Back to cases
SpotifyProduct SenseHard~35 min

Recommendation System Improvement

Spotify's recommendation system drives 35% of all listening hours. The ML team has a new model that improves click-through rate by 8% but reduces the diversity of recommended artists by 22%. The team is debating whether to ship it. You need to evaluate tradeoffs beyond CTR, define what 'good recommendations' means for a music platform, and propose a framework for balancing exploitation (playing what users already like) with exploration (helping users discover new music).

Duration
~35 minutes
Difficulty
Hard
Free
Your first session · then $3 per session
  • ~35 minute AI-powered mock interview
  • Realistic interviewer pushback
  • Stage-by-stage debrief with ratings
  • Signal coverage map
  • Reference response for comparison

Sign up free with Google · No card required

Skills tested
Tradeoff evaluation beyond single-metric optimizationExploration vs. exploitation framework for recommendationsMulti-stakeholder analysis (users, artists, platform)Long-term vs. short-term metric tension
Interview stages
1
Evaluate whether the CTR improvement is truly positive
2
Design the exploration vs. exploitation framework
3
Propose how to ship the improvement responsibly
What you'll get back

After the session, you'll get a detailed debrief: stage-by-stage ratings (Strong / Developing / Weak), a signal coverage map showing which key points you hit, specific feedback on your reasoning and communication, and a reference response to compare against.