Data Scientist Interview — Stats, ML, SQL, Product Sense
The data-scientist loop spans far more disciplines than other technical interviews — statistics, applied ML, SQL/data manipulation, and product/business judgement, often a coding round too. The exact mix depends heavily on whether the role leans 'analytics DS', 'ML DS', or 'product DS'. Read the JD carefully and tailor prep.
SQL & data manipulation (45–60 min). Live SQL on a realistic schema. Window functions, CTEs, anti-joins, percentile queries, cohort retention.
Statistics (45 min). Hypothesis testing, A/B test design, power, p-values, confidence intervals, multiple testing correction. Often a 'design an experiment' prompt.
Applied ML / modelling (60 min). Walk through a real ML project end-to-end. Bias/variance, regularization, feature engineering, evaluation metrics.
Product / case study (60 min). Open-ended: 'metric X is down 8% week-over-week, what's your investigation plan?' Tests structured thinking, metric framing, and business judgement.
Coding (45 min). Lighter algo round in Python — pandas/numpy gymnastics, or a focused problem like a sliding-window aggregator.
Top Data Scientist technical questions
These are pulled from interview-debrief patterns we see most often across Data & Analytics roles. They are not memorization fodder — interviewers reword them constantly. Practice the underlying skill, not the wording.
We ran an A/B test, p = 0.048, but the lift is +0.3%. Do we ship? What else do you need to know?
Explain bias vs variance with one ML example you've actually run into.
When would you use median over mean as your treatment-effect estimate?
Write a SQL query to compute 7-day retention for users by signup cohort.
Why does adding more features sometimes hurt model performance?
Explain why p-value < 0.05 doesn't mean the result is real with 95% probability.
We have churn prediction at 0.85 AUC. Sales says 'great'. What's missing?
Design an experiment to measure whether dark mode increases engagement.
Daily active users dropped 8% on Tuesday. Walk me through your investigation.
Logistic regression vs gradient-boosted trees — give two scenarios where each wins.
How would you detect data drift in a production model?
Explain SHAP values to a product manager who knows linear regression.
Behavioural questions
Tell me about a model you shipped that didn't deliver expected impact. What did you learn?
Describe a time you killed a project after analysis. How did you communicate it?
When has stakeholder pressure pushed you toward a result that didn't match the data?
Walk me through the most ambiguous problem you've scoped end-to-end.
Preparation tips for Data Scientist candidates
**Tighten your A/B test vocabulary.** Power, MDE, sample-ratio mismatch, novelty effect, network interference — these come up in nearly every product DS loop.
**Practice SQL on real schemas.** Pandas tricks won't save you on a SQL whiteboard. Window functions, lag/lead, and CTE chaining are table stakes.
**Lead the case study.** State the problem, propose 3 hypotheses, prioritize by likelihood × impact, and only then dig into one. Don't jump to numbers.
**Have one ML project at fingertips.** Be ready for 'pick a project — features, model choice, eval metric, what surprised you, what would you do differently'.
Practice with the AI mock interviewer
Panor's AI Job Assistant runs voice-based mock interviews tuned to the Data Scientist role. It ad-libs follow-up questions, calls out red flags in your answers, and produces a transcript with rubric-graded feedback. Resume × JD matching is also included — paste a target job description and the assistant rewrites your bullets in STAR format with keyword alignment scoring.
Strong candidates with relevant experience generally need 4–6 weeks of focused prep for a competitive Data Scientist loop. Career switchers should plan on 8–12 weeks, weighted heavily toward the data & analytics fundamentals.
Do I need to grind LeetCode?
For most Data Scientist loops in 2026, depth on a curated set of 60–80 problems beats grinding 400. Focus on the patterns the questions above test, not problem volume.
Is the format the same at startups vs Big Tech?
No. Big Tech tends to over-index on coding and system design; startups put more weight on judgement, speed, and 'will this person carry the team'. Read the JD and ask the recruiter for the explicit loop structure — they will tell you.