How to Prepare for a Data Scientist Interview, Round by Round
Prepare for a data scientist interview one round at a time, with SQL, pandas, statistics, A/B testing and ML problems you solve in code, a six-week plan and a free timed mock interview.
How do you prepare for a data scientist interview?
You prepare for a data scientist interview by practising each round separately and then rehearsing them together under a timer. The rounds are usually SQL, pandas or Python, statistics and probability, A/B testing, machine learning, and a product case. TensorTonic has coding problems, lessons and timed assessments for the technical rounds, including a free Data Science Mock Interview, and this guide links to them round by round.
Most candidates spend the bulk of their time on SQL because it is the easiest round to practise. The rounds that decide the outcome are often the ones with less practice material, such as experiment analysis and statistics written as code. A balanced plan gives every round its own practice and its own timed check.
What rounds are in a data scientist interview?
A data scientist interview usually has a SQL round, a statistics or probability round, an experiment design question and a product case, with a machine learning round in modeling-focused roles. Many loops also include a pandas or Python exercise, either live or as a take-home. This is what candidates commonly report, and individual loops vary, so ask your recruiter which rounds apply to you.
| Round | What it tests | Where to practise |
|---|---|---|
| SQL | Joins, aggregation, window functions and CTEs on business tables | SQL Sheet, SQL Sprint on Window Functions |
| Pandas and Python | Cleaning, grouping, merging and reshaping a DataFrame, plus basic Python | Pandas Sheet, pandas and NumPy sprint |
| Statistics and probability | Distributions, Bayes, confidence intervals and hypothesis tests | Probability and Statistics study plan, Applied Probability and Statistics |
| A/B testing and experiment design | Choosing a metric, sizing a test, reading the result and spotting bias | A/B testing lesson, A/B Test Analysis |
| Machine learning | Classic models, evaluation metrics and feature handling | Cracking ML, decision trees and random forests sprint |
| Product case | Defining metrics, diagnosing a metric change and recommending a decision | Practise aloud with the structure described below |
The title covers different jobs at different companies. Product analytics roles weigh SQL, experiments and product sense most heavily, while modeling roles add more machine learning. Our guide on ML engineer vs data scientist compares the roles and their interviews if you are deciding between them.
How do you prepare for the SQL round?
Prepare for the SQL round by writing queries against realistic tables until joins, aggregation, window functions and CTEs are routine. The SQL Sheet has 25 problems in five sections that run from SELECT fundamentals to window functions and advanced queries, and each section ends with a quiz. Every query runs against test tables, so you see whether the output matches.
| Topic | Practice |
|---|---|
| Filtering and sorting | WHERE Clauses, NULL Handling |
| Joins | LEFT JOIN, Self Join, Multiple Joins |
| Aggregation | GROUP BY, HAVING Clause, Conditional Aggregation |
| Ranking with window functions | ROW_NUMBER, RANK and DENSE_RANK |
| Comparing rows over time | LAG and LEAD, Running Totals |
| Structuring long queries | Common Table Expressions, Subqueries, CASE Statements |
| Dates and strings | Date Functions, String Functions |
Window functions separate strong answers from weak ones, because questions about the top items per group, changes from one period to the next and running totals all need them. Interviewers often follow a ranking question by asking what happens with ties, so make sure you can explain how ROW_NUMBER, RANK and DENSE_RANK differ. The 30-minute SQL Sprint on Window Functions tests this under time pressure.
How do you prepare for the pandas and Python round?
Prepare for the pandas round by solving the same kinds of questions you practised in SQL with a DataFrame. Interviewers often ask for one question in both, so grouping, merging and ranking in pandas should feel as familiar as the SQL versions. The Pandas Sheet has 25 problems in five sections, from building and inspecting a DataFrame to cleaning, grouping, merging and reshaping.
| Topic | Practice |
|---|---|
| Selecting and filtering | Loc vs iLoc, Boolean Indexing |
| Cleaning | Handle Missing Values, Drop Duplicates, Change Data Types |
| Grouping and aggregating | GroupBy Basics, Multi-Level GroupBy, Aggregation Functions |
| Combining tables | Merge DataFrames, Concatenate DataFrames |
| Reshaping | Pivot Tables, Melt Wide to Long, Cross Tabulation |
| Custom logic | Apply Custom Transforms |
Some loops include a short general Python round as well. The free Python Data Structures Sprint checks lists, dictionaries and similar basics under a timer, and the guide on how much Python you need has a checklist.
How do you prepare for the statistics and probability round?
Prepare for the statistics round by learning each concept well enough to compute it in code and explain it in one or two sentences. Interviewers commonly ask about conditional probability, Bayes theorem, confidence intervals, p-values and hypothesis tests, and they often ask you to interpret a result as well as calculate it. The Probability and Statistics study plan has 30 problems in six sections, from descriptive statistics and distributions to sampling, hypothesis testing and Bayesian updating.
| Topic | Lesson | Practice |
|---|---|---|
| Conditional probability and Bayes | Conditional Probability, Bayes Theorem | Bayes Theorem, Bayesian Updating |
| Distributions and the central limit theorem | Distributions, Central Limit Theorem | Percentiles |
| Confidence intervals | Confidence Intervals | CLT Confidence Interval, Bootstrap Confidence Interval |
| Hypothesis tests and p-values | Hypothesis Testing, P-Values, t-Tests | One-Sample t-Test, P-Value Analysis |
| Type I and type II errors | Type I vs Type II Errors | |
| Tests for categories | Chi-Square Test | |
| Correlation | Correlation | Pearson Correlation, Correlation Significance |
Practise saying definitions precisely. A p-value is the probability of seeing a result at least as extreme as yours if the null hypothesis were true, and candidates who describe it as the probability that the null hypothesis is true lose credit. The 45-minute Applied Probability and Statistics assessment runs from Bayes posteriors to a two-proportion z statistic.
How do you prepare for A/B testing and experiment design questions?
Prepare for A/B testing questions by practising the full sequence an experiment follows, from choosing a metric to making a decision. You should be able to state a hypothesis, pick a primary metric and guardrail metrics, estimate the sample size, run the right test and say what you would recommend. The A/B testing lesson explains the method with interactive visualizations, and the A/B Test Conversion Rates and A/B Test Significance problems have you compute the result.
- State what change is being tested and what you expect it to do.
- Choose one primary metric and a few guardrail metrics that must not get worse.
- Decide the unit of randomization, such as users or sessions, and explain why.
- Estimate the sample size from the baseline rate, the smallest effect worth detecting, the significance level and the power.
- Run the test for the planned duration without stopping early because the result looks good.
- Compute the effect, its confidence interval and the p-value, then recommend a decision.
Interviewers then ask what could make the result misleading. Common answers are peeking at the result before the planned end, a novelty effect that fades, a mismatch in the sample ratio between groups, and spillover when users in one group affect users in the other. The 60-minute A/B Test Analysis assessment covers sample size and spillover bias.
How do you prepare for the machine learning round?
Prepare for the machine learning round by implementing the classic models once and learning to choose and compute evaluation metrics. Data scientist interviews tend to ask about linear and logistic regression, decision trees, clustering, overfitting and metrics such as precision, recall and AUC. Cracking ML has 35 problems in six sections that cover linear models, classification, trees and ensembles, unsupervised learning, evaluation and feature engineering in NumPy.
| Topic | Practice |
|---|---|
| Linear regression | Linear Regression Closed Form, Linear Regression from Scratch |
| Logistic regression | Logistic Regression |
| Trees | Decision Tree Split |
| Clustering | K-Means Assignment, K-Means Centroid Update |
| Classification metrics | AUC-ROC, Precision, Recall and Average Precision |
| Preparing features | Feature Scaling, Imputing Missing Values |
Metric questions are common because they connect a model to a business decision. Our guide to ML evaluation metrics explains which metric fits which problem, and the ML coding interview questions guide lists the implementations interviewers ask for most.
How do you prepare for the product case round?
Prepare for the product case round by answering practice questions aloud with a fixed structure. The two common question types are defining success metrics for a product and explaining why a metric moved. This round is a conversation, so it is practised by speaking, ideally with another person who can interrupt and ask follow-up questions.
- For a metric definition question, clarify the goal of the product, name one primary metric, add guardrail metrics and say how you would measure each from the data.
- For a metric change question, first check whether the data is correct, then split the metric by time, platform, region and user segment to find where the change is concentrated.
- For a launch decision question, state the trade-off between the metrics that improved and the ones that got worse, and give a recommendation with its main risk.
The SQL and experiment rounds feed this one. A candidate who can say which query would confirm a hypothesis, or which test would settle a disagreement, gives a much more convincing product answer.
How should you plan data scientist interview prep week by week?
A six-week plan works well for data scientist interview prep, with one round as the focus of each week and a timed mock at the end. The order below starts with SQL because it is usually the first screen, and it finishes with full mock interviews. If you have less time, keep the order and shorten each block.
| Week | Focus | Practice |
|---|---|---|
| Week 1 | SQL fundamentals, joins and aggregation | The first three sections of the SQL Sheet |
| Week 2 | Window functions, CTEs and the same questions in pandas | The last two sections of the SQL Sheet, the Pandas Sheet and the SQL Sprint on Window Functions |
| Week 3 | Probability, distributions, confidence intervals and hypothesis tests | The Probability and Statistics study plan and the Applied Probability and Statistics assessment |
| Week 4 | A/B testing and experiment design | The A/B testing lesson, the two A/B problems above and the A/B Test Analysis assessment |
| Week 5 | Classic ML models and evaluation metrics | The linear models, classification and evaluation sections of Cracking ML |
| Week 6 | Product cases and full timed mocks | Two sets of the Data Science Mock Interview and product cases practised aloud |
How do you know you are ready for a data scientist interview?
You are ready when you can finish a mixed set of questions inside the time limit and explain each answer without notes. The free Data Science Mock Interview is a 75-minute sitting with five problems that mix SQL, pandas, NumPy, math and ML, and it ends with a scored report. It has four sets, so you can take one early as a baseline and another near the end of your plan.
The Interview section has a Data Scientist track with 11 timed assessments in total. The Data Science Mock Interview and the Python Data Structures Sprint are free, and the other sprints are part of Pro. The guide on ML mock interviews explains how to review a mock so the next one goes better.
Which platforms help with data scientist interview prep?
TensorTonic covers the technical rounds of the data scientist interview on one account, with SQL and pandas sheets, statistics problems solved in code, ML implementation problems and a Data Scientist track of timed assessments. Several other platforms are also used for data science interview practice.
- StrataScratch offers a bank of SQL and data science interview questions.
- DataLemur offers SQL interview questions for data roles.
- LeetCode has a database section with SQL problems alongside its algorithm problems.
- Kaggle hosts data science competitions and public datasets.
What mistakes do candidates make in data scientist interviews?
- Practising only SQL leaves the statistics, experiment and product rounds unprepared, and most loops include at least one of them.
- Writing a query without first asking about duplicates, NULL values and the grain of each table often produces a wrong count.
- Reciting a statistical definition without being able to compute it makes follow-up questions hard to answer.
- Reporting a p-value without the effect size and confidence interval gives an incomplete experiment answer.
- Jumping to a model in a case question before defining the metric and the decision makes the answer hard to follow.
- Practising without a timer hides how long each question really takes.
- Staying silent while solving gives the interviewer nothing to assess, so say your assumptions and your plan before you write code.
Practice this
- RANK and DENSE_RANK covers the ranking question that appears in many SQL rounds, including how ties change the answer.
- A/B Test Significance has you compute whether the difference between two variants is real, which is the core of the experiment round.
- GroupBy Basics is the pandas operation behind most questions that start with a per-user or per-day summary.
Then take the free Data Science Mock Interview to see how the rounds feel together under a timer.
Is SQL enough to pass a data scientist interview?
SQL alone is rarely enough to pass a data scientist interview. SQL is usually the first technical screen, and the later rounds commonly ask for statistics, an A/B test analysis, data manipulation in pandas and sometimes a machine learning question. Treat SQL as the round you must not fail, and give the other rounds a similar share of your preparation time.
Do data scientist interviews include machine learning coding?
Some data scientist interviews include machine learning coding, and it depends on the role. Modeling-focused roles often ask you to implement or explain a model such as linear regression, logistic regression or a decision tree, and to choose and compute an evaluation metric. Analytics-focused roles usually spend that time on SQL, experiments and product questions instead.
Do data scientist interviews include LeetCode-style questions?
Some companies include a general Python coding round in the data scientist interview, and it is usually easier than a software engineer algorithms round. The questions tend to use lists, dictionaries, strings and sorting, with less emphasis on advanced algorithms. Ask your recruiter whether the loop has this round, because many data scientist loops replace it with a pandas exercise.
How is a data scientist interview different from an ML engineer interview?
A data scientist interview tests how you answer questions with data, while an ML engineer interview tests how you build and ship models. Data scientist loops lean on SQL, statistics, experiment design and product cases. ML engineer loops lean on general coding, ML coding from scratch and ML system design.
Is there a free data scientist mock interview?
Yes, TensorTonic has a free Data Science Mock Interview. Each sitting is 75 minutes with five problems that mix SQL, pandas, NumPy, math and ML, and it ends with a scored report. There are four sets, so you can take it again with different problems after more practice.