Interview prep · Technology

Data Analyst interview questions and answers

Interviewers for data analyst roles want to know whether you can pull and clean data accurately, choose sound methods, and turn the results into a recommendation that non-technical people can act on. They will test your SQL, Python or R, and visualization skills, but they also look closely at how you handle ambiguity, catch errors, and work with stakeholders. Prepare examples that show both technical rigor and clear communication.

1. Question 1

   Tell me about yourself.

   What they're asking: They want a concise summary of your analytics experience, your strengths, and why this role fits you.

   Sample answer

   I'm a data analyst with [X years] of experience turning messy data into decisions for [type of team, e.g., marketing or operations]. Day to day, I write SQL to pull and shape data from [database type], use Python [or R] for cleaning and statistical analysis, and build dashboards in [Tableau or Power BI] that stakeholders actually use. Most recently, I [specific accomplishment, e.g., rebuilt a weekly sales report so manual prep dropped from six hours to one]. I enjoy translating numbers into a clear recommendation, so I always ask stakeholders what decision they're trying to make. I'm excited about this role because [reason tied to the company's data, product, or mission], and I'd like to bring my [strongest skill] to your team.

2. Question 2

   How would you write a SQL query to find the top 3 customers by revenue in each region, and what could go wrong?

   What they're asking: They are checking that you know window functions and can anticipate data-quality problems that distort results.

   Sample answer

   I'd use a CTE to aggregate revenue by customer and region, joining the customers and orders tables and excluding cancelled or refunded orders. In the next step I'd apply DENSE_RANK() OVER (PARTITION BY region ORDER BY total_revenue DESC) and keep rows where the rank is 3 or less. I'd pick DENSE_RANK or ROW_NUMBER depending on whether ties should all be included. Things that can go wrong include duplicate rows from a one-to-many join inflating revenue, NULL regions, customers who changed regions, and unclear date ranges. I'd compare row counts before and after each join and reconcile my total against a trusted figure. I'd also confirm with the stakeholder how they define revenue before running anything.

3. Question 3

   Walk me through how you clean and validate a messy dataset in Python or R before analyzing it.

   What they're asking: They want to see a disciplined, reproducible process rather than ad hoc fixes that could introduce errors.

   Sample answer

   I start by profiling the data before changing anything: row counts, data types, summary statistics, and missing values per column, using pandas [or dplyr in R]. Then I look for duplicates, impossible values like negative ages or future dates, inconsistent categories such as 'NY' versus 'New York', and outliers. For each issue I decide whether to fix, flag, or remove it based on its cause, and I never silently drop data. For missing values, I consider whether they're random or systematic before imputing. I keep every step in a documented script or notebook instead of editing by hand, so anyone can rerun it. Finally, I reconcile the cleaned data against source totals and note my assumptions in the deliverable.

4. Question 4

   How would you determine whether the results of an A/B test are statistically significant and worth acting on?

   What they're asking: They are testing your grasp of statistical analysis and whether you can separate statistical significance from business impact.

   Sample answer

   I'd start with the hypothesis and primary metric, such as conversion rate, pulling event data from Google Analytics or our warehouse. I'd confirm the test was set up well: random assignment, enough sample size, and a duration covering full weekly cycles. For conversion rates I'd use a two-proportion z-test or chi-square test, and for a skewed continuous metric like order value I'd consider a non-parametric alternative. I'd report the p-value with a confidence interval and effect size, because a significant result isn't always a meaningful one. I'd also watch for early stopping and unadjusted segment comparisons. Then I'd explain the outcome in plain language: what changed, how confident we are, and what I recommend.

5. Question 5

   Tell me about a time your analysis changed a decision.

   What they're asking: They want evidence that you go beyond reporting numbers and use critical thinking to influence outcomes.

   Sample answer

   Situation: At [company/team], leadership planned to cut [product line/campaign] because the overall [metric] looked flat. Task: I was asked to validate that decision before the budget meeting. Action: I segmented the data in SQL by [customer type/channel] and found that [segment] was growing while another segment masked it. I checked it with a [cohort analysis/significance test] to rule out noise, built a simple [Tableau/Power BI] view, and walked the director through it, including its limitations. Result: The team chose to [keep/reallocate] the budget, and over [time period] [metric] improved by [result]. It taught me to always check segments before trusting a topline number.

6. Question 6

   Describe a time you caught an error in your own data or report.

   What they're asking: They are assessing attention to detail and whether you take ownership of accuracy.

   Sample answer

   Situation: While preparing a monthly report for [team], I noticed [metric] was [X] higher than the previous month with no obvious cause. Task: I was responsible for the report's accuracy before it reached [executives/clients]. Action: I traced the numbers back through my SQL and found a join to [table] was duplicating rows for customers with multiple [records]. I fixed the query, added a row-count check and a reconciliation against [source system], and told my manager about the correction. Result: The report went out accurate, and the checks became a standard step in our team's process. It reinforced that I'd rather delay a report an hour than send a wrong number.

7. Question 7

   Tell me about a time you juggled multiple projects with competing deadlines.

   What they're asking: They want to know how you prioritize, communicate, and deliver when several stakeholders need work at once.

   Sample answer

   Situation: In [month/quarter], I was handling [three] requests at once: a [dashboard build], an [ad hoc analysis] for [team], and a [recurring report]. Task: I needed to deliver all of them without letting quality slip, even though some deadlines conflicted. Action: I listed each request with its deadline, business impact, and effort, then met with my manager and stakeholders to confirm priorities. I split the dashboard into phases, delivered a basic version first, and automated [the recurring report] with a scheduled query to free up time. I sent short status updates every [week]. Result: I delivered all three by [dates], and stakeholders said the phased approach let them give feedback earlier.

8. Question 8

   Tell me about a time your findings contradicted what a stakeholder expected.

   What they're asking: They are evaluating your communication and collaboration skills when the data delivers unwelcome news.

   Sample answer

   Situation: A [marketing/operations] manager at [company] believed [campaign/process] was driving [result], but my analysis showed the effect was mostly due to [seasonality/another factor]. Task: I had to share findings that contradicted their expectation while keeping a good working relationship. Action: I met with them before the presentation, asked how they'd reached their view, and walked through my method using a simple chart instead of statistical jargon. I acknowledged the data's limits and proposed a follow-up [test] to settle the question. Result: They agreed to run the test, which showed [outcome], and they began involving me earlier in planning. I learned that bringing people into the analysis early builds trust.

9. Question 9

   A stakeholder says the numbers in your dashboard don't match another report they use. What do you do?

   What they're asking: They want to see how you troubleshoot discrepancies and protect trust in your work.

   Sample answer

   I'd treat it as a trust issue and investigate right away. First, I'd ask for the other report or a screenshot, plus the filters, date range, and definitions behind it. Then I'd compare the two sources step by step: metric definitions such as how 'active user' or 'sale' is counted, time zones, filters, refresh timing, and whether one source removes duplicates or refunds. In Google Analytics, sampling or different attribution settings can also explain gaps. Once I find the cause, I'd explain it simply and fix my dashboard if it's wrong. I'd also document the definitions on the dashboard so the confusion doesn't recur. If both numbers are valid, I'd recommend agreeing on a single source of truth.

10. Question 10

    Website conversions drop sharply from one week to the next. How would you investigate?

    What they're asking: They are testing your structured problem solving and whether you validate data before drawing conclusions.

    Sample answer

    I'd first confirm the drop is real by checking whether tracking broke, such as a missing tag after a site release, a consent-setting change, or a data processing delay in Google Analytics. If the data is valid, I'd break the drop down by channel, device, landing page, geography, and new versus returning users to see where it's concentrated. I'd compare against the same period last year to account for seasonality and ask the web and marketing teams about site changes, campaigns, or outages. I'd share early findings along with my level of confidence, then recommend next steps, such as fixing a tag or rolling back a change, and suggest an alert to catch similar drops sooner.