Rethinking Spotify Search
Hugo Galvão and Daniel Doro document Spotify’s three iterations of search team organization as the team grew 5× between 2016 and 2021.
Evolution
Phase 1: Problem-Oriented Teams (2019)
Teams organized around specific user problems. Issue: problems are transient — new problems arise quarterly, forcing constant reorganization.
Phase 2: Stack-Oriented Teams (2020)
Each team owns one layer of the search stack (data ingestion, quality/ML, API, insights, podcast search). Revealed the centrality problem: cross-stack features required coordinating 5 different teams. High congestion + high centrality → slow delivery despite team growth.
Phase 3: Platform + Core (2021)
Two groups:
- Core Search — personalized search experience; metric: Spotify end-user satisfaction
- Search Platform — developer-facing infrastructure; metric: Spotify developer happiness + SLO compliance
Rationale: different users (end users vs. internal developers), different success measures, but shared mission. Platform autonomy enables other teams to experiment without coordinating with core search.
Key Data Points
- 2021 internal contribution teams: 100%+ growth vs. 2018, 500%+ vs. 2016
- System congestion and centrality metrics guided the decision to restructure
Key Insight
Growth in headcount does not translate to velocity if coordination costs grow faster. The platform model decouples infrastructure development from product experimentation.