Spotify Just Made The “Shuffle” Button A Thousand Times Smarter

Matthew Ogle wants Spotify to stop feeling so much like a music app. That might sound counterintuitive, since Spotify is pretty much synonymous with streaming music these days, but his quest is simpler than it sounds: More and more, he wants the all-you-can-stream music service to feel like an old friend.

When Ogle, a senior product lead at Spotify, oversaw the development of Discover Weekly last year, it was this synthesized feeling of years-long camaraderie that inspired the feature and its mix tape-esque vibe. Since the unexpectedly explosive success of Discover Weekly, the company has shifted to new personalized music features. First, there was Release Radar, an auto-updated weekly playlist of brand-new music based on your listening habits. Today, Spotify listeners are getting something new: Daily Mix, a fresh stab at music personalization that Ogle and his team hope will keep listeners even more addicted.

Daily Mix is an infinite, always-updating playlist (or set of playlists—each user can have up to six of them, depending on how broad their tastes are) that consists of two things: Primarily, it’s a shuffling list of songs Spotify already knows you like (because you listen to them a lot), but with new, Discover Weekly-style recommendations woven throughout. Think of it like the “shuffle” button on artificially intelligent steroids.

“For every user, we create a picture of their taste, which just looks at all the artists they play and the similarity distance between those artists and try to find some natural groupings,” says Ogle. “The idea with Daily Mix is that it takes all the zones of your listening and tries to make a bottomless playlist out of each one.”

Using Spotify’s taste analysis data science—one of the key technologies developed by the Echo Nest before Spotify bought them in 2014—Daily Mix first lumps your frequently played music into stylistically cohesive clusters. These are not quite genres, Ogle is quick to point out, since such human-concocted categorizations of music can be vague, overly broad, or imprecise. In fact, internal user studies showed that naming these lists after genres (“My Daily Hip-Hop Mix,” for instance) altered people’s expectations and complicated the experience. Instead, successive lists are named as simply as possible: Your Daily Mix 1, Your Daily Mix 2, and so forth.

Throughout the playlists, Spotify introduces less familiar (or often, completely new) songs using logic similar to its Discover Weekly algorithm. These songs are, according to the data, likely to scratch a similar itch as the one that their more familiar-sounding counterparts already do. For example, if you’ve been binging on Paul McCartney’s early solo stuff, your Daily Mix might throw in a lesser-known Beach Boys song that has a similar feel. Or if you’ve been listening to Rihanna’s latest album and recently went back to Janet Jackson’s Control to relive the album’s 30th anniversary, don’t be surprised to see an unfamiliar track by Nicki Minaj or an early Whitney Houston song show up next, even if you haven’t listened to her music in years.

“Our hope is that Daily Mix can help you fill in the gaps and explore around the music you already love,” says Ogle. Most Daily Mix playlists should be about 75% music you like and 25% new discoveries, although the ratio will vary depending on your listening habits and how deeply you’ve already dug into a cluster of related genres.

For those who aren’t familiar, Discover Weekly is a personalized playlist that analyzes your listening history and compares that to the behavior of a subset of active playlist creators. Scanning millions of playlists, the system finds tracks that are commonly listed alongside music with which a user is already familiar, and then groups those tracks together into a new, personalized list of songs. It essentially takes the classic “people who like that, also like this” logic of collaborative filtering, tapping into Spotify’s elaborate web of genres, artists, and musical tastes and crunching that data to create a digital mixtape for each user.

The feature was an immediate hit: In less than a year, Discover Weekly reached 40 million listeners (more than all Apple Music and Tidal subscribers combined) and generated 5 billion streams on Spotify.

With Daily Mix, Spotify is hoping to keep racking up impressive numbers like these, this time by blending discovery with familiarity. The effort comes at a pivotal time: Apple Music has quickly ballooned into Spotify’s biggest competitor over the course of the year, thanks to a blend of exclusive releases from big-name artists (a game that Spotify refuses to play) and superior, human-driven music curation in the form of Beats 1 radio and hand-curated playlists. The human-focused curation strategy has done well for Apple Music, which recently hit 17 million paying subscribers (compared to Spotify’s 40 million), but Spotify’s experiments in discovery appear to just be getting started.