Candidates describe the Google Data Scientist interview as a mix of SQL with window functions, A/B testing and statistics, product and ML cases, and a Googleyness behavioral round. Formats vary by team and level.
$177K
Avg. Base Comp
$320K
Avg. Total Comp
4-5 rounds
Typical Rounds
Not reported
Process Length
Across the reports for Google's Data Scientist interview, the most consistent theme is experimentation and statistics. Several candidates describe designing an A/B test end to end: choosing the success metric, estimating sample size, handling novelty effects and seasonality, and deciding what to do when metrics conflict, such as click-through rising while session time falls. Others report sample ratio mismatch checks, randomization-unit validation and questions on p-values, sampling and the central limit theorem.
SQL appears often, and candidates mention window functions in several reports. Examples include a 7-day rolling average of daily active users, retention metrics, and the time to a user's third transaction. Candidates also say the interviewer wanted to hear the reasoning about definitions and edge cases, not just the query. One report describes writing SQL in a Google Doc-style setup instead of running it.
Product and ML cases tie to Google products. Candidates mention improving or measuring Google Maps features, a YouTube homepage recommender and a YouTube Shorts engagement model. Interviewers reportedly push back on assumptions, so expect to defend metric and model choices out loud. Python is described as light list processing or data cleaning in some reports and as ML-related coding in others. One candidate was asked to design a spam classifier and write bi-gram pseudocode.
Most candidates mention a Googleyness or behavioral round about collaboration, disagreement and leadership. Loops differ by level and team, and outcomes range from offers to rejections, so treat each report as one candidate's loop.
Synthesized from 40 candidate reports by our editorial team.
Had an interview recently?
Share your experience. Unlock the full guide.
Real interview reports from people who went through the Google process.
The whole process took about three weeks, which felt manageable. It started with a standard HR phone screen where they checked salary expectations and my background — pretty straightforward. Then came a technical video call focused on Python and SQL, which I felt okay about.
The tricky part was a 45-minute data structures and algorithms problem during one of the technical rounds. It was a graph problem, solidly in the LeetCode Medium to Hard range. What caught me off guard was how little guidance the interviewer gave. I had to work through the approach, implementation, and debugging mostly on my own, which added to the pressure. There wasn't much room to bounce ideas off them or course-correct if I was heading in the wrong direction.
The final round was a virtual onsite with the hiring manager and two team members. That one was more conceptual — we talked through past projects, machine learning fundamentals, and cultural fit. One question that stuck with me was about precision versus recall in a fraud detection context, which forced me to think concretely about the trade-offs rather than just regurgitate definitions. It was a good question, honestly, because it made me justify my reasoning.
Overall, it felt like a solid interview process, but I didn't make it through. The algorithm round and final round combination was demanding, and I stumbled on execution.
Prep tip from this candidate
Practice graph algorithms thoroughly and expect to solve Medium-Hard LeetCode problems with minimal interviewer guidance — you'll need to talk through your approach clearly and debug independently. Also prepare concrete examples for ML trade-off questions like precision vs. recall in specific business contexts (e.g., fraud detection), not just textbook definitions.
Share your own interview experience to unlock all reports, or subscribe for full access.
Sourced from candidate reports and verified by our team.
Topics based on recent interview experiences.
Featured question at Google
Write a query that returns all neighborhoods that have 0 users.
| Question | |
|---|---|
| 2nd Highest Salary | |
| Top Three Salaries | |
| First Touch Attribution | |
| First to Six | |
| Merge Sorted Lists | |
| Experiment Validity | |
| String Shift | |
| 500 Cards | |
| Last Transaction | |
| Button AB Test | |
| Top 3 Users | |
| Raining in Seattle | |
| Third Purchase | |
| Job Recommendation | |
| Minimum Change | |
| Impression Reach | |
| Jars and Coins | |
| Lazy Raters | |
| WAU vs Open Rates | |
| Find the First Non-Repeating Character in a String | |
| Network Experiment Design | |
| Daily Retention Summary | |
| Bucket Test Scores | |
| Complete Addresses | |
| RMS Error | |
| Find the Index with Equal Left and Right Sum | |
| Find Bigrams | |
| Reducing Error Margin | |
| Delivery Estimate Model |
Synthesized from candidate reports. Individual experiences may vary.
Several candidates report a short recruiter or HR call covering their background, role expectations and logistics. One candidate says it included a Bayesian statistics question, and another says they were emailed straight to a technical screen with no HR call. Formats may vary by team and level.
Candidates report one or two technical screens. Reported content includes SQL with window functions, light Python, probability and statistics, A/B testing and an ML case. One junior candidate describes a scenario-only round with no coding. Expect to explain reasoning aloud, and note that the recruiter's description may not match the interviewer's actual format.
Candidates commonly describe designing an experiment for a feature or thumbnail change. Reported follow-ups cover metric choice, guardrails, sample size, power, novelty effects, seasonality, sample ratio mismatch and conflicting results. Interviewers may keep layering questions onto a single scenario.
Reports describe product cases such as measuring a Google Maps feature, debugging a 10% engagement drop, or framing a YouTube recommendation or engagement model. Candidates say interviewers probe assumptions on data availability, cold start, drift and evaluation, and want a clear problem framing before modeling.
Candidates report a behavioral round near the end of the loop, often called Googleyness. Reported topics include leadership, difficult decisions, disagreeing with a PM and handling pushback on data. Candidates suggest preparing STAR-style stories with measurable impact. Outcomes after the loop reportedly vary widely.
