Big Tech platforms Data Scientist Interview Loop
Most Big Tech data scientist interviews follow the same outline. There's a recruiter call, then a technical screen that's SQL with a product or statistics question attached. Then comes an onsite of four to six 45-minute interviews: SQL and coding, statistics and experiments, a product case, and behavioral. What changes is who has the final say and what gets the most weight. Amazon tests Leadership Principles in every round. Microsoft brings in a senior interviewer who can outweigh strong technical rounds. Google decides by committee and goes deepest on statistics. Meta puts product judgment above math. Apple depends on the team.
Practice by round
Every one of these screens is SQL basics under time, in a plain editor with no autocomplete. Nobody fails on a hard question. They fail by being slow on an easy one.
Retention, engagement, funnels and cohorts, using window functions and CTEs. Then come the follow-ups. What could confound this metric, and what would you check before you believe it? Amazon and Google also ask how your query holds up on a much bigger table.
| Question | Difficulty |
|---|---|
| Share of Active Users | Medium |
| Users Exclusive Per Client | Medium |
| Total Sales In Different Currencies | Medium |
| Daily Active Users | Medium |
| Manager of the Largest Department | Medium |
| Successful Lower Priced Product | Medium |
| Top Three Classes | Medium |
| Same Brand Purchases | Medium |
| Recommendation System | Medium |
| Blocked Users | Medium |
| World Tours | Medium |
| Retention Rate | Hard |
| Consecutive Days | Hard |
| Player with Longest Streak | Hard |
| Popular Posts | Hard |
The product case. You define a success metric for a product, decide whether to launch, judge success for Shorts or a notification system, or work out why two metrics disagree. Meta scores product thinking above everything else here. Amazon sets the case in its own business and expects you to put a number on the impact. Google wants to see how you scope the problem.
| Case study | Difficulty |
|---|---|
| Translator Improvement | Easy |
| Skype Usage | Easy |
| Recommended News Metrics | Medium |
| Shorts Success Metrics | Medium |
| Storage Plan Usage | Medium |
| Two Metrics Discrepancy | Medium |
| Close Friends Metrics | Medium |
| Ad Success Metrics | Medium |
| Best Friend | Medium |
| Health of Groups | Medium |
Statistics and probability: A/B testing, sample size and test length, sampling distributions and Type I error, skewed data and outliers, and z versus t. Google keeps asking until you explain how things work. Amazon's version is broader and shallower. The real round may also ask what happens when the two test groups affect each other, so practice that too.
| Question | Difficulty |
|---|---|
| Definition of Variance | Easy |
| Margin of Error | Easy |
| Conversion Rate | Easy |
| Negative R Squared | Easy |
| Precision and Recall | Easy |
| Standard Error of the Mean | Easy |
| Z and t distributions | Easy |
| Outliers Detection | Easy |
| Assessing Multicollinearity | Easy |
| Bias-Variance Tradeoff | Medium |
| Panel vs Cross-sectional Data | Medium |
| Right vs Left-Skewed Distributions | Medium |