'For You' Recommender in an Audio and Podcast App
An audio app with 6M users and 400,000 episodes observes that 80% of listening is concentrated in 2% of the catalogue. Users who do not find something new in their first week leave. The Head of Product wants a 'For you' screen that increases minutes listened and 30-day retention.
Where the money comes from
What to do now
- 1Know the cost per minute of each type of content.
- 2A/B test of the "For you" screen versus the current home page.
- 3Add content margin and diversity to the ranking.
For your committee
One page with the decision, the impact, the risks, the steps and the KPI to track.
The print dialog opens: choose "Save as PDF".
Watch out: more volume is not always more profit
More minutes is not always more profit: if you recommend content with expensive royalties, cost grows with usage. Balance retention against the cost per minute of the recommended content.
KPI for the committee
D30 retention and margin per active user.
Detail for your data team
Proposed technique: Collaborative Filtering
Why: With millions of interactions, collective behaviour is the best signal: users with similar listening share tastes. Collaborative filtering (matrix factorisation) uncovers latent tastes that no metadata describes and scales to huge catalogues.
Cost of getting it wrong: Bad recommendations = short sessions and early churn. Recommendations that are too safe = a bubble that reduces discovery and catalogue diversity.
Data needed: Listening events: user, episode, seconds listened, whether finished, skips.; Saves, shares and show subscriptions.; Episode metadata: category, length, language, transcript.