Music Tech

Inside Pandora’s Plan To End Abandoned Playlists

image from www.openlettersmonthly.comSubtitled "how I learned to stop indecision and love the Genome," Pandora data analyst Glenn Peoples offers a peek inside “Add Similar Songs,” a unique and powerful playlist-building feature built into the new Pandora Premium service.

_____________________________________

Pandora_2016_logo wide

By Glenn Peoples, Music Insights and Analytics at Pandora

If famed disc jockey John Peel were alive today, would he build playlists on streaming services? Probably, yeah. Would he ever need help building a playlist? Who knows? He was a music encyclopedia and revered tastemaker. But nearly all music lovers are mere mortals compared to Peel. Making a good playlist isn’t easy. Now people can turn to “Add similar songs,” a playlist-building feature on the just-launched Pandora Premium service.

To construct a playlist, a person uses part music knowledge, part recall and part effort. If you can’t think of an appropriate song, if you can’t remember a particular song or artist name, and if you don’t have the time to invest, trying to make a good playlist can produce more frustration than reward. Playlist creators need a helping hand when they hit the streaming equivalent of writer’s block. But not just any song can be picked. Playlist creation is a personal experience, like John Cusack’s character carefully crafting his mixtapes in the movie High Fidelity. It has to be a song right for you. 

Pandora’s “Add similar songs,” a feature in Pandora Premium, adds five songs to a playlist. Click again and get five more songs. It’s not magic. “Add similar songs” uses a user’s listening history and the Music Genome Project to predict what songs are right for each playlist and each person.

Playlist indecision means people often start a playlist but soon leave it abandoned and unheard. The playlist doesn’t go away; it’s still accessible, like an ignored vinyl record at the bottom of the stack. “We know most playlists out there have fewer than five songs,” says vp product Chris Becherer, speaking of insight gained from his time at Rdio, the on-demand service whose assets were acquired by Pandora in 2015. Through market research, Pandora also knows playlists are one of the top factors when a person is considering a premium service. So, Pandora has built an innovative feature for its soon-to-launch Pandora Premium service. “Add similar songs” does what the name suggests: it adds to a playlist more songs chosen specifically for that listener.

After “Add similar songs” picks songs, a person can remove a song — just swipe to the left — and change the sequence of songs. For more variety, add a different song and run “Add similar songs” for more recommendations. “We do have high confidence these are the right songs, but, this is on-demand, not radio, and the user is in control. If they don’t want it, they can swipe it [to remove the song from the playlist], and keep going, or re-order, or add their own song for variety and hit the feature again,” says Becherer.

The user will get similar songs, yes, but not always the most obvious songs. “Add similar songs” taps into the Pandora’s Genome ability to help people discover new music. In one case, using the late Chuck Berry’s “No Particular Place to Go” as the seed, “Add similar songs” returned classic hits by obvious Berry compatriots like Chubby Checker, Little Richard and Jerry Lee Lewis. It also added “The Wanderer” by Dion, a contemporary of Berry’s who infused pop into rock and R&B, and the less-heralded classic “The Book of Love” by The Monotones. In a matter of seconds, Premium built a fine collection of songs from rock’s early years that spanned Berry’s rock and R&B styles with pop and doo-wop, too.

Yet, for all the outward simplicity, “Add similar songs” involves a great deal of science. Each group of five songs comes from real-time analysis of Pandora’s Music Genome Project—a taxonomy of songs’ musical attributes—users’ listening habits and their thumb histories. Call them on-the-fly recommendations. Like Pandora’s decade-old radio service, “Add similar songs” is personalized for each particular playlist and listener. Knowing what song to play next, a challenge as old as radio, is at the core of Pandora’s mission. The Music Genome Project has for more than a decade recommend songs and learned from listeners’ feedback, always iterating, always improving.

Market research has confirmed that playlists are central to the music streaming experience. In a survey of 1,500 people in the US, UK and France, MIDiA Research found 68 percent of streaming service subscribers listen mainly to playlists while 60 percent also listen to albums. The album used to be the standard collection songs. An artist would release ten or so songs, a number that has increased in the digital era, call it an album, and send it to retail. A person who wanted just one or two songs would buy the CD or LP. When download stores become popular in the early 2000s, consumers could purchase single tracks and craft playlists on software such as iTunes. Today, outside of the vinyl LP resurgence, there are no indications the playlist won’t be streaming’s standard unit for the foreseeable future.

The “Add similar songs” button tool is built for moments of indecision, for quickly creating a playlist, and for needing curatorial help. Pandora doesn’t want subscribers to face decision paralysis, a phenomenon described in acclaimed book The Paradox of Choice. In it Barry Schwartz described the difficulty in making a choice when presented with an abundance of choices. Having more choices gives a person more power and autonomy. There’s no need to buy that brand of detergent when others brands of equal satisfaction are available. But too much choice has its problems. A person’s decision-making can become burdensome and the level of satisfaction can fall. “At this point, choice no longer liberates, but debilitates,” he wrote.

In a well-worn example of the paradox of choice, a Columbia Business School professor ran an experiment in which shoppers were given either 6 or 26 varieties of jam. One might think the people presented with the most options would buy more jam. It turned out the shoppers presented with fewer choices bought 8 times similar jam than the shoppers with 26 varieties.

What’s the lesson to a music service? Reduce the number of options? Probably not. Premium’s catalog has a wealth of music spanning genres. Another solution would be to build a feature to solve the problem. Here the story of 401(k) savings plans provide a good example. A 2004 study by four Harvard economists details the indecision problem. When workers of three companies had to opt into 401(k) plans, enrollment ranged from 57 percent to 69 percent. But when workers were automatically enrolled their 401(k) participation rate was 85 percent. Automatic enrollment turned out to be a good tool to combat the indecision faced when choosing between a menu of options.

Obviously, music is neither jam nor a retirement plan. But with music and streaming services, listeners can face the same kind of indecision seen in other environments. When the mind hits a wall, automation can step in and finish the job.

Share on: