Tech Insider / Spotify
There's a playlist on Spotify I love called Discover Weekly. It's updated every Monday with a mix of songs, some I know and some I've never heard, crossing into almost every genre with no discernible pattern. Like magic, it just knows what I want to hear.
It's one of the reasons why I'm listening to Spotify more than ever. And I'm not alone.
The nine-year-old streaming service has doubled its paid subscribers to 20 million in the past year, and has an additional 55 million free users who sit through occasional ads and forgo the ability to play any song in an album or playlist without shuffle enabled.
What's most remarkable is that Spotify is seeing this kind of growth and buzz during what was supposed to be the summer of Apple Music.
While Apple claims its new service can sort through millions of streamable songs better than anyone else, thanks to the help of human tastemakers, Spotify has kept pace and more with a series of innovations ranging from Discover Weekly — launched in July — to its own human-powered playlists.
We visited Spotify's New York office to see how the Swedish company is trying to make the most personalized music listening experience ever.
How people really listen to music
Spotify's progress in sorting its library of 35 million songs can be traced back to The Echo Nest, a music intelligence company that was created within the MIT Media Lab a decade ago. Spotify bought The Echo Nest last March in what was reported to be a $100 million deal.
Jim Lucchese, CEO of The Echo Nest, tells Tech Insider that his team of about 70 people are focused on delivering “the right listening experience at the right time" within Spotify.
They do this by analyzing the makeup of every song, how people are talking about music online, and how people are listening to it. While the company continues to work with clients like Rdio, Microsoft, Sirius, and Vevo, as it did before it was sold, its most cutting-edge work is developed and honed for Spotify.
Ajay Kalia, who oversees the project, tells us they realized early on that there’s an important distinction between the music you listen to and music you actually like.
For example, just because I play a lot of instrumental, ambient music while I'm at work doesn’t mean that I have a particular affinity for those kinds of artists. And just because your significant other plays a lot of country music while you're both in the car doesn't mean you want a bunch of country playlists shoved at you.
"We believe that it’s important to recognize that a single music listener is usually many listeners, and a person’s preference will vary by the type of music, by their current activity, by the time of day, and so on,” says Kalia. "Our goal then is to come up with a nuanced understanding of each portion of your taste."
Kalia showed me my taste profile, and it was surprisingly accurate in terms of which artists I like. Different kinds of artists that span multiple genres but are still somewhat related are grouped into what’s internally referred to as “taste clusters.”
My largest cluster is 98 artists, which includes primarily hip hop artists like Kanye West, Drake, Miguel, and Jay Z. My listening is relatively well distributed within the cluster, meaning that I listen to each artist with roughly the same frequency.
If this cluster is what I actively seek out and listen to the most, whether it be in playlists or the search bar, then my second largest cluster could be considered what I listen to when I just want to have something on in the background, or what Kalia calls a "lean back" listening experience.
My second cluster has about 20 artists that are primarily instrumental, like This Will Destroy You, Explosions in the Sky, and Caspian. I usually work while listening to that kind of music, and The Echo Nest groups it separately and considers my activity within that cluster differently than the first cluster.
Its decision to group clusters separately is informed by how this kind of music is usually found in playlists like “Deep Focus,” explains Kalia. People primarily listen to ambient, instrumental music in Spotify without skipping a lot of tracks and for longer periods of time, which shows The Echo Nest that such music encourages a more passive listening experience than a playlist filled with the latest pop hits.
Building a better playlist
There’s another Spotify playlist I subscribe to called “RapCaviar,” which is described as "the freshest fifty songs on the rawest hip hop playlist ever."
This playlist, which is updated weekly, is the brainchild of Spotify’s resident hip hop expert, Tuma Basa, a veteran DJ with decades of experience at networks like BET and MTV. It's one of 4,500 playlists, with names like "Air, Sex & Water" and "Pop Punk Mania," produced by Spotify's team of 32 music experts around the world.
Although Spotify has had humans making playlists for years, its efforts got a major boost last year with the introduction of Truffle Pig, an internal tool from The Echo Nest that breaks music down into thousands of categories like “wonky,” “chillwave,” “stomp and holler,” or “downtempo."
Doug Ford, Spotify’s director of music programming, walked us through how a Spotify playlist comes together.
“Every curator has to come up with a hypothesis,” Ford says. "You brand it with an image and you give it a cool description that gives you an indication of what you’re diving into if you’re about to listen.”
For example, Ford says he used Truffle Pig extensively to make the playlist “Your Favorite Coffeehouse,” but that his most important criteria for picking a song was if it "gave me the chills."
Ford never could have combed through the tens of millions of available tracks on Spotify without help from Truffle Pig, but it was his “data-supported gut,” as he puts it, that made the final call.
Spotify's most popular playlist, "Today's Top Hits," is 100% human curated, says Ford, but the songs that make it into the list are surfaced with the help of Spotify’s complex algorithms in a matter of seconds.
“It really kills me when I see people say that Spotify is just algorithms,” Ford says. “Well, what does that make us? We’re running a team of curation experts.”
What you hear from everyone at Spotify is that humans using data insights are key to curating music on a large scale. Naturally, they're also using data to evaluate how well playlists are working.
"You can tell instantly if
it’s a win or not because we’re so fed by data," Ford says. "All
of the playlists live or die by the
Looking forward, Spotify is
thinking about a big expansion of the playlists it
Ford describes a future in which Spotify treats genres less like a collection of playlists and more like radio. He tells me to envision Tuma Basa’s RapCaviar expanded to include the culture of hip hop globally, with "top hits” playlists for reach major region of the world, a “gold” playlist with songs that are older but still relevant hits, a playlist for crossover pop and R&B tunes, and an “early bets” playlist that features what Basa thinks is going to be hot.
“I’m envisioning it like we own a big, large network,” he says, hinting at what’s to come.
Just what you want to hear
Spotify's work with The Echo Nest has culminated (so far) in this summer's launch of Discover Weekly, that custom playlist that magically predicts what I and every other Spotify listener wants to hear. Even though the feature is barely a month old, the overall reception in reviews and on social media has been overwhelmingly positive.
“Discovery can feel like work, and we wanted it to feel very human and natural like the selections that are powering it,” Matt Ogle, the guy in charge of Discover Weekly, tells TI.
Ogle previously worked at Last.fm, a startup that pioneered a lot of work on analyzing music data, and a now-shuttered startup called This Is My Jam, which let you choose one song a week to share with your friends. Since joining Spotify earlier this year, Ogle has tried to combine both approaches to blend big data with a human touch.
Ogle gives a simple analogy for how Discover Weekly works: You’ve been playing song A and song C a lot, but it turns out that when other people play those songs together in their playlists there’s a song B that you’ve never heard before.
Discover Weekly gives you song B.
“We look at what you’ve been listening to and what are the songs playing around these songs that you’ve been jamming on,” Ogle tells Tech Insider. “We’re trying to find the missing tracks.”
During internal testing, his team realized that if you don’t recognize a single artist in a playlist, you might question if it’s actually geared for you. That’s why the playlist is intended to have a mix of mostly new tracks with a few songs you’ve heard before.
“Having a little bit of familiarity is key to building trust. It can be exhausting to just listen to stuff you’ve never heard of before,” he says.
Now it's true that the songs on Discover Weekly are chosen
by algorithm, not humans, but Ogle insists that this misses
the point since the whole thing is built on data created by
humans — it's just that algorithms are connecting the
dots on a massive scale.
There’s something compelling about this humans versus robots narrative
“There’s something compelling about this humans versus robots narrative: a lovingly curated playlist versus an algorithm screwing up your sexy time,” says Ogle. "That whole distinction no longer really describes how we work. Discover Weekly is humans all the way down. Every single track that appears in Discover Weekly is because other humans being have said, ‘Hey this is a good song, and here’s why.’"
As popular as the new feature is, it's just the beginning for Spotify.
“We want to make sure that as mainstream or hipster as you are, Spotify can cater to you,” says Ogle. "That will involve humans and machines, and ideally it involves them skipping down the street hand-in-hand, because practically that’s the only way to get things done.”
“I see Discover Weekly as one of the first products from this new era of personalization, but ultimately we’d love for everything you interact with on Spotify to feel like there’s a bit of you in it.”
Music streaming is here to stay
Whatever differences there are between Spotify, Apple Music, and their competitors, it's clear that streaming music is the future.
A recent Nielsen study found that on-demand music streaming in the US increased by 92% during the first half of 2015 compared to the same period last year. And revenue from streaming music services grew 29% to $1.87 billion last year, accounting for 27% of the industry's total revenues, according to The Recording Industry Association of America, a music industry trade organization.
Revenue from sales of downloadable music, like you'd buy from iTunes, peaked in 2012 and has declined each the last two years, according to the RIAA.
The growth of streaming doesn't mean Spotify won't have roadblocks as it continues to build the on-demand listening experience it thinks people want. It has yet to roll out the video platform it debuted back in May, which will include a mix of podcasts and short video clips from partners like ESPN and Comedy Central.
Ultimately, Spotify doesn't seem too panicked about its place in the music streaming race or going head to head with most valuable company in the world.
“Hopefully, it will spur a lot of innovation on both sides,” says Discover Weekly’s Matt Ogle when asked about what Apple Music means for Spotify. "I think it’s pretty cool that the biggest company in the world agrees with us now that this is how people are going to consume music in the future.”