Data Analyst Interview Questions (2026)
DA interviews are roughly 70% SQL, 20% business reasoning, and 10% spreadsheet or visualization. If you're prepping for DE roles, you've covered the SQL side; the part that catches most engineers off guard is the business framing — interviewers want to hear you translate a vague product question into a precise query. This guide includes 10 fully-explained questions spanning SQL, business metrics, Python, visualization, and case studies.
SQL questions: top 3, MoM growth, signup-to-purchase
Three SQL questions calibrated to DA interview difficulty. All three test window functions plus a clarifying-question moment that interviewers reward.
Q1. Write a query to find the top 3 products by revenue in each category for the last 30 days.
Join orders to products. Filter to last 30 days via WHERE order_date >= CURRENT_DATE - INTERVAL '30 days'. Aggregate SUM(revenue) GROUP BY category, product_id. Apply ROW_NUMBER() OVER (PARTITION BY category ORDER BY total_revenue DESC), filter rn <= 3. Tests window functions and whether you handle ties explicitly — ask: if two products tie for third, should both appear? That determines ROW_NUMBER vs DENSE_RANK.
Q2. Calculate month-over-month growth rate for total revenue.
Aggregate revenue by month with DATE_TRUNC. LAG(monthly_revenue) OVER (ORDER BY month) gives previous month. Growth = (current - previous) / previous * 100. Handle the first month (LAG returns NULL) with COALESCE or CASE. Tests window functions, date truncation, division-by-zero awareness. Mention negative growth rates are valid and shouldn't be filtered.
Q3. Find users who made a purchase within 7 days of signing up.
Join users to orders on user_id. Filter where order_date BETWEEN signup_date AND signup_date + INTERVAL '7 days'. DISTINCT user_id to avoid double counts. Subtle: does 'within 7 days' include day 0? Clarify with the interviewer. Also decide first purchase only vs any purchase within the window. These clarifying questions show analytical rigor.
DA vs DE interview structure
What's the same, what's different. If you're preparing for both, this tells you where your DE prep already covers DA, and what extra you need.
| Dimension | Data Analyst | Data Engineer |
|---|---|---|
| SQL depth | Intermediate (window functions, CTEs) | Advanced (optimization, indexing, plan reading) |
| Python | pandas, numpy, basic scripting | ETL, file I/O, error handling, APIs |
| Business metrics | Heavy (DAU/MAU, retention, LTV, funnels) | Light (mostly embedded in modeling) |
| Visualization | Dedicated round (chart choice, dashboard design) | Almost never tested directly |
| A/B testing | Common case-study round | Rare; sometimes in design |
| System design | Almost never | Dedicated round (pipelines at scale) |
| Data modeling | Light (star schemas at high level) | Dedicated round (grain, SCD, normalization) |
| Stakeholder communication | Dedicated round (explain analysis) | Behavioral round only |
Business metrics questions: LTV, DAU investigation
Two questions that test how you define metrics and structure investigations. Both reward systematic thinking over quick answers.
Q4. Define customer lifetime value (LTV) and explain how you'd calculate it from a transactions table.
LTV = average revenue per customer per period × average customer lifespan. Simple: SUM(revenue) / COUNT(DISTINCT user_id). More sophisticated: segment by acquisition cohort, calculate revenue per cohort over time, fit a curve to project. LTV is most useful compared to customer acquisition cost (CAC). The LTV/CAC ratio measures whether growth is sustainable — below 3:1 is a warning sign.
Q5. Your dashboard shows a 20% drop in DAU. Walk through how you'd investigate.
Structured investigation, not guessing. Step 1: verify the data — pipeline healthy, no missing data, no schema changes. A 20% DAU drop can be a pipeline bug, not real user behavior. Step 2: segment by mobile/desktop, new/returning, geography. Step 3: external events (holidays, competitor launches, app store removal). Step 4: correlate with product changes (deploy, feature flag, A/B test on login). Step 5: adjacent metrics — did sessions drop or did engagement per session drop? The framework shows you don't jump to conclusions.
Where DA and DE interviews overlap
Five overlap areas. SQL is the biggest by far. Business metrics knowledge improves DE schema design even though it's not directly tested in DE rounds.
SQL is the core overlap
SQL is the backbone of both DA and DE interviews. The difference is depth. DA interviews test aggregation, joins, window functions, and CTEs at an intermediate level. DE interviews add optimization, schema design, and advanced window frames. If you can pass a DE SQL interview, you can pass a DA SQL interview. The reverse isn't always true because DE questions push further into edge cases and performance.
Business metrics and KPIs
DA interviews heavily test defining, computing, and interpreting metrics: DAU/MAU, retention, churn, conversion funnels, average order value, customer lifetime value. DE interviews test this less directly but it appears in data modeling rounds — when asked to design a schema for an e-commerce analytics platform, the implicit test is whether you know which metrics the schema must support.
Python for data manipulation
DA interviews test pandas, numpy, and basic Python manipulation. DE interviews test Python more broadly: file I/O, error handling, API integration, scripting. The overlap is transformation logic — both need to filter, aggregate, reshape, and join. DE Python prep covers DA Python requirements plus more.
Visualization
DA interviews often include a round on interpreting or creating visualizations: chart type choice, identifying misleading axes, communicating insights to non-technical stakeholders. DE interviews almost never test viz directly. But understanding what makes a good dashboard helps DEs build better pipelines because you understand the downstream consumer.
Stakeholder communication
DA interviews heavily test explaining technical findings to non-technical people. DE interviews test this in behavioral rounds — describe a time you explained a complex system to a non-technical stakeholder. Same skill, different context: DAs explain analysis results, DEs explain pipeline design and outage root causes.
Python questions: event sequences, IQR outliers
Two pandas-heavy Python questions. Both come up in DA Python rounds and in DE coding rounds that lean toward analyst-style transforms.
Q6. CSV with user_id, event_type, timestamp. Write Python to find the most common sequence of 3 events per user.
Read with pandas. Sort by user_id, timestamp. Group by user_id. For each user, build rolling windows of 3 consecutive events. Convert each triplet to a tuple, count via collections.Counter. Return top N. Key: per-user grouping — events from different users mustn't form cross-user sequences. Edge cases: users with <3 events (skip), duplicate timestamps (secondary sort key), repeated event types (valid).
Q7. Detect outliers in a numeric column using IQR.
Q1 = 25th percentile, Q3 = 75th. IQR = Q3 - Q1. Bounds: Q1 - 1.5×IQR, Q3 + 1.5×IQR. Values outside are outliers. pandas: q1 = df[col].quantile(0.25); q3 = df[col].quantile(0.75); mask = (df[col] < q1 - 1.5*iqr) | (df[col] > q3 + 1.5*iqr). Discuss alternatives: Z-score (assumes normal), modified Z (median absolute deviation), domain-specific thresholds. IQR holds up on non-normal distributions — that's why it's the standard.
DA vs DE compensation by level (US, 2026)
Approximate ranges. The gap is real and widens with seniority because DE talent is harder to find.
| Level | Data Analyst | Data Engineer |
|---|---|---|
| Entry-level (0–2 yrs) | $65K – $90K | $85K – $120K |
| Mid-level (3–5 yrs) | $85K – $115K | $115K – $160K |
| Senior (5–8 yrs) | $100K – $140K | $140K – $200K |
| Staff / Principal (8+) | $140K – $200K (rare beyond) | $200K – $400K (FAANG) |
Visualization and case study questions
Three case-study questions: chart choice, feature impact analysis, and funnel investigation. These are where DA interviews go beyond pure technical skill.
Q8. Visualize relationship between marketing spend and new signups over 12 months. What chart and why?
Dual-axis line chart: months on x, spend on left y, signups on right y. Shows correlation over time. Alternative: scatter plot with spend on x, signups on y, labeled by month — better for identifying linear relationship without time. Dual-axis is better for trends and lag effects (month N spend → month N+1 signups). Mention dual-axis risk: different scales mislead. Always label axes; consider normalizing to percentage change if scales differ greatly.
Q9. PM wants to know which features drive retention. How would you approach this?
Define retention first (7-day, 30-day, custom). Identify features with usage logs: onboarding completion, search, notifications, social sharing, content consumption. Compare retention rates between users who used each feature vs those who didn't. CRITICAL: control for confounders — power users use every feature and have higher retention regardless. Use cohort analysis by signup week to control for time effects. Distinguish 'users who use feature X have higher retention' (correlation) from 'feature X causes higher retention' (causation, requires experiment).
Q10. Conversion funnel shows 60% drop-off between 'add to cart' and 'checkout.' What do you investigate?
Segment the drop-off: by device (mobile checkout has higher friction), user type (new vs returning), product category (some have pricing issues), geography (payment methods vary). Check technical issues: JS errors, page load >3s, payment gateway failures. Industry context: e-commerce cart abandonment averages 70%, so 60% may be below average. If the rate recently increased: new checkout flow, price changes, removed payment options, shipping surprises. Recommend instrumenting step-by-step tracking to identify exactly where users leave.
Four patterns that distinguish strong DA candidates
What separates a passing DA interview from a strong one. Each pattern is independent.
Always define metrics before calculating
When an interviewer says 'churn,' ask: monthly churn or annual? Voluntary or involuntary? Including paused subscriptions? Strong candidates spend 30 seconds clarifying definitions. Weak candidates assume and get the calculation right for the wrong metric — which interviewers count as a fail.
Frame analyses around the business decision
'Which feature drives retention' isn't an analysis goal. The goal is 'should we invest more in this feature?' Strong candidates restate the business question, propose the analysis, and explain how the result changes the decision. Interviewers reward candidates who think about the action the analysis enables.
Distinguish correlation from causation explicitly
'Users who use feature X have higher retention' is correlation. 'Feature X causes retention' requires a controlled experiment or a natural experiment with strong identification. Mentioning the distinction in a case study unprompted shows analytical maturity. Stronger candidates also describe what experiment would be needed.
Quantify the assumption space
When proposing a metric, state the assumptions. 'This LTV calculation assumes users churn at the average rate; for power users it's an underestimate.' Showing the bounds of your analysis is more valuable than showing the analysis itself in DA interviews.
Data analyst interview FAQ
How different are data analyst and data engineer interviews?+
Can a data analyst transition to data engineering?+
Should I apply for data analyst jobs if I want to be a data engineer?+
Do data analysts and data engineers earn similar salaries?+
How important is SQL for data analyst interviews?+
Your SQL gets sharp here
- 01
Active recall beats re-reading by 50%
Cognitive-science meta-reviews (Dunlosky et al., 2013) rank practice testing as a top-tier study technique, while re-reading and highlighting rank near the bottom
- 02
76% of hiring managers reject on the coding task, not the resume
From HackerRank's 2024 Developer Skills Report. Candidates who look strong on paper still fail the live screen if they haven't done timed, executable practice
- 03
Five problem shapes cover 80% of data engineer loops
Dedup, sessionization, top-N-per-group, slowly-changing dimensions, partition tricks. Writing the shapes by hand turns the unfamiliar into pattern recognition