Google Data Scientist Interview Loop
The Google data scientist interview has a recruiter call, a technical screen in a shared doc, not a code editor, and an onsite of three to five 45-minute interviews. Team matching comes after you pass. It has the most statistics of the big platforms, and the follow-up is what sets it apart. Every interviewer scores how you think as well as what you know, so naming a method only starts the question.
What decides it
Depth when the follow-ups come. The candidates who don't get through can usually name a method but can't explain it.
Practice by round
SQL basics at speed, with no autocomplete and nothing to run your query against. Think language speakers per country, logins by account, or the oldest survivor per class. The real screen also gives you a dataset to explain. Scope it before you answer, because the questions you ask are marked as much as what you write.
Medium and hard SQL: percentiles and medians of hours studied, making a friends network symmetric, flags per video, the longest streak per player, and fare differences through a self-join. This is the level of both the screen and the onsite coding round, so practice here carries over most directly.
You answer a business problem with a measurement plan. That might be success metrics for Shorts, recommended news or a storage plan, two metrics that disagree, or a campaign where you define success before you measure it. Treat it as a conversation. Senior candidates lose marks most often for solving before they've scoped.
| Case study | Difficulty |
|---|---|
| Translator Improvement | Easy |
| Recommended News Metrics | Medium |
| Shorts Success Metrics | Medium |
| Storage Plan Usage | Medium |
| Two Metrics Discrepancy | Medium |
Statistics and probability. You'll be asked how a method works, not only when you'd use it. Topics include sampling distributions and Type I error, standard error and margin of error, F1 from a confusion matrix, spam-filter probabilities, and coin tosses and elevators. The real round adds an experiment-design conversation, so practice it out loud.