Spotify runs hundreds of experiments at any given time, so they need techniques to coordinate experiments too, for example, to ensure that users cannot be in two experiments at the same time. Most specifically, Spotify requires experiment coordination strategies that enable programs of exclusive experiments (i.e. sets of experiments that are exclusive to each other). In this post, Maarten Schultzberg, Oskar Kjellein, and Johan Rydberg describe how bucket reuse (where unique user IDs are hashed into buckets upon which random sampling is performed) enables such coordination – allowing Spotify to run more experiments with compromising quality. They also mention other initiatives, like a tool to review the history of available buckets, which will enable more statistically rigorous experimentation.