How Spotify Uses GenAI and ML to Annotate a Hundred Million Tracks

Spotify built a scalable annotation platform to label over 100 million tracks by combining human expertise with GenAI automation. The platform features a three-tier workforce structure (core annotators, quality analysts, project managers), flexible tooling for multimodal tasks, and infrastructure that integrates directly with ML pipelines. This hybrid approach increased annotation throughput by 10x while improving quality through structured escalation paths and agreement scoring, enabling faster ML model development and iteration cycles.