Data analyst interviews test whether you can turn a messy business question into a correct, useful answer, and then explain it to someone who does not care how you got there. Technical skill gets you through the screen. Judgement and communication get you the offer.
This article is organised around the questions candidates most often ask about these interviews. Under each one you will find the interview questions you are likely to face and what a good answer includes. Sample answers are illustrative. SQL and Excel tests are common in this process and each deserves its own preparation; here they get a brief mention so the focus can stay on statistics, cleaning, metrics, visualisation and communication.
What does a data analyst interview process usually look like?
Processes vary, but most include some combination of these stages:
- Recruiter or HR screen: background, motivation, salary expectations, notice period.
- Technical screen: live SQL, a spreadsheet task, or questions on statistics.
- Case or take-home exercise: a dataset and a business question, with a deadline.
- Presentation or panel: you walk through your findings and are questioned on your choices.
- Hiring manager and team fit: how you work with stakeholders, prioritise and handle ambiguity.
Ask the recruiter which stages apply and which tools the team uses. It changes what you should practise.
What statistics questions will I be asked?
Analyst roles rarely need advanced theory, but interviewers will check that you will not mislead anyone. Expect these.
“When would you use the median rather than the mean?”
When the distribution is skewed or has extreme values. Income, house prices, order values and response times are the classic examples: a handful of very large values pull the mean up, while the median shows what a typical case looks like. A good answer also says you would report both if the difference matters to the decision.
“How do you deal with outliers?”
The weak answer is “remove them”. The NIST/SEMATECH statistics handbook defines an outlier as an observation that lies an abnormal distance from other values, and advises that before you consider removing such points, you should try to understand why they appeared, because they often contain useful information about the process or about how data was collected (NIST).
A strong answer covers how you would detect them, for example with a box plot, where NIST describes points beyond Q1 minus 1.5 times the interquartile range, or Q3 plus 1.5 times it, as mild outliers, and points beyond 3 times the interquartile range as extreme outliers. Then explain how you would decide what to do: check for data-entry or tracking errors, confirm with the data owner, and either correct, exclude with a note, or keep and analyse separately.
Illustrative answer: “In a sales dataset, I found a handful of orders worth a hundred times the average. They turned out to be bulk B2B orders recorded in the consumer channel. I didn’t delete them; I split them into a separate segment and told the sales team, because their own dashboards were inflated too.”
“What does a p-value tell you?”
This catches many candidates. In 2016 the American Statistical Association published a statement with six principles on the use of p-values (Amstat News). Among them: p-values can indicate how incompatible the data are with a specified statistical model; they do not measure the probability that the studied hypothesis is true; business or policy decisions should not be based only on whether a p-value passes a threshold; and a p-value does not measure the size of an effect or the importance of a result (APS summary of the ASA statement).
In an interview, say what a p-value is, say what it is not, and then show that you would also report the effect size and a confidence interval, and ask whether the difference is big enough to matter to the business.
“Correlation is not causation. Give an example.”
Choose an example from business, not ice cream and sunburn. For instance: customers who use a loyalty app spend more, but that may be because loyal customers download the app, not because the app makes them spend. Then say how you would get closer to cause: a controlled experiment (A/B test), or at least comparing similar groups before and after a change.
“How would you design an A/B test?”
A good answer lists the steps: define one primary metric and the minimum change worth detecting; decide the sample size and duration before you start; randomise users into groups; avoid stopping early because the result looks good; check for problems such as unequal group sizes; and report the effect with its uncertainty, not just “significant” or “not significant”.
How will they test my data-cleaning skills?
Often with a deliberately messy file. Interviewers want to see that you check before you analyse. Be ready to talk through:
- Duplicates: how you define a duplicate (same ID? same ID and timestamp?) before you remove anything.
- Missing values: whether a blank means zero, unknown or not applicable, and why that changes the result.
- Inconsistent formats: dates in mixed formats, currencies, units, spelling variants of the same city or product.
- Joins that multiply rows: checking row counts before and after joining two tables.
- Documentation: recording every cleaning step so someone else can reproduce your work.
Common question: “Your total revenue figure doesn’t match finance’s. What do you do?” A strong answer: confirm definitions (gross or net, with or without tax, refunds, currency conversion, date of order or date of payment), reconcile on a small sample, and agree a single definition with finance before publishing.
What business and metrics questions should I expect?
These test whether you understand why the analysis matters.
“Sales dropped 15% last month. How would you investigate?” Structure the answer: first confirm the drop is real (tracking changes, data delays, a public holiday); then break it down by segment, such as region, channel, product and new versus returning customers; then compare with the same period last year to rule out seasonality; then form hypotheses and test them. End with what you would report and to whom.
“Which metrics would you track for [our product]?” Do your homework on the company. Suggest a small set: one headline metric tied to the business goal, a few supporting metrics, and at least one guardrail metric that should not get worse (for example, refunds or complaints when you push conversion).
“Tell me about an analysis that changed a decision.” Use the STAR structure, and make the result a decision, not a chart.
What will they ask about dashboards and visualisation?
Expect questions on chart choice and on the tools themselves. If the job uses Power BI, look at what Microsoft tests for its PL-300 Power BI Data Analyst certification. As of the skills update dated 20 April 2026, the exam covers preparing the data, modelling the data (including DAX calculations), visualising and analysing the data, and managing and securing Power BI (Microsoft Learn). Those four areas make a useful revision list even if you are not sitting the exam.
Typical questions:
- “When would you use a bar chart instead of a pie chart?” (Comparing many categories, or values that are close together.)
- “How do you design a dashboard for a senior manager?” (Start with the decisions they make; put the few metrics that support those decisions at the top; add filters, not more charts.)
- “What is the difference between a calculated column and a measure in Power BI?” (A column is computed row by row and stored; a measure is calculated at query time in the current filter context.)
- “How do you control who sees what?” (Workspace roles and row-level security, if you have used them.)
What about the take-home or live case?
Treat it as a work sample. Reviewers look at:
| They check | What good looks like |
|---|---|
| Did you answer the question asked? | A clear answer on the first page, before any method |
| Did you check the data? | A short note on quality issues and how you handled them |
| Is it reproducible? | Clean, commented code or a documented workbook |
| Are the conclusions honest? | Limits stated; no claim of cause from correlation alone |
| Is it readable? | A handful of well-labelled charts, not twenty |
Respect the time limit. If you stopped short of something, say what you would do next. That is often worth more than the extra chart.
How do I answer “explain this to a non-technical stakeholder”?
You may be asked to explain a concept such as statistical significance, or your take-home result, as if to a marketing manager. Lead with the answer and the action, use one number rather than five, avoid jargon, and say how confident you are in plain words.
Illustrative answer: “The new checkout page increased completed orders by about two percentage points. We’re fairly confident it’s a real effect, not noise, and it would add roughly this much revenue a month at current traffic. I’d recommend rolling it out and monitoring refunds for the first month.”
Questions to ask in a data analyst interview
- Where does the data come from, and who owns data quality?
- Which tools does the team use day to day: SQL engine, BI tool, Python or R, spreadsheets?
- Who are the main stakeholders, and how are analysis requests prioritised?
- Is there a single agreed definition for key metrics such as revenue and active users?
- What is a recent analysis that changed a decision here?
- How is the analyst team’s work reviewed?
Before you go in
Pick two projects you can discuss in depth: the question, the data, the problems, your method, the result and what changed. Have them ready in two versions, a technical one and a plain-English one. Then make sure your CV tells the same story with the same tools and outcomes. A data analyst resume builder can help you structure it, and an ATS resume score check will show whether your tool names and keywords match the job advert.
Frequently asked questions
What technical questions are asked in a data analyst interview?
Expect SQL and spreadsheet tasks, statistics questions such as mean versus median, outliers, p-values and A/B test design, data-cleaning scenarios and questions about your BI tool.
How should I answer a question about outliers?
Explain how you would detect them, for example with a box plot and interquartile range, then say you would investigate why they appeared before deciding to correct, exclude with a note or analyse them separately.
Is a take-home assignment common for data analyst roles?
Many employers use one. Answer the question on the first page, note data quality issues, keep your work reproducible and state the limits of your conclusions.
Do I need a certification to become a data analyst?
Usually not, but a certification such as Microsoft’s PL-300 can show tool skills. Practical projects you can discuss in depth tend to carry more weight in interviews.
What questions should I ask in a data analyst interview?
Ask where the data comes from, who owns quality, which tools are used, how requests are prioritised and whether key metrics have agreed definitions.
Sources
- What are outliers in the data? (NIST/SEMATECH e-Handbook of Statistical Methods)
- Statistics Organization Speaks Out on P Values (Association for Psychological Science (reporting the American Statistical Association statement))
- ASA Releases ‘Statement on Statistical Significance and P-Values’ (Amstat News, American Statistical Association)
- Study guide for Exam PL-300: Microsoft Power BI Data Analyst (Microsoft Learn)