A Google data scientist interview has five rounds. The first round—a 45-minute phone screen—eliminates roughly 80% of candidates. The most common reason? Candidates prep for the wrong questions. They memorize neural network architectures and get blindsided by a basic SQL window function or a probability puzzle a statistics undergrad could solve.
This guide covers the real data science interview questions that show up repeatedly across Google, Meta, Amazon, and mid-size tech companies—organized by topic so you can identify your gaps and fix them before the call.
How Data Science Interview Questions Are Structured
Most data science interviews follow a predictable format, even if the specific questions vary:
- Technical screen (30–60 min): SQL queries, probability, and basic stats. This is the first filter.
- Take-home or timed coding challenge: Python/pandas, exploratory data analysis, sometimes a small modeling task.
- Onsite or virtual loop (3–5 rounds): Machine learning concepts, product/business case questions, past project deep-dives, and a final culture fit round.
Understanding this structure matters because data science interview questions at the phone screen stage are almost never about deep learning. They test whether you can think analytically and write clean SQL. Save your neural network knowledge for round three.
SQL Data Science Interview Questions
SQL is tested in nearly every data science interview, regardless of company size. Expect at least one question involving aggregation and at least one involving joins or window functions.
Commonly asked SQL questions
- "Write a query to find the second highest salary in a table." The answer interviewers want uses a window function (
DENSE_RANK()), not a correlated subquery—even though both work. - "Find users who logged in on two consecutive days." Tests self-joins and date arithmetic. Most candidates struggle with the self-join approach.
- "Calculate a 7-day rolling average of daily active users." Tests
AVG() OVER (ORDER BY date ROWS BETWEEN 6 PRECEDING AND CURRENT ROW). - "Find the percentage of users who completed a purchase after viewing a product." Conversion funnel query—requires conditional aggregation or a CTE.
The pattern: interviewers want to see that you default to window functions for ranking/rolling calculations, that you understand when to use CTEs versus subqueries, and that you can think about NULL handling without being prompted.
Statistics and Probability Data Science Interview Questions
This is where most candidates underestimate the difficulty. Statistics questions in data science interviews are not about memorizing formulas—they test whether you can reason under uncertainty and communicate findings to a non-technical audience.
High-frequency probability and stats questions
- "You flip a fair coin 10 times and get heads every time. What's the probability the next flip is heads?" Classic test of whether you understand independence. Answer: 50%. The previous flips are irrelevant.
- "Explain p-value to a product manager." If you reach for a textbook definition, you'll fail. They want: "It's the probability of seeing results at least this extreme if the treatment had no real effect. It's not the probability your hypothesis is true."
- "We ran an A/B test. The p-value is 0.03. Should we ship the feature?" Strong candidates ask: What's the sample size? What was the pre-specified alpha? Did we run multiple tests? Was this a sequential test? A lone p-value is never sufficient.
- "What's the difference between Type I and Type II errors, and which is worse?" Type I = false positive (ship a feature that doesn't work). Type II = false negative (reject a feature that would have worked). Which is worse depends entirely on context—and saying so is the right answer.
- "A/B test showed a 2% lift in click-through rate. Is it practically significant?" Tests whether you distinguish statistical significance from business impact. A 2% CTR lift on an ad generating $10M/year is worth shipping. On an internal tool used by 12 people, maybe not.
Machine Learning Data Science Interview Questions
ML questions in interviews are mostly conceptual—you won't usually be asked to implement gradient descent from scratch, but you'll need to explain it clearly.
Questions that separate strong candidates
- "Explain the bias-variance tradeoff." High bias = underfit (model too simple, wrong on training and test data). High variance = overfit (model memorizes training data, fails on new data). The goal is finding the sweet spot. Interviewers want a concrete example—decision tree depth is a good one.
- "When would you use a random forest over logistic regression?" Logistic regression when you need interpretability or when the relationship is approximately linear. Random forest when you have nonlinear relationships, missing data, or mixed feature types. Neither is universally better.
- "How would you handle class imbalance?" Expected answers: oversampling the minority class (SMOTE), undersampling the majority, adjusting class weights in the model, or choosing precision/recall over accuracy as the evaluation metric.
- "What evaluation metric would you use for a fraud detection model?" Not accuracy—a model that labels everything as "not fraud" achieves 99.9% accuracy in most datasets. Use precision, recall, F1, or AUC-ROC depending on the cost of false positives vs. false negatives in the specific business context.
- "Explain regularization and when you'd use L1 vs. L2." L1 (Lasso) drives some coefficients to exactly zero—useful for feature selection when you have many irrelevant features. L2 (Ridge) shrinks all coefficients but keeps them non-zero—better when most features are relevant.
Product and Business Case Data Science Interview Questions
These are the questions data scientists most consistently underestimate. Companies want someone who can connect analysis to business decisions, not just run models.
Case-style questions you should practice
- "Daily active users dropped 15% last Tuesday. Walk me through how you'd diagnose it." Structure matters here: Is the data pipeline broken? Is it isolated to a platform, geography, or user segment? Did a product change ship that day? Is it a seasonal pattern? Interviewers want a systematic debugging approach, not a single hypothesis.
- "How would you measure the success of a new recommendation feature?" Metrics depend on the goal: click-through rate, conversion, session length, return visit rate, or revenue per user. Strong candidates ask clarifying questions before naming metrics.
- "Instagram is considering adding a 'downvote' button. How would you evaluate this?" A classic product sense + experimentation question. You'd design an A/B test, define guardrail metrics (don't tank engagement overall), and think through second-order effects on creator behavior.
Top Courses to Prep for Data Science Interview Questions
The fastest way to close gaps is targeted practice, not reading textbooks. These courses address the specific skill areas tested in interviews.
Database Design and Basic SQL in PostgreSQL
SQL is tested in virtually every data science interview, and this course covers the exact concepts—joins, aggregation, subqueries, and database design—that interviewers reach for. PostgreSQL syntax translates directly to interview environments.
Introduction to Data Analytics
A strong conceptual foundation for the exploratory analysis and business case questions that trip up candidates with strong modeling backgrounds but weak analytical frameworks. Covers the reasoning process, not just the tools.
Applied Plotting, Charting & Data Representation in Python
Take-home data science challenges almost always include an EDA component. This course builds the visualization fluency to turn raw data into clear, defensible findings quickly—exactly what graders are evaluating.
Executive Data Science Specialization
The product and business case questions in data science interviews require translating technical work into business language. This specialization builds that communication layer, which is often the deciding factor between two equally technical candidates.
COVID-19 Data Analysis Using Python
A hands-on project-based course using real-world messy data—the kind you encounter in take-home challenges. Working through a complete analysis pipeline with real data is more useful than toy datasets.
FAQ: Data Science Interview Questions
How long should I spend preparing for data science interview questions?
Four to six weeks of focused prep covers most ground: one week on SQL (LeetCode SQL 50 is a good structured list), one week on stats and probability (think about communication, not just formulas), one week on ML concepts, and the remaining time on mock interviews and case practice. Spreading prep over longer than 8 weeks without mock interviews tends to reduce retention without improving performance.
What programming language do data science interviews expect?
Python is the default. pandas and NumPy are expected. You should be able to manipulate DataFrames, handle missing data, write functions cleanly, and do basic visualization. R is occasionally accepted but don't assume it. SQL is tested separately from Python coding—treat them as distinct prep tracks.
Do data science interviews include machine learning coding, or just concepts?
It depends on the role and level. Most entry- and mid-level interviews test ML conceptually—you explain bias-variance tradeoff or describe how gradient boosting works, you don't implement it. Senior roles and research-track positions sometimes include implementing algorithms from scratch in Python. Check the job description: "applied scientist" roles skew toward implementation; "data scientist" roles skew toward application and business impact.
How important is domain knowledge in data science interviews?
For product companies (Meta, Google, Airbnb), extremely important. They want someone who can connect data to product decisions, and that requires understanding what the metrics mean in context. Finance and healthcare roles may expect some domain vocabulary. For pure ML engineering or research roles, domain knowledge matters less than technical depth.
What questions should I ask at the end of a data science interview?
Ask about the data infrastructure you'd be working with, how the team decides which problems to prioritize, and how data science work gets incorporated into product decisions. These signal that you're thinking about real impact, not just getting hired. Avoid asking about salary or work-from-home policy in technical rounds.
Are take-home data science challenges open-book?
Almost always yes—you're expected to use documentation, Stack Overflow, and any tools you'd normally use. What's evaluated is your reasoning, your EDA structure, how you communicate uncertainty, and whether your conclusions are defensible. Graders often care more about the write-up than the model accuracy.
Bottom Line
The candidates who pass data science interviews aren't always the ones with the deepest ML knowledge. They're the ones who can write clean SQL without Googling basic syntax, who can explain a p-value without sounding like a textbook, and who tie every analytical decision back to a business outcome.
Audit your weak spots honestly: if SQL window functions aren't automatic, that's your first priority. If you'd struggle explaining your last project's results to a non-technical PM, that's your second. The Database Design and Basic SQL course handles the technical foundation; the Executive Data Science Specialization handles the communication layer. Address both, and you'll be in the top third of candidates before you walk in the door.