things about this company — its data maturity, a product you use, a domain you know. Then what you would want to learn in year one. Common mistake: Saying you want to move into data science. The panel hears that they are hiring a two-year stopgap.

Data Analyst interview questions — and what a strong answer includes

Tell me about an analysis that actually changed a decision.

Explain the difference between an inner join and a left join, and when you would use each.

A key metric dropped 20% last week. Walk me through how you would investigate.

Tell me about a dashboard you built that nobody used. What did you learn?

How would you explain a window function to someone who only knows basic SQL?

Describe a data quality problem you found and what you did about it.

A stakeholder asks for a number the data cannot actually give them. What do you do?

Explain how you would set up and read a simple A/B test.

Which tools do you use, and how do you choose between Excel, SQL and a BI tool for a task?

Tell me about a time you presented a finding that the audience did not want to hear.

How do you make sure a report is correct before you send it?

Describe a time you had to work with messy or incomplete data to meet a deadline.

How do you prioritise when three teams each want an analysis this week?

Why data analysis, and why this company?

Answer these out loud, get coached

More technology roles

Before the interview

Get new data analyst questions by email

Data Analyst interview questions — and what a strong answer includes

Tell me about an analysis that actually changed a decision.

Explain the difference between an inner join and a left join, and when you would use each.

A key metric dropped 20% last week. Walk me through how you would investigate.

Tell me about a dashboard you built that nobody used. What did you learn?

How would you explain a window function to someone who only knows basic SQL?

Describe a data quality problem you found and what you did about it.

A stakeholder asks for a number the data cannot actually give them. What do you do?

Explain how you would set up and read a simple A/B test.

Which tools do you use, and how do you choose between Excel, SQL and a BI tool for a task?

Tell me about a time you presented a finding that the audience did not want to hear.

How do you make sure a report is correct before you send it?

Describe a time you had to work with messy or incomplete data to meet a deadline.

How do you prioritise when three teams each want an analysis this week?

Why data analysis, and why this company?

Answer these out loud, get coached

More technology roles

Before the interview

Get new data analyst questions by email