When Spotify first debuted its Discover Weekly personalized playlist in July, it felt like a hit. On Twitter and in real-life conversations, even casual music listeners could routinely be heard singing its praises. Well, now there are some numbers to back up the hype: In its first five months, Discovery Weekly resulted in 1.7 billion streams.
In the quest to crack the code of online music discovery and curation, plenty of companies have thrown plenty of ideas up against the wall, always involving some blend of machine and human smarts. So what makes Discover Weekly different?
“We brought discovery to people in a very familiar format, which is the playlist,” says Matthew Ogle, Discover Weekly product owner at Spotify. “We took a lot of the work and cognitive load out of discovery. You don’t need to learn a new interface. It just works.”
And it’s not just the music geeks who are eagerly diving into their weekly music recommendations. When newer, more casual listeners try Discover Weekly, they listen about 80% as much as hardcore super-users.
For the unfamiliar, Discover Weekly is a playlist of songs that automatically appears each Monday in every Spotify user’s account. It analyzes that person’s listening history, focusing on the music he or she has played recently. It then compares that insight to the playlisting behavior of others. 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 playlist. It essentially takes the tried-and-true “people who like that, also like this” logic of collaborative filtering and applies it to the process of making a mixtape.
The end result, Ogle tells me, is a playlist of songs designed to feel like it was handcrafted for you by a good friend. Thanks to the magic and scalability of machine learning, Spotify is able to create these playlists for millions of users at once.
Much of the technology that powers Discover Weekly has lived within Spotify for quite some time. But until July, it mostly resided within the depths of the “Discover” tab of the service, where new albums and artists are laid out in a grid, sort of the digital equivalent of browsing the bins of a record store guided by a music-savvy best friend.
The recommendations found in the Discover tab are quite good, but they’re buried in the product, waiting to be found by the listeners who are most eager to hunt for new tunes. The Discover Weekly playlist removes all of that friction by showing up in a familiar format, right next to the music a user already listens to.
“For all the special sauce and the algorithmic work, the fact that we’ve kept it simple and that it’s just a playlist has really helped it resonate with people,” says Ogle.
The adoption of Discover Weekly has also benefited from some seemingly minor design details. The playlist’s thumbnail, for example, pulls in the user’s Facebook profile picture and automatically generates a more personalized graphic. It may sound inconsequential, but adding this detail resulted in a 17% increase in engagement.
Ogle knows a thing or two about how music discovery works. Before joining Spotify as a senior product owner in January, he founded a small but buzzworthy music discovery site called This Is My Jam. Before that, he worked stints at the Echo Nest and Last.fm, two venerable players in the digital music curation space.
As successful as Discover Weekly has been, the feature is far from perfect. Soon after it launched, users vocally clamored for some kind of Pandora-style voting mechanism for dismissing songs they don’t like (or endorsing the ones they love). This feature, which already exists within Spotify’s “Radio” feature (a copycat of Pandora), would be very easy to implement. But after putting together a few prototypes internally, Ogle says the team is still torn on whether or not to add UI elements to Discover Weekly, for fear of making it less of a lean-back experience.
In the meantime, the team behind Discover Weekly is looking at other data signals to incorporate into the feature. Almost half of Discover Weekly listeners wind up saving a song from the playlist to their personal collection, a “thumbs-up”-esque signal that the team plans to incorporate into the underlying algorithm. Similarly, they’re looking into whether to interpret song skips as a “thumbs-down.”
From a competitive standpoint, the timing of Discover Weekly’s successful launch could hardly have been better: Just a month earlier, the world’s biggest tech company, Apple, took direct aim at Spotify with the launch of the Apple Music subscription service, which features some extraordinarily solid music curation features of its own. Meanwhile, Pandora–whose pioneering model of human-informed, data-powered curation pretty much invented Internet radio as we know it–is gearing up to join Spotify in the on-demand space as well, as evidenced by its recent acquisition of Rdio’s assets and talent.