In tune with the data - how Spotify uses data to know your music tastes better than you do.
Exploring Spotify's success with utilizing user data to power engagement across their products.
I spend a significant amount of time on Spotify – curating playlists, discovering new music, and seeing what my friends are listening to. Spotify has done a great job of engaging their users by surfacing data in an interesting way, whether it be an ever-expanding set of personalized playlists at our fingertips or my favorite holiday of the year Spotify Wrapped.
Spotify said it best — “The beauty of these experiences is our ability to deliver the right piece of music for that exact moment in time, and maybe even connect you with your next favorite artist in the process.” They've made the connection of using the story that data tells to not only recommend more songs within their application, but even get users attending concerts, buying merchandise, and sharing insights to social.
Today, we’ll talk about how Spotify uses data to power its products, how you can generate your own Discover Weekly inspired playlist, and the new AI driven product they are releasing.
Spotify uses data by:
Utilizing a users’ listening history to better understand preferences then surface new songs and artists through products like Discover Weekly.
Sharing user data back to users through products like Wrapped, Blends and Event Recommendations for relevant performances.
Delivering daily, personalized playlists for every user across a plethora of categories.
Knowing your user better than they know themselves
There are times Spotify recommends me a song and I say, wow, I never would have thought to listen to a song like this. This is one of my favorite “aha!” moments in the beauty that is storytelling through data. My listening history exposes a story that even I don’t fully know I’m telling. As a user, you have a multitude of data points that Spotify can learn from – what music you listened to today, what music you listened to a year ago, what songs you like and don’t like, how long you listen to a song before you skip it, how many times you play a song after hearing it for the first time, what your friends are listening to, and so much more. Spotify is learning through these data points and building models to predict what you'll listen to next, which songs you’ll listen to the most of, and which songs they shouldn’t recommend. By knowing a user better than they know themselves, you generate trust with your user that the “method behind the madness” works, and isn’t just spitting out nonsense in hopes of driving engagement.
A set of researchers took a deep dive into user behavior by analyzing a trove of data from Spotify. Some interesting areas they looked into included how users switch between devices, what time of day users listen the most, and more. These metrics continue to uncover the story of Spotify user’s everywhere.
Read more here.
Speaking of storytelling – where is Spotify’s “a year ago today you listened to Driver’s License by Olivia Rodrigo 42 times in a row” feature? I’d definitely be embarrassed by most of the results, but would merit a good laugh.
Sharing user data back to the user
As I mentioned, Spotify Wrapped is seriously like a holiday to me. I invest a lot of time in listening to music and podcasts in the Spotify app, so it does feel like a present to see how all my absurd listening patterns played out over the course of the year. I loved learning that I listened to almost 90,000 minutes of music last year and exclusively played Fred Again for basically the entire month of October. It’s fascinating seeing how your own data and app usage can be returned to you, but in a fun, easy to digest, and shareable manner. Shareable is key here. Generating Wrapped is not just a feat of data engineering, but an incredible marketing effort. It’s a mutually beneficial product — share user data back to the users, present it in a way that encourages sending it to your friends and posting it on social, and have free Spotify advertising across Twitter and Instagram for the days following the release.
For a note on user adoption of personalized products, I really liked the Spotify Engineering team’s breakdown of learnings in 2015 after the launch of Discover Weekly. They highlighted a few key points that drove success including:
“Reusing a format people already know, in the simplest way possible way, can help your idea reach a large audience quickly.”
“A personal image can help create a sense of ownership, and draw more people into the experience.”
“Involving marketing, design and other functions early on is crucial for success.”
You can read that post in full here.
Build your own Discover Weekly inspired playlist
We’ve talked about API’s before (if you aren’t sure what they are, check out this past edition of Day to Data that gives some background).
One of my favorite API’s to play with is Spotify’s. They have a massive amount of easy to access data and fun use cases, as well as incredible documentation (a developer’s dream!). For a while now, I’ve been working through rebuilding Spotify’s Discover Weekly function as a side project. You could spend years working on a perfect recommendation system, which is a concept I hope to cover in a post in the future.
I plan on creating a walk through of how to generate this playlist from start to finish in the future, but for now, I want to point out the publicly available tool you can use if you'd like to explore it.
Spotify’s API endpoint for Get Recommendations allows users to automatically generate recommendations based on their listening history. Spotify returns to a user a set of IDs for songs they'd recommend. I tried it out and it was a new set of songs that I hadn't heard before, but not quite as good as my Discover Weekly.
What’s coming next from Spotify
Spotify recently announced their new AI driven product DJ that is using machine learning and generative AI as a way to bring a personalized music experience to users. (Sounds a little bit like a radio station right? I could talk all day about how some things are just trending right back to where they began).
They’re combining generative technology from OpenAI, voice from Sonantic (acquired by Spotify) and their already powerful personalization engines to create an experience for users everywhere. I’m excited to give it a try and report back how the product performs.