Interview Coach
Data Scientist Interview Questions
Data Scientist loops test statistical judgement more than modelling technique. The rounds that fail candidates are the case study — where you must translate a vague business question into a measurable one — and the experimentation round, where interviewers probe whether you understand what an A/B test can and cannot tell you. Modelling questions are usually shallower than candidates expect and focus on evaluation choices rather than architectures.
Build your story bank freeWhat data scientist interviews are scored on
Business case framing
Given a vague prompt — "engagement is down" — define the metric, form hypotheses, and state what data would distinguish them. The highest-weighted round at most companies.
Statistics and experimentation
A/B test design, power and sample size, p-values and their misinterpretation, multiple comparisons, and novelty effects. Interviewers commonly ask you to critique a flawed experiment.
SQL and data manipulation
Window functions, cohort queries, and joins against a realistic schema. Almost always a standalone round.
Modelling judgement
Choosing and defending an evaluation metric, handling class imbalance, explaining why a model is failing in production.
Technical data scientist interview questions
Questions of this shape recur across data scientist loops. Practise them aloud — interviewers score how you reason, not only where you land.
- Daily active users dropped 8% week over week. How do you find out why?
- Design an A/B test for a new onboarding flow. How do you size it and when do you stop it?
- Your model has 95% accuracy on a dataset where 96% of cases are negative. What do you report instead?
- Write a SQL query returning each user's first and third purchase dates and the gap between them.
- Explain p-value to a product manager, and then explain what it does not mean.
- A model that performed well offline is underperforming in production. What are your hypotheses?
- How would you measure the effect of a feature you cannot randomise?
Behavioural questions for data scientist roles
Prepare one STAR story per theme. A single strong story usually answers two or three of these prompts.
- Tell me about an analysis whose conclusion the stakeholders did not want to hear.
- Describe a time your model or analysis turned out to be wrong.
- Tell me about translating an ambiguous business question into something measurable.
- Describe a project where the data quality was much worse than expected.
- Tell me about a time you chose not to build a model.
Numbers that make data scientist answers credible
A STAR answer without a result is a story. These are the measures that carry weight in this role.
Questions worth asking your interviewer
- How are experiments run here — is there a platform, and who decides what ships?
- How often do models make it to production versus staying as analyses?
- Who owns the data pipelines the science team depends on?
- How is data science success measured on this team?
Data Scientist interview FAQs
What is the hardest round in a Data Scientist interview?+
The business case. You are given something deliberately vague — "retention dropped, find out why" — and scored on how you decompose it: segmenting to localise the change, separating seasonality from a real shift, distinguishing a data-collection bug from a product regression, and stating what evidence would confirm each hypothesis. Candidates who jump straight to modelling fail this round. Interviewers are testing whether you would spend three weeks on the right question or the wrong one.
How much SQL do Data Scientists need?+
Deep working fluency, and it is nearly always a separate round. Expect window functions, self-joins, cohort and funnel queries, and date arithmetic against a schema you see for the first time. Many companies screen on SQL before any statistics round, so it is the most common early elimination point. Being able to write a retention cohort query from scratch is a reasonable bar to hold yourself to.
Data Scientist versus Machine Learning Engineer — which should I apply for?+
Data Scientists answer questions; ML Engineers ship systems. If your strength is experimentation, causal reasoning, and communicating findings that change decisions, target Data Scientist. If it is training pipelines, serving infrastructure, latency, and model monitoring in production, target ML Engineer. The interviews differ accordingly — DS loops weight statistics and case studies, MLE loops weight system design and coding. Applying to both with one CV usually reads as unfocused to each.
What should a Data Scientist CV show?+
Decisions changed, not models built. "Redesigned the pricing experiment framework, cutting time-to-decision from 6 weeks to 9 days and catching a $2M annual revenue leak from an under-powered test" beats any list of algorithms. Name the business metric you moved, the size of the population affected, and the decision that followed. Keep the technique list compact — hiring managers assume you know regression and gradient boosting.
Need the CV before the interview?
See data scientist CV examples, before/after bullets, and the metrics reviewers look for.
Other interview guides
Turn your experience into answers
CVEdge reads your CV, drafts STAR stories from what you actually did, and matches them to the data scientist job you're interviewing for.
Start building free