Last updated: September 2026. Written by Josh Hutcheson, OnlineCourseing editor. Job figures, books and practice platforms re-checked at the source on 21 September 2026. See our review methodology.
By Josh Hutcheson · E-Learning Specialist
Reviewing online learning platforms since 2019. Review methodology
THE SHORT ANSWER
Bottom line: data science interviews test five things: SQL and Python, statistics and probability, machine learning concepts, product sense, and communication. Find out which rounds your target company runs, give SQL and product questions the most practice because they appear almost everywhere, and be ready to explain every project on your CV in detail.
- Time needed: about 4–6 weeks of steady practice if you already work with data.
- Most tested: SQL (joins, aggregations, window functions) and metrics or A/B-test questions.
- Often underestimated: explaining your own projects and trade-offs clearly.
- Job outlook: data scientist jobs projected to grow 35% from 2025 to 2035 (BLS).
Compare data science courses →
What a data science interview process looks like
Before you sign up for another data science course, read this.
I've taken DataCamp, Dataquest, Coursera ML, and the Udacity nanodegrees. Get my Tuesday picks — plus reader-only codes when they drop.
No spam. Unsubscribe anytime.
The format varies by company and by role (data analyst, product data scientist, machine learning engineer), but most processes combine the rounds below. Ask the recruiter which ones you will face and in what order.
| Round | What it tests | How to prepare |
|---|---|---|
| Recruiter screen | Background, motivation, role fit | A clear two-minute story of your experience and why this role |
| SQL or Python technical | Querying and manipulating data correctly and quickly | Timed SQL practice; pandas for cleaning and aggregation |
| Statistics and probability | Hypothesis testing, distributions, experiment design | Review core concepts and explain them in plain language |
| Machine learning | How models work, when to use them, how to evaluate them | Know the common models, bias and variance, and evaluation metrics |
| Product or case study | Choosing metrics, diagnosing changes, designing experiments | Practise structured answers to metric and A/B-test questions |
| Take-home assignment | End-to-end analysis and communication | Clean code, clear assumptions, a short readable write-up |
| Behavioural | Collaboration, influence, handling ambiguity | Prepared stories in STAR format about real projects |
Step 1: Match your preparation to the role
“Data scientist” covers very different jobs. A product or analytics data scientist spends most interviews on SQL, metrics and experiments; a machine learning engineer faces coding and system design much like a software engineer; a research role goes deep on statistics and modelling. Read the job description line by line, list every tool and technique it names, and rate yourself on each. The gaps are your study plan.
Step 2: Get fast and accurate at SQL
SQL is the most consistently tested skill in data science and analyst interviews because it is used every day in the job. You should be able to write, without help, queries using:
- Inner, left and self-joins, and know when each returns duplicates.
- GROUP BY with HAVING, and conditional aggregation with CASE.
- Window functions: ROW_NUMBER, RANK, running totals, LAG and LEAD.
- Subqueries and common table expressions for multi-step problems.
- Date handling: grouping by week or month, and calculating retention.
Practise on interview-style problems with a timer. DataLemur, built by Nick Singh, has SQL and data science interview questions tagged by company (DataLemur), and our guide to the best SQL courses covers structured options if your foundations need work.
Step 3: Refresh statistics and probability
Expect questions you can answer on a whiteboard: what a p-value means, when to use a t-test, how to calculate a confidence interval, what happens to statistical power when the sample is small, and simple probability puzzles. Interviewers care as much about clear explanations as correct formulas, so practise explaining each concept to someone non-technical. Our statistics courses guide lists refresher options.
Step 4: Know machine learning at the level of the job
For most data scientist roles you need to explain, not derive, the common models: linear and logistic regression, decision trees, random forests and gradient boosting, and clustering. Be ready to discuss overfitting, the bias–variance trade-off, feature engineering, class imbalance, and choosing an evaluation metric (precision and recall versus accuracy, for example). Machine learning engineer roles add deployment, monitoring and scaling questions.
Step 5: Build product sense
Product questions separate strong candidates from good coders. They come in a few recurring forms, and a structure helps with each:
| Question type | Example | A structure that works |
|---|---|---|
| Define a metric | “How would you measure the success of a new feature?” | Clarify the goal; pick one primary metric and two or three guardrail metrics |
| Diagnose a change | “Daily active users dropped 10%. Why?” | Check data and logging first; segment by platform, region and user type; then form hypotheses |
| Design an experiment | “How would you test a new checkout flow?” | Hypothesis, randomisation unit, sample size, duration, success criteria, risks |
| Trade-off | “Engagement rose but revenue fell. Ship it?” | Weigh long-term against short-term effects; propose a decision rule |
Step 6: Know your own projects inside out
Interviewers will pick a project from your CV and dig in: why you chose that approach, what the data problems were, how you validated the result, what you would do differently, and what impact it had. Rehearse a two-minute summary of each project and be ready to go deeper on any step. If you cannot explain a project in detail, remove it from your CV. If you need stronger projects, our guide on becoming a data scientist without a degree covers portfolio work that gets noticed.
Step 7: Handle take-home assignments well
- Read the brief twice and write down the question you are answering.
- State your assumptions and any data quality problems you find.
- Keep the analysis as simple as the question allows; a clear baseline beats an unexplained complex model.
- Write clean, commented code that someone else can run.
- Lead the write-up with the answer and recommendation, then the evidence.
- Stay within the suggested time, and say what you would do next with more.
Step 8: Prepare behavioural stories
Data scientists work with product managers, engineers and executives, so interviewers test how you influence decisions. Prepare five or six stories in STAR form (situation, task, action, result): a time your analysis changed a decision, a time data contradicted a stakeholder, a project that failed, a tight deadline, and a time you explained something technical to a non-technical audience.
Sample questions, and what a strong answer covers
These are typical of each round. The point is not to memorise answers but to see what interviewers are listening for.
SQL: “Find each user’s second purchase date.”
A strong answer uses a window function (ROW_NUMBER partitioned by user and ordered by purchase date) in a common table expression, then filters to row 2. Mention how you would handle two purchases with the same timestamp, and users with only one purchase.
Statistics: “Explain a p-value to a product manager.”
Say it is the probability of seeing a result at least as extreme as the one observed if there were really no effect, and that a small p-value makes “no effect” hard to believe. Add what it is not: the probability the result is true, or a measure of how large or important the effect is.
Machine learning: “Your fraud model is 99% accurate. Is it good?”
Point out that fraud is rare, so predicting “not fraud” every time could also score 99%. Switch to precision and recall, discuss the cost of false positives versus missed fraud, and suggest a precision–recall curve to choose a threshold.
Product: “Sign-ups fell 15% last week. What do you do?”
Start by checking whether the drop is real: tracking changes, outages or a holiday. Then segment by platform, country, traffic source and new versus returning visitors to find where the drop sits, and only then form hypotheses, such as a release or a marketing change, that you can test.
Experimentation: “How long should this A/B test run?”
Explain that duration comes from the sample size needed to detect the smallest effect worth acting on at your chosen significance level and power, divided by daily traffic, and that tests should cover at least one full weekly cycle. Mention the risk of stopping early when results first look significant.
A five-week preparation plan
| Week | Focus | Goal |
|---|---|---|
| 1 | SQL fundamentals to window functions | Solve medium SQL problems in under 15 minutes |
| 2 | Statistics, probability and A/B testing | Explain core concepts out loud without notes |
| 3 | Machine learning concepts and evaluation | Explain five common models and when to use each |
| 4 | Product sense and metric questions | Answer ten practice cases with a clear structure |
| 5 | Projects, behavioural stories and mock interviews | Two full mock interviews; project summaries rehearsed |
Add Python data manipulation practice throughout, and extra coding and system design if you are applying for machine learning engineer roles; our guide on how to crack a coding interview covers that side.
Resources worth your time
- Ace the Data Science Interview by Kevin Huo and Nick Singh collects 201 real interview questions with full solutions across SQL, statistics, machine learning and product sense (official site). It is the closest thing to a standard text for this interview.
- DataLemur offers SQL and data science interview practice questions tagged by company, with many available free (DataLemur).
- DataCamp’s Data Scientist Certification is taken in R or Python and, per DataCamp, consists of two timed two-hour exams (two attempts each), a practical exam on a real-world data project and a recorded presentation of your findings, all within 30 days of registering. It is included in a DataCamp Premium subscription. Useful if you want structured practice that mirrors a take-home plus a credential to show employers.
- Kaggle datasets and notebooks are a free way to practise analysis and build portfolio projects.
See DataCamp’s Data Scientist Certification →
For a ranked comparison of paid programs, see our guide to the best data science courses and our IBM Data Science Professional Certificate review.
Common mistakes
- Spending most prep time on advanced machine learning while SQL and product questions decide the interview.
- Jumping to a model in a product question before defining the goal and the metric.
- Listing projects on a CV that you cannot explain in depth.
- Over-engineering a take-home assignment and missing the question it asked.
- Answering statistics questions with formulas but no intuition.
- Ignoring data quality problems when diagnosing a metric change.
Is data science still worth it?
The demand data is strong. The US Bureau of Labor Statistics projects employment of data scientists to grow 35% from 2025 to 2035, much faster than the average for all occupations, with a median salary of $120,230 in 2025 (BLS). Entry-level roles are competitive, which is exactly why preparation for the interview itself pays off. Our analysis of whether data science is a good career covers the wider picture.
Frequently asked questions
How do I prepare for a data science interview?
Find out which rounds the company runs, then prepare for each: SQL and Python coding, statistics and probability, machine learning concepts, a product or case question about metrics and experiments, and behavioural questions. Most people need four to six weeks of steady practice, with SQL and product sense getting the most time because they appear in almost every loop.
What questions are asked in a data science interview?
Expect SQL queries on joins, aggregations and window functions; Python data manipulation; probability and statistics questions such as hypothesis tests and confidence intervals; machine learning questions on how common models work and how to evaluate them; and product questions about which metric to track, why a metric moved, or how to design an A/B test.
Is SQL important for data science interviews?
Yes, often more than machine learning. SQL is tested in most data science and data analyst interviews, usually in the first technical round, because it is used daily in the job. Practise writing queries by hand under time pressure, including window functions and self-joins.
Do data science interviews include coding like software engineering interviews?
Usually lighter. Data science coding rounds focus on data manipulation in Python or SQL rather than algorithm puzzles, although machine learning engineer and some research roles include algorithm questions similar to software engineering interviews.
What is a data science take-home assignment?
A task you complete on your own time, typically analysing a dataset and presenting findings or building a simple model within a few days. Graders look for clear reasoning, clean code, sensible assumptions and a short, readable write-up, not the most complex model.
How do I answer product sense questions in a data science interview?
Clarify the goal of the product, define a primary metric and a few guardrail metrics, and explain how you would measure a change, usually with an A/B test. If asked why a metric dropped, check data issues first, then break the metric down by segment, platform and time before proposing causes.
Is data science still a good career in 2026?
The job market data is strong. The US Bureau of Labor Statistics projects employment of data scientists to grow 35% from 2025 to 2035, much faster than average, with a median salary of $120,230 in 2025. Competition for entry-level roles is real, so projects and interview preparation matter.
The verdict
Preparing for a data science interview is mostly about covering the right ground: fast, accurate SQL, statistics you can explain simply, working knowledge of common models, structured answers to product questions, and projects you know inside out. Spend four to six weeks on those in proportion to how often they are tested, practise out loud, and do at least two mock interviews before the real one.
See the best data science courses →
Related guides: How to crack a coding interview · Best SQL courses · Best data science courses · Data science jobs · Is data science a good career? · Is data analyst a good career? · How to write a resume