Creating Personal Playlists using Machine Learning
Daily Mix by Spotify / 2016 - 2017
Role: Product Design Lead
Descended from Radio, Daily Mix is our take on user's music taste, organised through 1 or more bottomless playlists depending on the variety of music you stream. Mixes are regularly updated and evolve along with your listening habits. Daily Mix makes it simple to quickly tune into – and tune between – user's favourite stuff, any day of the week. We keep similar artists grouped together, no maintenance required, and we throw in some discovery tracks too.
Press
"Spotify Just Made The "Shuffle" Button A Thousand Times Smarter" - FastCompany
"Spotify’s new Daily Mix promises a never-ending, never-boring playlist" - The Next Web
The design challenge
In 2016, despite the growing popularity of Spotify, traditional Radio & Pandora were still our biggest competitors
Project Overview:
Goals & Objectives
Together with Product and Engineering teams, we came up with some key goals and objectives that will help grow the business.
People Goals:
Get people to listen to more music (Consumption)
Get people to keep using Spotify over time (Retention)
Get people to love our product over time. (Happiness)
Business Goals:
Teach our algorithms about what people love and how people listen to music — to help improve our recommendations service
Provide an alternative product to compete with traditional radio and Padora
Research
For this project we did a couple of studies before going to the drawing board to help us understand our users more and to help us make more informed decisions
The research we did includes
⟶ Field Studies (Atlanta, USA)
⟶ Diary Studies
⟶ Lab Testing (of our Spotify and competitor apps like Pandora)
What we found out
People love to discover and listen to music with as little effort as possible just like radio. However, it’s so difficult to do this in Spotify.
🚩 This is just one of my early sketches of Daily Mix (2016)
Personal, Easy, Infinite
After the initial research, we felt that we could make an informed decision what we think is (1) the right problem to solve (2) the right hypothesis and (3) the right solution for the problem.
Our hypothesis
Play and discover music that what like without having to think about it.
🚩 One of the first prototypes we tested in the product. This was the bare minimum we could do. Immediately, we saw some usability issues when we tested these in the lab (e.g. The navigation was confusing - people could find how to get back to the 2nd screen here)
🚩 We kept improving the usability and visual design as we kept learning more from our users. We also working brand to come up with the name "Daily Mix" and unique cover art.
🚩 We also looked at Feedback Models — how users can tell us what they like and done like. We ended up using the ❤️and 🚫because it was straightforward and easily understood.
What has happened since?
This was the final product we eventually launched to the public in 2016. We continued iterating and evolving until 2017. The results were really good and it is one of our most beloved and well-known products at Spotify.
💪 Daily Mix is one of our most used products
💪 We increased engagement by 26pp & retained new users
💪 To this day, Daily Mix powers Spotify’s machine learning algorithms to improve our recommendation sevice
💪 More personal music listening for people all over the world