On Spotify, listening is everything. With their recent launch in Nigeria, they wanted to understand the most popular African music genres and subgenres within Nigeria. I was tasked with discovering the most popular African music genres and creating an experience that allows users to teach the Spotify Algorithm how to categorize songs across these genres.
In conducting heuristic evaluation, insights on how songs are currently be categorized was identified and these were the key insights;
#1 Songs are categorized by a combination of genres, moods, trends, events etc.
#2 There is no place on the app where you can categorically identify the genre a song falls under.
Conducted market research with Ten participants to understand the most popular African genres known and also how they listen to music. The interview gave rich qualitative data.
We have over 17+ African genres, with Afrobeat being the most popular within Nigeria with sub genres such as Afro pop, Afro soul and Afro fusion etc.
All ten of our users are not conscious of African music genres because they create playlist based on mood, but become conscious when there is an interruption while listening to a playlist or recommendation. This led me to believe that users can identify music genres while listening.
HOW MIGHT USERS TEACH THE ALGORITHM TO CATEGORIZE SONGS ACROSS GENRES.
The idea was to design a gamification experience where we utilize trivia questions for the identification of the genres and for every correct response users get points with the top trivia winners ranked on a global leaderboard.
#APPROACH 1: FOLLOW UP ON A LIKE INTERACTION
Considering people's attention span are so short and often times after playing a song they might exit the application and get into an activity.
#APPROACH 2: UTILIZE A BUTTON ON THE NOW PLAYING VIEW
The now playing view contains a lot of details about a song and if a user would want more information about a song, that is the first call to action.
During my brainstorming session, I explored several concepts and during a discussion session with our stakeholders, we discovered that presenting users with reward options would get them biased and might be the cost to the business. I finally decided on two variants which usability testing was conducted on for initial response.
These are the key insights after conducting usability testing with four different users;
"It will correct the errors must music apps are currently making
" it would create a fairly accurate grouping of songs
"Earning points would get me more engaged on the App
"A song trending is not enough reason to add it to my playlist
To align with our user’s expectations and the business goals, I redesigned and expanded upon the original idea to have more use cases. We moved the design to high fidelity and carried out two rounds of usability testing with four users each. These were the main insights discovered.
The initial idea was a leaderboard experience where users are ranked based on their points but considering it's a voluntary exercise, less competitive users might abandon it and the goal is motivation. Some users liked the experience that doesn’t reveal their position because a leaderboard experience where they are stacked against others will decrease their motivation but others felt it would make them more competitive. For users with less competitive spirit, I decided on highlighting just the top players to encourage participation.
The initial idea was to reward users with premium access equivalent to their points but I discovered that presenting users with reward options would get them biased and might be the cost to the business. After much research, it was discovered users liked bragging rights on social platforms and it was a great opportunity.
The first approach was following up on a like song interaction but after research it was discovered that users who would want to go through the experience again might constantly be adding and removing songs from their favorites which would affect the current flow. This led to adding an entry point on the now playing view.
It was challenging to find a way to present the experience to satisfy both users' wants ( Getting value or reward from the experience) and stakeholders' wants ( training the algorithm).
The goal was to give the users the value of the experience up front to help drive clicks towards the experience.
The goal with this iteration was to balance the user need with the business need and also encourage user motivation.
I want to commend you for using the user feedback to validate and invalidate your ideas cause I can see your iterations and versions are quite different. You surfaced the opportunity within the existing experience of the app which is really important, like you’re taking the experience that the listener has and kind of incorporating it.
( Ashley Graham - Senior Researcher at Spotify)
75%
of users that tested the prototype, found this entry point efficient. The follow up interaction after liking a song got them engaged on the pop up at first contact.
45%
of users that tested the prototype, found this entry point unique and an insightful way to learn the genres of the current song playing.
50%
of users that tested the prototype, said they liked that the value of the experience was highlighted up front.
It was really a great experience to work on this project. Despite the focus was on African genres, I believe the experience could be applicable to all genres on the App. One of the key learnings for me was to bin my assumptions whenever I felt my own confirmation bias coming into play. Looking back ,one thing I'd definitely do differently is consider other Spotify users like the artist and advertisers, and not only listeners like I did. A mix badge will offer several perspectives, opportunities and a wider range of learnings. Another key learning is that users are not the best to teach the algorithm but can validate the authenticity of the genre.
There are so many ways to teach the algorithm and I would love to explore other directions. I'm imagining an experience were users select the instruments they can identify from a particular song because every genre has there own sequence and also standard instruments that make up the beat. Taking Afro as an instance the components are Brass, Conga, Guitar and shakers. Allowing the users identify the instruments and having the algorithm match this responses against an already defined component tag on the database.
