Netflix Data Scientist Interview Loop
Netflix's loop is long. A recruiter call that covers culture, one or more hiring manager conversations, and a live SQL round. Then a second round of SQL with a case, and an onsite of four to six sessions. Two of those are technical cases, the rest are conversations with the team. The hiring manager decides and needs the whole team to agree. Recent candidates report seven to twelve interviews over two months or more. Read the culture document before the first call.
What decides it
Judgment and how you talk through it. The conversational rounds carry as much weight as the technical ones, and the team has to be unanimous.
Practice by round
SQL on viewing data: title ratings, documentary fans, completion rate by title, new profiles by country and month. The first technical round is pure SQL, so the basics have to be fast and correct.
| Question | Difficulty |
|---|---|
| Abigail Breslin Nominations | Easy |
| Title Ratings | Easy |
| Documentary Fans | Easy |
| Completion Rate by Title | Easy |
| New Profiles by Country and Month | Easy |
| Titles Never Rated | Easy |
| Average Rating by Genre | Easy |
| Longest Session by Genre | Easy |
Harder SQL and Python, like heaviest viewers in a month or minutes watched by content type. The second technical round pairs a query like this with a case, so practice moving from the numbers to a point of view.
The product cases: success for a recommendation row, testing personalized row ordering, falling completion for new titles, and whether to show a Trending Now row. The real onsite case is an experiment to design and power. Treat each of these as a design question rather than a metrics list. The mini-analysis on new release watch time and the data project are the hands-on version.
| Case study | Difficulty |
|---|---|
| Success for a Recommendation Row | Medium |
| Falling Completion for New Titles | Medium |
| Should We Show Trending Now? | Medium |
| Testing Personalized Row Ordering | Hard |
| Data project | Difficulty |
|---|---|
| New Shows and Movies | Easy |
| New Release Watch Time | Medium |
Statistics for experimentation and causal inference: what R-squared does and does not tell you, judging whether a new algorithm is better, detecting drift in a recommendation model, and a synthetic control for a change in a single market. Netflix data science leans toward causal inference, so expect questions that go past a plain A/B test.
| Question | Difficulty |
|---|---|
| Better Algorithm | Easy |
| Rate a Movie | Easy |
| Interpreting R-Squared | Easy |
| Detecting Drift in a Recommendation Model | Medium |
| Synthetic Control for a Single-Market Change | Hard |