Root-Causing a 25% Fall in Spotify Song Streams

Spotify's song listens are down 25%. As the product manager responsible for listening, how would you work out what caused the drop? ```hint Pin down the shape of the drop Ask what exactly counts as a listen, which period the 25% is measured against, and whether the fall happened overnight or slowly. Each answer rules out whole groups of causes before you look at user behavior. ``` ### Clarifying Questions - Is the 25% a fall in total song streams, in streams per active listener, or in minutes of music played, and is it compared with the previous few weeks or with the same period last year? - Did the decline appear on a single day or build over several weeks, and is it worldwide or concentrated in certain markets or platforms? - Does the metric count every play start, or only plays that pass a minimum listening time? - Are podcasts and audiobooks excluded, so that listening time moving to them would show up here as fewer song listens? ### What a Strong Answer Covers - Confirming the drop is real before diagnosing behavior: the metric definition, logging and pipeline changes, counting-rule changes and the choice of comparison window - Breaking song listens into active listeners, sessions per listener and songs per session, and finding which factor moved - Segmenting by platform and app version, market, free versus Premium, new versus existing listeners and listening surface (playlists, search, autoplay, connected devices) - Internal causes (releases, experiments, recommendation changes, outages, catalog removals) and external ones (seasonality, competitor moves, operating-system changes), each paired with the data that would confirm or rule it out - An ordered plan of checks, a conclusion tied to evidence, and the fix and prevention for the likely cause ### Follow-up Questions - The drop turns out to be limited to free-tier Android listeners on the newest app version. How do you prove the release caused it, and do you roll back before you are sure? - Active listeners are flat but songs per session fell sharply. Which parts of the product do you examine first, and why? - Podcast and audiobook listening rose by about as much as song listens fell. Is that a problem to fix or a shift to accept?