Data Analyst Interview Questions 2026: SQL, Case Studies & Behavioral (With Answers)

Getting a data analyst role in 2026 means clearing three distinct hurdles in the same loop: SQL technical rounds, open-ended business case problems, and behavioral stories. Most candidates over-prepare on one and bomb the other two.

This guide covers all three — with real questions, structured answers, and a practice link so you can rehearse before the clock starts.

The Three-Round Reality

The most common failure mode: treating data analyst interviews like SQL exams. They're business problem-solving interviews that happen to use SQL.

SQL Interview Questions for Data Analysts

SQL rounds are pass/fail in most companies. These are the patterns that come up most often.

1. Calculate Monthly Retention Rate

What they're testing: Window functions, date arithmetic, CTEs.

WITH monthly_active AS (
  SELECT
    user_id,
    DATE_TRUNC('month', event_date) AS activity_month
  FROM events
  GROUP BY 1, 2
),
retention AS (
  SELECT
    m1.activity_month AS cohort_month,
    COUNT(DISTINCT m1.user_id) AS users_in_month,
    COUNT(DISTINCT m2.user_id) AS users_retained_next_month
  FROM monthly_active m1
  LEFT JOIN monthly_active m2
    ON m1.user_id = m2.user_id
    AND m2.activity_month = m1.activity_month + INTERVAL '1 month'
  GROUP BY 1
)
SELECT
  cohort_month,
  users_in_month,
  users_retained_next_month,
  ROUND(100.0 * users_retained_next_month / users_in_month, 2) AS retention_rate_pct
FROM retention
ORDER BY cohort_month;

Use CTEs over nested subqueries — they show you think about readability and debuggability.

2. INNER JOIN vs LEFT JOIN — When to Use Each

Strong answer: Choose LEFT JOIN when you need to preserve the full universe of your base table — e.g., showing all users regardless of whether they completed a purchase. INNER JOIN there silently drops users with no purchases.

Common follow-up trap: "Which is faster?" — The answer depends on data distribution and indexes, not the JOIN type. Saying "LEFT JOIN is slower" signals a misunderstanding.

3. Find Duplicate Rows

SELECT user_id, COUNT(*) AS occurrences
FROM users
GROUP BY user_id
HAVING COUNT(*) > 1;

Extension interviewers add: "Now delete the duplicates, keeping the most recent." This requires ROW_NUMBER() + CTE + DELETE from ranked results.

4. Running Total / Cumulative Sum

SELECT
  order_date,
  revenue,
  SUM(revenue) OVER (ORDER BY order_date) AS running_total
FROM orders;

Add PARTITION BY user_id for per-user running totals.

5. Diagnose a Sudden Traffic Drop

This is a diagnostic business case, not a SQL exercise. Structure your answer:

1. Confirm the signal: Is it across all sources or one channel? All pages or one section?
2. Segment: Break by source/medium, device type, geo, landing page.
3. Correlate with events: Deploys, campaigns, algorithm updates.
4. Hypothesize + test: Form 2–3 hypotheses, then pull SQL to validate each.
5. Communicate finding: What's the most likely cause? What's the confidence level?

This question separates candidates who pull data from candidates who think in hypotheses.

Business Case Interview Questions

Define a Metric for User Engagement

Weak answers: "Daily active users" or "time on site."

Better answer: Engagement metrics should map to business outcomes. For a SaaS product, a stronger engagement metric is "core action completion rate" — the percentage of sessions where the user completed the action the product promises. This predicts retention better than time-on-site.

How Do You Handle Missing Data?

Structure your answer around the reason for missing data:

• MCAR (Missing Completely at Random): Safe to drop rows or use mean imputation
• MAR (Missing at Random): Use model-based imputation
• MNAR (Missing Not at Random): The missingness itself is signal — encode it as a feature

Never say "I drop missing rows" without explaining why it's statistically safe to do so.

Behavioral Interview Questions

Describe a Time Your Analysis Changed a Business Decision

Weak: "I analyzed the churn data and made a dashboard."

Strong: "We were about to launch a win-back email campaign targeting all churned users from the past year. I analyzed the underlying churn reasons and found that 40% had churned due to billing failures, not product dissatisfaction. I segmented these users and recommended a separate reactivation flow. The targeted campaign had 3.2× the reactivation rate of the generic blast."

The key structure: what decision was being made → what data revealed → what you recommended → what happened.

Stats & Probability Questions

What Is a p-value?

A p-value is the probability of observing your result (or more extreme) if the null hypothesis is true. A p-value of 0.04 does NOT mean there's a 96% chance your hypothesis is correct.

Statistical vs. Practical Significance

A result can be statistically significant but practically meaningless if the effect size is tiny. Always pair significance with effect size. A 0.1% conversion lift that's statistically significant is not worth building a feature for.

How to Practice Before Your Loop

Reading answers is not the same as saying them out loud under time pressure. Data analyst interviews have a pacing component — most SQL rounds are 30–45 minutes for 2–3 problems.

Read the full article here

Been using ManyOffer to sharpen my own answers — if you want AI mock interviews with real LP feedback, they have a deal running through July worth checking out: Claim 1 free month here