Routines that fit the day
Rainy Manhattan, office focus
Contextual rows adapt to location and schedule: commute playlists for rainy mornings, focus tracks for the workday.
For Ü
Making Amazon Music Part of Everyday Life Through Personalization
Problem
How might we make users feel that Amazon Music really understands them?
Retention depends on whether casual listeners feel the service fits into their lives. When For You recommendations miss the mark, Amazon Music becomes interchangeable with any other streaming app.
Design a customer experience strategy that improves retention across tiered offerings and makes personalization feel human, not algorithmic.
29% disengaged
Among lapsed casual users, 29% cited poor recommendations as a reason for leaving.
75% felt competitors won
Among lapsed users, 75% felt Spotify, Apple Music, or YouTube Music offered stronger personalization.
Casual listeners matter most
Active and passive casual listeners form the largest segment, making them the greatest opportunity for improving retention.
Research
Two kinds of casual listeners
We surveyed users and mapped behaviors to two personas. Both stream casually, but they relate to music differently.
Active Listener
“Music bridges me, myself, and others.”
Wants recommendations that reflect taste, mood, and identity, not just play history.
Passive Listener
“Music is just there in the background of my life.”
Needs context-matched playlists without searching or building libraries.
Recommendations feel generic
For You rows that repeat artists or ignore recent activity feel like guessing.
Ecosystem data is underused
Shopping, reading, and viewing signals rarely surface in Music in ways users trust.
Competitors lead on control
Spotify and Apple Music offer clearer taste levers; Amazon has the data but not the experience.
Strategy
For Ü: personalization that knows you
For Ü connects Amazon's ecosystem to everyday listening: cross-service recommendations users can trace to their data, routine-based playlists, and Maestro AI for frictionless creation.
Transparent by default
Users choose which Amazon services feed recommendations and can adjust anytime.
Context over catalog
Playlists respond to mood, weather, and daily routines, not just listening history.
Creation without friction
Maestro Beta generates tracklists from text or image prompts; users control saving and refining.
Design Pillar 01
Cross-ecosystem recommendations
Amazon already has signals about what users read, watch, and shop for. With explicit consent, For Ü makes selected signals visible inside Music and explains how they shape recommendations.
Design Pillar 02
Routine-based For You
Passive listeners open Music because the moment needs a soundtrack. For Ü organizes the For You tab around daily context: weather, commute, work rhythm, and emotional cues.
- Routines that fit the day
- Quote-based recommendations
Design Pillar 03
Maestro Beta: AI playlist creation
Active listeners want to shape their sound without building playlists track by track. Maestro Beta accepts text or image prompts and keeps final control with the listener.
- Create with prompt or image
- Combine image and prompt
- AI reads the mood
- Preview before saving
- Saved to your library
- Share your daily pick
Create with prompt or image
Text + visual input
Users pick an inspiration card or type a prompt (city walk, mood, genre) to start generation.
Impact
Five participants rated their likelihood of returning at 7.8/10 the next day and 7.0/10 within the next month.
We interviewed five participants to evaluate the design's retention potential.
“I would return to the app to find soundtracks from shows I've recently watched.”
“It's cool to create playlists with AI. I would like to try different photos.”
“I like being able to view and edit my routine because I usually choose music based on the situation.”
“Sometimes, images say what I can't put into words.”
Conclusion
Making Amazon Music feel indispensable
For Ü treats personalization as trust: show where recommendations come from, match music to daily rhythms, and offer AI creation without removing agency.
This was a four-week concept project. Next steps would include testing whether recommendation settings feel transparent, validating routine-based triggers with listening data, and aligning the Maestro concept with production saving flows.
