Several days ago, a friend admitted to me that he’s grown progressively disinterested in discovering new music as it’s become more accessible. The confession came as somewhat of a surprise from someone whom I respected for years as a true music nerd. We used to spend inordinate amounts of time downloading reams upon reams of .MP3s and compiling them into genre- and mood-specific playlists. We constantly listened, discussed and swapped new songs we were into. We’d hear a song we liked while out at the bar, and by the next day it’d be in heavy rotation in our personal playlists. Sometimes, we’d torrent entire discographies of artists we barely listened to. Why did we download songs so indiscriminately? So we could have them. Our appreciation for music was second only to our appetite for hoarding it — how could you consider yourself a true music fan if you didn’t have obscure Elliott Smith EPs handy?
But our commitment to finding interesting new music dissipated in recent years. A large part of this was due to new restraints on our free time — working a full-time job isn’t conducive to being a music snob. This is not uncommon — most people’s interest in what’s new and hip wanes as they age, their tastes calcifying. The larger problem, though, was that with streaming (namely Spotify), there was simply too much music to sift through. Part of the joy of pirating music was the effort and patience required — discovering an artist via word of mouth or a music forum, then hunting down the appropriate song files and waiting for our modem router to download it. I felt accomplished afterward. But with streaming, access was instant. Where do you begin when the entirety of recorded music history is a few mere keystrokes away? The technology was streamlined, but the process had turned daunting and arduous.
Spotify knew its selection was overwhelming for users looking to discover new music, which is why in July 2015 they introduced Discover Weekly. Every week, Discover Weekly delivers me, and millions of others, a 30-song playlist catered to our unique music palates. The songs tend to be a mix of things I’ve either never heard or have long since forgotten, from artists I either know and love, have been meaning to check out or have never heard of before.
Discover Weekly solves for what’s called “the paradox of choice.” Turns out, people don’t actually want total freedom to choose. When presented with a seemingly infinite number of options, we enter “decision paralysis” and become dissatisfied. Ever spend more time toggling through Amazon, HBO, Hulu, iTunes and Netflix than you do actually watching? That’s decision paralysis. What people really want is a limited number of options from which they can choose; otherwise you’ll constantly fear you’re missing out on something better.
Which is exactly what Spotify’s Matt Ogle theorized when Spotify hired him away from This is My Jam (a music discovery platform he founded) in early 2015 to revamp Discover, its struggling recommendation feature. While Ogle saw potential in the technology powering Discover — which recommended row upon row of artists and albums to users — he wasn’t surprised Discover’s abundance of recommendations were widely overlooked. “I came in with the idea that Maybe this was a bit too much work,” Ogle says.
Ogle’s solution was to make playlists — by using Discover’s pre-existing recommendation engine, Spotify could generate personalized “mixatpes” for each of its millions of users. “We wanted to make a great mixtape and we thought, Why don’t we look at the billions of mixtapes our users are making and power [Discover] Weekly by that?”
Discover Weekly creates playlists by analyzing a user’s listening behavior and comparing it to that of other like-minded users. Let’s say you’ve been listening to lots of Gary Clark, Jr. lately, for instance. Discover will find other Gary Clark, Jr. fans and identify the songs and artists they’ve recently added to their personal playlists (e.g. The Black Keys, “Them Shoes,” Heartless Bastards). Discover filters out the artists you’ve already heard, reducing the list to 30 songs (about two hours worth of music).
Perhaps the biggest key to Discover Weekly’s success has been this limited selection. “[30 songs] felt like a very digestible amount of music and that really made a difference,” Ogle says. “We also decided that it should feel special — kind of like a gift someone made for you.”
Discover is in stark contrast to Pandora’s exhaustive taxonomy process (known as the Music Genome Project): Each song is ascribed up to 450 distinct musical characteristics — such as “electric rock instrumentation,” “punk influences” and “minor key tonality” — and Pandora recommends songs that share characteristics. But Spotify’s relies on the hivemind of its users rather than a thorough dissection of each song’s elements.
Today, Discover Weekly delivers 75 million new mixtapes to users every Sunday night — 3 percent of all the streams on Spotify now come from Discover Weekly playlists, according to Spotify. And Discover Weekly has made Spotify more “sticky,” Ogle claims. Discover Weekly listeners are more loyal to Spotify, he says, and more willing to subscribe to its ad-free service. Many users have ditched their own, personally-curated playlists for the new feature. Half those who try their Discover Weekly playlist end up listening to at least 10 of its 30 songs, adding at least one to a personal playlist.
But can it truly expand your horizons? One of Discover’s current shortcomings is that it often serves up music in a small collection of closely-related genres, Ogle says, and doesn’t cater to “adventurous” listeners with more eclectic taste. “Some people complain that they listen to heavy metal 20 percent of the time but barely get any in their Discover Weekly playlists,” Ogle says. “So we might make a version that is a more even distribution.”
The more immediate goal for Spotify is to improve Fresh Finds, its collection of genre-specific new music playlists. Fresh Finds uses “natural language processing” to analyze prominent music websites and determine whether the sentiment around a certain song or artist is positive or negative. The feature will also examine the listening patterns of Spotify’s “hipster cohort,” according to Ogle — 50,000 Spotify users who have demonstrated a knack for seeking out up-and-coming music.
For those of us who used to demonstrate that knack, this is great news. Not only has my immense music collection been rendered obsolete by streaming, I’m more than a tad embarrassed about how hard I tried to cultivate an image as Guy Who Knows Music. I don’t want to read Pitchfork reviews and scour forums and download the songs all the other cool kids are listening to anymore; I want an algorithm to reduce my taste to a series of ones and zeroes and then tell me what I should listen to. Good thing we finally have one that seems to work.
John McDermott is a staff writer at MEL where he last wrote about how much people actually earn driving for Uber.