Skip to content

Teardown · Music-streaming apps

Discovery in music-streaming apps

Streaming apps must balance comfort and novelty. A look at how home feeds, mixes and search shape what people hear.

Last reviewed

  • Explore vs exploit
  • North Star
  • Kano

Context

Music libraries in streaming apps are effectively unlimited, so the real product is selection: what to put in front of a listener right now. Most listening is familiar music; discovery is the smaller but strategically important share that keeps the catalogue — and listeners' tastes — growing.

Users and jobs

User Job
The commuter "Play something I like, now, with no effort."
The explorer "Help me find something new that fits my taste."
The occasion listener "Give me music for this moment — a workout, focus, a party."

One home screen serves all three, often for the same person on different days.

Flow walkthrough

  1. Home: recently played, personalised mixes, editorial playlists, new releases.
  2. Search: by song, artist, mood or genre.
  3. Playback: radio or autoplay continues from what you chose.
  4. Feedback signals: likes, skips, saves and playlist adds.

What works

  • Recently played at the top serves the commuter immediately.
  • Personalised mixes blend familiar and new tracks, lowering the risk of trying something new.
  • Autoplay turns a single choice into a long session without further effort.

Friction and hypotheses

Friction Hypothesis
Home feed dominated by familiar music Explorers may feel the app is repetitive and use it less for discovery
Skips treated as negative signal regardless of context A skip during a workout may mean "wrong tempo", not "dislike"
Weak explanation of recommendations Users are less likely to try a suggestion they can't relate to their taste

Metrics we'd watch

  • North Star candidate: time spent listening to music the user chose to keep (saves, repeat plays) — closer to value than raw minutes.
  • Discovery health: share of listening from artists new to the user who are saved or replayed within 30 days.
  • Guardrails: session length and skip rate for familiar-music sessions.

If we were the PM

  1. "Because you liked…" labels on discovery rows. Kano lens: a delighter for explorers at low cost. Trade-off: screen space and label quality.
  2. Context-aware skip signals that weight skips differently by activity or time of day. Trade-off: model complexity and explainability.
  3. An explicit "something new" control on autoplay to let the user set their own novelty level. Trade-off: an extra decision for users who want none.