top of page
Spotify Redesign
While Song Radio uses behavioral data to generate recommendations, it fails to predict how user preferences can change over time. To better capture these unexpected shifts in behavior, I incorporated both behavioral and explicit data in the recommendation algorithm. Behavioral data refers to all listening behaviors on the app, while explicit data refers to all feedback or reports sent by users. By creating a balance between the two, recommendations can accurately reflect user needs, even as they change in real-time.
Details
Role
UX Researcher
Tools
Qualtrics, Jamboard, Figma, Teams
Problem
Unhappy Users
I conducted an online survey using Qualtrics to understand how users felt about Song Radio recommendations.
-
​Respondents were unhappy with 44% of recommendations.
-
Respondents disliked 39% of recommendations that did not share characteristics, while they disliked 5% of those that did.
-
89% of respondents wanted more control over recommendations.
-
Respondents focused on the artist, genre, instruments, and other characteristics when deciding whether or not they liked a recommendation.
The Current Algorithm
While Spotify relies on behavioral inputs for recommendations, like searching, listening, skipping, and saving, it does not take into account spontaneous and short-term changes in user behavior. As a result, it fails to give recommendations that match the evolving needs of users.​​​​​​​
How do Spotify recommendations work?
Solution
Humanizing The Algorithm
Integrating both behavioral data and explicit feedback in the algorithm could generate more flexible recommendations + revolutionize the music landscape as it is not common practice to collect user feedback in 2024.
Provide Commentary
As users receive recommendations based on their behavior, they would have the ability to report their preferences and directly alter the algorithm's output.
Express Yourself
Not only would this address habitual and spontaneous aspects of human behavior, but it would also allow users to communicate their needs at any given moment.
Approach
Introducing Filters
One way to add explicit feedback to Song Radio is by integrating filters into the system. This means restructuring the algorithm/navigation and redesigning the user interface for compatibility.
Phase 1: Structure
Navigation
Organizing The Flow
To receive recommendations, I created a flow where users can pick a song, choose filters, submit their preferences, and listen to recommendations.
Search
Select
Submit
Listen
Defining The Filters
I created two sets of filters: one with sound types and another with song characterics.
Tailoring The Labels
I conducted hybrid card sorting sessions to create smaller, identifiable groups.
Phase 1.5: Structure Algorithm
Building The Anatomy
I drafted a sample structure of the new algorithm that outlines all possible scenarios, filters, definitions, and pathways.
Phase 2: Design UI
Establishing Guidelines
I developed two personas to inform my decision making, one for new users and another for active ones.
I recently signed up for Spotify to discover new music. Before I joined the community, I would listen to top radio hits, but now I want to expand my interests. I am open to all types of music and I think Spotify's Song Radio will help refine my music taste.
New User
Behaviors
-
Listens to the radio
-
Expresses curiosity
Needs
-
To understand how Song Radio works
-
To find Song Radio tab on the app
Wants
-
To learn about new music
-
To find his style
I have been using Spotify for over two years. I enjoy listening to music every day and sometimes use Song Radio to see what's out there. I don't often get the urge to explore new music, but I'm still open to surprises.
Active User
Behaviors
-
Uses Spotify daily
-
Uses Song Radio occasionally
-
Expresses flexibility
Needs
-
To understand how the new update works
-
To navigate the interface with ease
Wants
-
To find unexpected releases
-
To discover music at their leisure
Recalling the behaviors, needs, and motivations of my target audience helped me establish ground rules for the design process.
Easy To Find
Easy To Understand
Appeal To All Users
Sketching My Ideas
I brainstormed different ideas for the UI, staying true to Spotify's identity.
Mapping The Path
I created a task flow, illustrating four unique scenarios on Song Radio:
Users can choose between (1) getting started, (2) watching a tutorial, (3) viewing saved content, and (4) browsing their listening history.
Phase 3: Test Prototype
Measuring Efficiency
I conducted 3 sessions of moderated usability testing with new, regular, and expert Spotify users. Participants were asked to complete a series of tasks via Google Meets.
Expert User
Regular User
New User
Task 1: I want to learn more about Song Radio.
Task 2: I want to use Song Radio.
Task 3: I want to find my Song Radio history.
Task 4: I want to find my Song Radio saved songs.
Making Discoveries
I extracted themes and patterns from the behavioral data, uncovering the following insights:
17% of all pages were problematic.
-
Users had a hard time interpreting new signs, symbols, and buttons.
-
Users repeatedly got stuck in the filtering process, unsure of what to do next.
-
Task completion and ambiguous design were the most challenging issues.
-
Filters page included the most issues.
More than 50% of issues were linked to new users.
-
Those unfamiliar with Spotify's interface struggled more.
Conclusion
Moving Forward
-
It's important to reference the existing product's brand identity because it’s easily recognized and understood by all users.
-
Comprehensive instructions are necessary for onboarding users, including tooltips, tutorials, get help buttons, and more.
bottom of page