何韦葶
Flow 3.0: Giving Users More Control Over Music Recommendations
I led the redesign of Flow, Deezer’s flagship personalized listening experience, to address a recurring trust issue: users felt recommendations were random, repetitive, and hard to influence. This project focused on restoring trust in recommendations by making Flow more understandable, adjustable, and aligned with users’ listening intent.
Scope
Recommendation feature & Engagement
Team
Product, Engineering, Global Editorial Team
Overview
a Trust Gap Between Users and Flow
Flow is Deezer’s flagship personalised listening algorithm averaging 1 million users per day.
Despite strong adoption, internal research showed declining trust and long-term engagement due to perceived randomness and lack of control.
User interviews highlighted recurring friction:
- Genre and mood inconsistency within a single session
- Overrepresentation of already-known tracks
- No way to steer or correct recommendations
- (~30 qualitative interviews)
These issues weakened perceived relevance and reduced repeat usage.
Before
After
Before
After
Context & Constraints
Flow felt like a black box
Users couldn’t understand why music was recommended, nor influence it in the moment, leading to disengagement instead of exploration.
Context
- Flow had to remain fast, lightweight, and emotionally engaging
- Used by both passive listeners and power users
- Deeply coupled with recommendation algorithms
Constraints
- No complex settings or heavy configuration flows
- Minimal interruption to playback
- Changes had to scale across millions of sessions
Key Design Moves
Designing for Intent, Not Configuration
The solution wasn’t more settings, it was meaningful, lightweight control. Users don’t want to configure recommendations. They want to nudge them. Control needed to be: lightweight, contextual and reversible.
- Anchor Flow Sessions with Intent
Added genre selection in addition to moods giving visual feedback to the user. This clarified the direction of the algorithm.
- Make Discovery Explicit
Added a toggle for users to choose between: exploring new music or listening to favorites.
This surfaced recommendation logic.
- Direct visual feedback
Users now have the power to shape their listening sessions like never before, allowing them to feel fully in control of their experience. This influence enhances engagement and personal connection, making each session uniquely tailored to their preferences.
Outcome
Users can now enjoy greater control over their flow, with more genres, favorites, and discovery options.
Postsoft launch results (30 days):
+10% increase in songs saved from Flow
+40% increase in regular Flow listenersQualitative feedback confirmed improved trust:
“Now I feel like Flow understands what I want right now.”
By shifting Flow from a predictive system to a collaborative one, the redesign: increased engagement without adding complexity, improved perceived relevance and discovery, restored trust in personalization.
Claire Lecerf ⓒ 2026
Get In Touch
何韦葶
Flow 3.0: Giving Users More Control Over Music Recommendations
I led the redesign of Flow, Deezer’s flagship personalized listening experience, to address a recurring trust issue: users felt recommendations were random, repetitive, and hard to influence. This project focused on restoring trust in recommendations by making Flow more understandable, adjustable, and aligned with users’ listening intent.
Scope
Recommendation feature & Engagement
Team
Product, Engineering, Global Editorial Team
Overview
a Trust Gap Between Users and Flow
Flow is Deezer’s flagship personalised listening algorithm averaging 1 million users per day.
Despite strong adoption, internal research showed declining trust and long-term engagement due to perceived randomness and lack of control.
User interviews highlighted recurring friction:
- Genre and mood inconsistency within a single session
- Overrepresentation of already-known tracks
- No way to steer or correct recommendations
- (~30 qualitative interviews)
These issues weakened perceived relevance and reduced repeat usage.
Before
After
Before
After
Context & Constraints
Flow felt like a black box
Users couldn’t understand why music was recommended, nor influence it in the moment, leading to disengagement instead of exploration.
Context
- Flow had to remain fast, lightweight, and emotionally engaging
- Used by both passive listeners and power users
- Deeply coupled with recommendation algorithms
Constraints
- No complex settings or heavy configuration flows
- Minimal interruption to playback
- Changes had to scale across millions of sessions
Key Design Moves
Designing for Intent, Not Configuration
The solution wasn’t more settings, it was meaningful, lightweight control. Users don’t want to configure recommendations. They want to nudge them. Control needed to be: lightweight, contextual and reversible.
- Anchor Flow Sessions with Intent
Added genre selection in addition to moods giving visual feedback to the user. This clarified the direction of the algorithm.
- Make Discovery Explicit
Added a toggle for users to choose between: exploring new music or listening to favorites.
This surfaced recommendation logic.
- Direct visual feedback
Users now have the power to shape their listening sessions like never before, allowing them to feel fully in control of their experience. This influence enhances engagement and personal connection, making each session uniquely tailored to their preferences.
Outcome
Users can now enjoy greater control over their flow, with more genres, favorites, and discovery options.
Postsoft launch results (30 days):
+10% increase in songs saved from Flow
+40% increase in regular Flow listenersQualitative feedback confirmed improved trust:
“Now I feel like Flow understands what I want right now.”
By shifting Flow from a predictive system to a collaborative one, the redesign: increased engagement without adding complexity, improved perceived relevance and discovery, restored trust in personalization.
Claire Lecerf ⓒ 2026
Get In Touch
何韦葶
Flow 3.0: Giving Users More Control Over Music Recommendations
I led the redesign of Flow, Deezer’s flagship personalized listening experience, to address a recurring trust issue: users felt recommendations were random, repetitive, and hard to influence. This project focused on restoring trust in recommendations by making Flow more understandable, adjustable, and aligned with users’ listening intent.
Scope
Recommendation feature & Engagement
Team
Product, Engineering, Global Editorial Team
Overview
a Trust Gap Between Users and Flow
Flow is Deezer’s flagship personalised listening algorithm averaging 1 million users per day.
Despite strong adoption, internal research showed declining trust and long-term engagement due to perceived randomness and lack of control.
User interviews highlighted recurring friction:
- Genre and mood inconsistency within a single session
- Overrepresentation of already-known tracks
- No way to steer or correct recommendations
- (~30 qualitative interviews)
These issues weakened perceived relevance and reduced repeat usage.
Before
After
Before
After
Context & Constraints
Flow felt like a black box
Users couldn’t understand why music was recommended, nor influence it in the moment, leading to disengagement instead of exploration.
Context
- Flow had to remain fast, lightweight, and emotionally engaging
- Used by both passive listeners and power users
- Deeply coupled with recommendation algorithms
Constraints
- No complex settings or heavy configuration flows
- Minimal interruption to playback
- Changes had to scale across millions of sessions
Key Design Moves
Designing for Intent, Not Configuration
The solution wasn’t more settings, it was meaningful, lightweight control. Users don’t want to configure recommendations. They want to nudge them. Control needed to be: lightweight, contextual and reversible.
- Anchor Flow Sessions with Intent
Added genre selection in addition to moods giving visual feedback to the user. This clarified the direction of the algorithm.
- Make Discovery Explicit
Added a toggle for users to choose between: exploring new music or listening to favorites.
This surfaced recommendation logic.
- Direct visual feedback
Users now have the power to shape their listening sessions like never before, allowing them to feel fully in control of their experience. This influence enhances engagement and personal connection, making each session uniquely tailored to their preferences.
Outcome
Users can now enjoy greater control over their flow, with more genres, favorites, and discovery options.
Postsoft launch results (30 days):
+10% increase in songs saved from Flow
+40% increase in regular Flow listenersQualitative feedback confirmed improved trust:
“Now I feel like Flow understands what I want right now.”
By shifting Flow from a predictive system to a collaborative one, the redesign: increased engagement without adding complexity, improved perceived relevance and discovery, restored trust in personalization.
Claire Lecerf ⓒ 2026
Get In Touch