Explain How Spotify Personalizes Music Recommendations, From Signals to Ranking
Company: Spotify
Role: Product Manager
Category: Machine Learning
Difficulty: medium
Interview Round: Other
In a product manager interview at Spotify, you are asked to explain how Spotify's recommendation system works: how the app decides which songs, artists and playlists to put in front of each listener. Explain it at the depth a product manager needs to work with the engineers who build it, covering what the system learns from, how it chooses what to show, and how you would know whether it is working.
```hint Trace one recommendation end to end
Pick one surface, such as a weekly playlist of tracks the listener has not heard, and follow a single song: what data made it a candidate, how it was ranked against other candidates, and what signal the listener's play or skip sends back.
```
### Clarifying Questions
- Which surface should I anchor on: a weekly playlist of new-to-you tracks, the Home screen, autoplay after an album ends, or search?
- How deep should I go: the main signals and model families, or also the ranking pipeline and how changes are evaluated?
- Should podcasts and audiobooks be included, or only music?
- Should I describe only what Spotify has shared publicly, or may I reason about how I would expect the rest to work?
### What a Strong Answer Covers
- The listener signals the system learns from, explicit and implicit (plays, skips, saves, playlist adds), and how noisy each one is
- The complementary model families, collaborative filtering from listening and playlist co-occurrence and content understanding from audio and text, and why each is needed
- A staged pipeline from candidate retrieval through ranking to re-ranking for diversity and business rules, tied to specific product surfaces
- Cold start for new listeners and new releases, and the balance between exploiting known tastes and exploring
- How a product manager would measure quality: offline and online metrics, long-term satisfaction, and guardrails such as artist exposure and narrowing tastes
### Follow-up Questions
- An unknown artist uploads a debut single with no plays. How could it ever reach listeners?
- After a model change, total plays rose but skips on the weekly discovery playlist also rose. Do you ship it?
- How would you keep recommendations from narrowing a listener to a single genre over time?
Overview: A technical question from a Spotify product manager interview asking you to explain how its music recommendations work. It tests knowledge of listening signals, collaborative filtering and content-based models, the retrieval and ranking pipeline behind personalized playlists, cold start, and how a product manager would evaluate recommendation quality.