What to expect in a data analyst interview
The questions usually fall into three groups: technical questions (SQL, Excel, statistics, and sometimes Python or R), analytical questions (case-style problems about how you'd use data to answer a business question), and behavioural questions (how you communicate, work with stakeholders and cope when a task is unclear).
Some employers also set a take-home task, where you clean and analyse a dataset and present what you found. Give it as much care as the interview, and be ready to explain every choice you made.
Explaining what you found
A lot of an analyst's work is explaining results to people who don't work with data. Practise talking through your findings as if the person listening has never opened a spreadsheet.
Technical skills
Questions 1–8 · Technical
Question 01
"What SQL queries do you use most often? Can you walk me through an example?"
What they're really asking
Have you actually used SQL, or just listed it on your CV?
How to answer it
Walk through a query you've written yourself, even for a uni subject or a project of your own, and show which building blocks you're comfortable with: SELECT, WHERE, GROUP BY, ORDER BY, JOIN types (INNER, LEFT, RIGHT), aggregate functions (COUNT, SUM, AVG) and subqueries. If you've used window functions (ROW_NUMBER, RANK, LAG and LEAD), say what for, such as ranking products within each category. For example: "I used a LEFT JOIN from last quarter's customers to this quarter's orders and kept the rows with no match, to find customers who'd stopped ordering."
Question 02
"What is the difference between a LEFT JOIN and an INNER JOIN?"
What they're really asking
Whether you know what each join returns, and when you'd use it.
How to answer it
An INNER JOIN returns only the rows that match in both tables. A LEFT JOIN returns every row from the left table, with NULLs in the right table's columns where there's no match. Use an INNER JOIN when you only want matched records, and a LEFT JOIN when you need to keep everything from your main table. A common use of a LEFT JOIN is finding records that have no match in the other table: join, then keep the rows where the right table's key is NULL. Either join can return more rows than the left table has if a row matches several rows on the right.
Question 03
"How would you handle missing or null values in a dataset?"
What they're really asking
Real data has gaps. Do you think about why before you delete or fill them?
How to answer it
It depends on how many values are missing and why, so say you'd check that first. The options are removing the rows (fine if only a few are missing, with no pattern to which ones), imputing values (the mean or median for a number, the most common value for a category), using a placeholder such as "Unknown" in a text field, or keeping the nulls when a blank means something (an empty "cancellation date" means the order wasn't cancelled). If the gaps follow a pattern, say one store's system never recorded the field, deleting or filling them can skew your results.
Question 04
"What is the difference between mean, median, and mode? When would you use each?"
What they're really asking
Basic statistics, plus whether you know when each one gives a fair picture.
How to answer it
The mean is the average. It suits roughly symmetrical data, but a few extreme values can pull it a long way. The median is the middle value once the data is sorted. Outliers barely move it, so it's better for skewed data like incomes or house prices. The mode is the most common value, which is useful for categories, such as the most popular plan. The usual example is pay: if a few executives earn ten times what everyone else does, the mean overstates what a typical employee earns, and the median gives a fairer picture.
Question 05
"How would you check whether two variables are correlated?"
What they're really asking
Whether you can apply basic statistics to a practical question.
How to answer it
Start with a scatter plot to see whether there's a relationship and what shape it is. Then calculate a coefficient: Pearson's r for a straight-line relationship between two continuous variables, or Spearman's rank correlation for ordinal data (rankings, ratings) or a relationship that keeps rising or falling without being a straight line. Both run from -1 to +1, and values near either end mean a strong relationship. A Pearson's r near 0 rules out a straight line but not a curve, which the plot would show. Then add that correlation on its own doesn't show causation: ice cream sales and drownings rise together because both go up in hot weather.
Question 06
"What visualisation tools have you used and what makes a good data visualisation?"
What they're really asking
Can you use a chart to get a point across?
How to answer it
Name the tools you've used, such as Tableau, Power BI, Excel, Python (matplotlib or seaborn), Looker or Data Studio (formerly Looker Studio). For the second part, talk about choosing the right chart for the data (bars to compare, lines for trends over time, a scatter plot for how two numbers relate), a title that states the finding rather than just the subject, nothing on the chart that doesn't carry information, and a level of detail that suits the audience. Aim for each chart to answer one question.
Question 07
"What is the difference between a dimension and a metric in data analysis?"
What they're really asking
A basic idea you'll use every time you build a report in a BI tool.
How to answer it
A dimension is a descriptive field you use to group or filter data, such as country, product category, customer segment or date. A metric (also called a measure) is a number you aggregate, such as revenue, number of orders or session length. In a typical analysis you group by dimensions and aggregate metrics: "total revenue (metric) by product category and month (dimensions)". A BI tool may treat postcodes or customer IDs as metrics because they're stored as numbers, and add them up, so check how it has classified each field.
Question 08
"Walk me through how you would approach a data cleaning task."
What they're really asking
Cleaning data takes up a lot of an analyst's time. Do you have a method for it?
How to answer it
Walk through the steps in order. Get to know the data: the number of rows and columns, the data types, and summary statistics for each field. List the problems: nulls, duplicates, inconsistent labels ("USA" and "United States"), fields stored as the wrong type, and outliers to check (some are errors, some are real). Fix what affects your analysis, since not every problem needs fixing. Write down every change, so someone else, or you in a few months, can repeat or check it. Check the result by comparing row counts and important totals with the source.
Practise your behavioural answers out loud
InterviewZap asks behavioural questions like 14–20 below, written from your CV and the job description. Answer by typing or out loud, and after each answer you'll see what was missing and a stronger way to put it.
Start Practising Free →No credit card. Free to start.
Analytical thinking
Questions 9–13 · Analytical Thinking
Question 09
"How would you define and measure the success of a new product feature?"
What they're really asking
Whether you can link data to a business decision.
How to answer it
Start with the problem the feature solves and who it's for. Then pick metrics that fit: adoption (are people finding it?), engagement (do they keep using it?) and its effect on a business outcome such as retention, revenue or task completion. Agree the baseline and how long you'll measure for before launch. Watch other metrics too, in case the feature helps one number and hurts another. If you can, compare against a group that didn't get the feature (see question 12), because a simple before-and-after can be thrown off by the time of year.
Question 10
"If one of our main metrics dropped sharply last week, how would you investigate?"
What they're really asking
Do you work through it in a sensible order, or jump to a theory?
How to answer it
First check the data itself: a tracking change, a broken pipeline or a reporting bug can look exactly like a real drop. Then narrow it down by splitting the metric by region, platform, user type and product area to see where the drop is. Look for causes outside the data, such as a product release, a marketing change, something a competitor did, a public holiday or the time of year. Form a hypothesis and test it before you report a cause.
Question 11
"What is the difference between correlation and causation? Give an example."
What they're really asking
Do you know when data shows cause and effect, and when it doesn't?
How to answer it
Correlation means two things move together. Causation means one makes the other happen. For example, users who read more help articles might stay longer, but making everyone read the articles probably won't change how long they stay. Something else, like how motivated a user is, may drive both, or the link may run the other way, since people who stay longer have more time to read. To show cause and effect you usually need a controlled experiment such as an A/B test, or a careful natural experiment, because observational data alone rarely settles it.
Question 12
"How would you design an A/B test to evaluate a change to a sign-up flow?"
What they're really asking
Whether you understand how to run a fair experiment.
How to answer it
State a hypothesis (changing X will lift sign-up completion by Y%). Choose one primary metric: the sign-up conversion rate. Randomly split users into control and treatment groups. Work out the sample size before you start, from the current conversion rate, the smallest change worth detecting, the significance level (often 5%) and the power you want (often 80%). Run it for the planned length, in whole weeks if you can. Stopping early because the result looks good raises the chance of a false positive. Then analyse the results: check the difference is statistically significant and big enough to act on. Mention guardrail metrics too, the other numbers you'd watch in case the change hurts something else.
Question 13
"How do you prioritise what to analyse when you have multiple requests from stakeholders?"
What they're really asking
Can you decide what comes first when you have more requests than time?
How to answer it
Rank requests by their business impact and how soon they're needed. For each one, ask what decision the analysis will feed into, how big that decision is, and when it has to be made. A number someone is curious about can wait behind analysis for a major product decision. Tell people where their request sits and when they can expect it. If two requests really are equal, ask your manager to decide rather than guessing.
Behavioural questions
Use the STAR method for these.
Questions 14–20 · Behavioural
Question 14
"Tell me about a time you turned data into a clear recommendation."
What they're really asking
Can you take an analysis all the way to a recommendation someone could act on?
How to answer it
Pick one project, even a uni assignment or something you did on your own. Say what the business question was, what you found in the data, and what you recommended and why. Spend most of the answer on the recommendation and how you reached it, and keep the technical detail short. If someone acted on it and you know the result, say so. If you haven't had a real business problem to work on yet, say what you'd have recommended if your coursework had been for a real business.
Question 15
"Describe a time you had to explain a complex analysis to a non-technical audience."
What they're really asking
Whether you can explain your work to people who aren't analysts.
How to answer it
Describe the audience, the analysis, and what you changed about how you explained it. Things that help: the conclusion before the method, an analogy, one clear chart instead of a whole dashboard, and answers ready for the questions they're likely to ask. You'll know it worked if they understood it well enough to make a decision. If you can, describe a moment when your explanation changed what someone thought or did.
Question 16
"Tell me about a time your analysis produced a surprising or counterintuitive result."
What they're really asking
When the data says something unexpected, do you check it before you believe it or dismiss it?
How to answer it
Use a real example: a result that contradicted what you expected or what other people assumed. Walk through how you checked it (ruling out data errors before you accepted it), how you explained it to others, and what happened next. Show that you tested the result properly, instead of either believing it straight away or dismissing it.
Question 17
"How do you approach a problem when you don't have the data you need?"
What they're really asking
What you do when the data you'd like doesn't exist.
How to answer it
Be honest about what's missing. Then talk through what you could do: use a proxy metric, state your assumptions and test how much the answer changes if they're wrong, look for another data source, or narrow the question to something the data can answer reliably. Be clear about what you can and can't conclude from the data you have, and say how sure you are of each finding.
Question 18
"Tell me about a personal or side project involving data."
What they're really asking
Do you work with data even when nobody's asked you to?
How to answer it
Any project counts, even a small one: a Kaggle competition, a dataset you explored for fun (your Spotify listening history, football results, local house prices), a dashboard you built, or a blog post where you wrote up an analysis. Talk about why you picked it and what you found, more than how sophisticated it was. Then say what you'd look at next if you had more time.
Question 19
"How do you make sure your analysis is reproducible?"
What they're really asking
Could someone else check your work, or repeat it?
How to answer it
Describe what you do: commented, documented code instead of one-off scripts, version control (Git), raw data kept separate from cleaned data, assumptions and data sources written down, and date ranges and other changing values set as parameters rather than typed into the code. A uni project kept in Git with a README is enough to show you work this way. A good test: could you rerun the analysis in six months and get the same numbers?
Question 20
"Do you have any questions for us?"
What they're really asking
Are you interested in this job, and in how the team works with data?
How to answer it
Questions you could ask: "What tools and databases does the team use day to day?" / "Do people across the business come to the team with questions, or is the team still building that trust?" / "What would someone in this role work on in their first three months?" / "What's the most interesting analysis the team has done in the past year?"
Before you apply
Build a small portfolio. Two or three analyses on GitHub or a personal site, each with a short write-up of what you did and why, give you real work to talk about in the interview, and something to show besides the list of tools on your CV.