Data scientist interview questions, with answer tips
Data science interviews test three things: whether you can do the technical work, whether you can solve messy real problems, and whether you can explain your work to people who aren't data scientists. Here's what to expect at each stage.
What a data science interview process looks like
| Stage | What it tests |
|---|---|
| Recruiter or hiring manager call | Motivation, background and whether your experience fits the role |
| Technical screen | SQL and Python, often live or as a short online test |
| Take-home task or case study | How you frame a problem, analyse data and present conclusions |
| Technical interview | Statistics, machine learning and your past projects in depth |
| Behavioural interview | How you work with stakeholders, handle setbacks and make decisions |
Technical and SQL questions
- How would you find the second-highest salary in each department?Window functions such as DENSE_RANK() with PARTITION BY. Talk through ties and empty departments.
- What's the difference between a LEFT JOIN and an INNER JOIN, and when would a join create duplicates?Show you understand one-to-many relationships and check row counts after joining.
- How would you handle missing data in a dataset?Start with why it's missing, then the options: drop, impute or model it, and the bias each could introduce.
- How do you make your analysis reproducible?Version control, environments, parameterised notebooks or pipelines, and documentation.
Statistics and machine learning questions
- Explain the bias-variance trade-off.Simple models underfit (high bias), complex models overfit (high variance). Mention how you'd detect and manage each.
- How would you explain a p-value to a non-technical manager?Keep it plain: how surprising the result would be if there were really no effect. Avoid saying it's the probability the result is true.
- Precision or recall: which matters more for a churn model?It depends on the cost of each error. If retention offers are cheap, favour recall; if they're expensive, favour precision. Tie it to the business.
- How would you deal with a heavily imbalanced dataset?Choose the right metric first (not accuracy), then options like resampling, class weights and threshold tuning.
- What is data leakage, and how do you prevent it?Information from the future or the target sneaking into training. Use time-based splits and build features only from data available at prediction time.
- How would you design an A/B test for a new feature?Hypothesis, primary metric, sample size and power, randomisation unit, test length, and guardrail metrics.
Case study questions
Case questions such as "How would you predict which customers are likely to leave?" test how you think, not whether you know one right answer. A strong structure:
- Clarify the goal: what decision will the model support, and how will success be measured?
- Define the target: what counts as churn, and over what time window?
- Data and features: what you'd use, such as usage, billing, support contacts and tenure, and what to watch for, such as leakage.
- Approach: start with a simple, explainable baseline before anything complex.
- Evaluation: the right metric for the business cost of errors, and a time-based test set.
- Deployment and monitoring: how it would be used, and how you'd spot performance drifting over time.
Behavioural questions for data scientists
- Tell me about a time your analysis changed a decision.The decision, your evidence, how you presented it and what happened as a result.
- Describe a model or analysis that didn't work as expected.Honesty, how you diagnosed the problem, what you fixed and what you'd do differently.
- How have you explained a complex model to a non-technical audience?Focusing on the decision and the trade-offs, using examples rather than maths.
- Tell me about a time stakeholders wanted a different answer from the one the data gave.Holding your ground with evidence while understanding their concerns.
- How do you prioritise when several teams want analysis from you?Impact, urgency and effort, and being transparent about trade-offs.
Answer behavioural questions with the STAR method. There's a full data science example in our competency-based interview guide, about persuading a product lead to pause a campaign when a churn model started drifting.
Tips that make a difference
- Lead with impact, not method. "Cut churn in a key segment" lands better than "I used gradient boosting".
- Know your projects inside out. Expect deep follow-ups on every project on your CV.
- Think out loud in technical and case questions. The reasoning matters as much as the answer.
- Ask about their data. Questions like "How mature is the data platform?" show you've done this for real.
Frequently asked questions
What questions are asked in a data scientist interview?
Usually a mix of SQL and Python questions, statistics and machine learning concepts, a case study or take-home task, deep dives into your past projects, and behavioural questions about working with stakeholders.
How do I prepare for a data science case study interview?
Practise a consistent structure: clarify the business goal, define the target, discuss data and features, start with a simple baseline, choose a metric that reflects the cost of errors, and explain how you'd deploy and monitor the model.
Do data scientists get behavioural questions?
Yes, and they often decide close calls. Expect questions about explaining your work, handling disagreement over results, prioritising requests and learning from analyses that went wrong.
How technical are data science interviews in the UK?
It varies by company and level. Most include a SQL or Python test and questions on statistics and machine learning, while senior roles focus more on project depth, judgement and stakeholder skills.