Inside Pandora’s New Social Music Mind-Meld

Pandora already acts like an all-knowing DJ. This week, it launched tools and tweaks to its Music Genome Project algorithm that could factor your friends’ song choices into a streaming social megamix. Founder Tim Westergren and CTO Tom Conrad explain.

Inside Pandora’s New Social Music Mind-Meld


Pandora got social this week. Now your friends are coming along on the Choose Your Own Adventure-style ride through an 800,000-plus song catalog.

The predictive streaming music service is rolling out a new look, new
guts, and, most importantly new social features that will slowly transform the personalized Internet radio experience into a tool for sharing tastes.

The magic of Pandora is that you tell it you like the Clash, and it
spits out five other artists you probably love but might not have
known or listened to in years. Behind it all is Pandora’s algorithm, or “Music Genome Project.” It’s powered by two sets of people: Pandora’s music experts on the front end listening to ripped music tracks and logging their attributes into the Genome; and users on the back end–now at 100 million–picking paths through those attributes by clicking thumbs-up and thumbs-down buttons for individual songs. The Genome has “10 billion pieces of thumb feedback behind it,” Pandora cofounder and chief strategy officer Tim Westergren says.

When it works best, it feels like hearing a favorite song on the radio or getting a rarefied mix tape from a record store clerk. With social, the feeling is multiplied by the number of friends whose musical adventures you “follow” in your “Music Feed.”

The Genome gets a tune-up this week, too, Pandora chief technology officer Tom Conrad tells Fast Company. It now has a third dimension–a “social algorithm”–that helps listeners filter out or include artists according to social-cultural characteristics, or “social connections from a cluster that give a sort of wisdom of the crowd,” Conrad says. So, for example, even though the bands Iron and Wine and Postal Service might never be matched for a listener based on Genome data and thumbs-up behavior, they may be linked via the new social filter. And even though Massive Attack and Jennifer Lopez may share some common characteristics through the Genome analysis, the “social algorithm” knows that the former’s fans would never want to be seen near J.Lo and will factor that into new social-based recommendations.

With friends helping make choices–and adding to Pandora’s data set–song predictions get exponentially smarter for every user. Not only are you telling it whether you like its choices but whether your friends would like those choices, too.


The “big a-ha! moment” for Conrad was realizing that listeners’ mentors may be very different from their circles of friends. And that’s an idea that works with Pandora’s core values. The new Pandora still “surfaces and supports personal musical discovery” by suggesting ‘fitting’ people or gurus to follow and letting users control interactions with mentors, Conrad says. (Like Twitter, there is even a private “DM” type function for mutual followers.)

In a conversation prior to the rollout of Pandora’s new social tools, Westergren was careful to underscore Pandora’s core mission to “empower personalized music discovery.” If social tools aid in that, great, “but our musicalogical approach is blind to musical popularity,” Westergren said. (As Westergren told TV’s Stephen Colbert, “What you like is cool.”)

The latest features could end up wowing both users and critics who were underwhelmed by Pandora’s recent $2.6 billion-valued IPO–as recently as Tuesday, at least one analyst was branding Pandora stock with a “sell” rating, helping drive a 3% stock price dip. Some have pointed to Apple’s impending cloud music services and today’s U.S. Spotify launch as stiff competition. But Pandora’s new social functions were an unexpected twist, a smarter alternative to Spotify’s playlists, which too can be shared on Facebook, Twitter, and other networks. Apple’s lame social attempt, Ping, doesn’t even register.

Plus, Westergren hinted last week that Pandora’s social offerings are still taking shape. “It’s something that we’re working on. Because we think there’s value there.” And Conrad foresees iterations of social Pandora giving listeners the ability to build stations based not just on what music they like but also on what friends/followees they like. “My social station” could be built on combined interests of a user, plus his friends, plus, say, Beck or Bono.

But it’s a thorny path through the discovery Pandora has worked hard to engineer. And the idea of adding friends to the equation gets into the philosophical differences between personalization and socialization–do I like my music more because my friends like it and I want that to be known to my community? Westergren was clear to point out that the power is still in the hands of the individual and that Pandora’s personalization champions one’s “guilt-free” listening and liberation from peer pressure. “Pandora’s no sort of cultural filter … not a tastemaker.”

Conrad, too, says that personalization is the foundation for the total product and user experience; new social tools exist only to “enhance the listener’s personalization results.” You won’t see a “trending songs” home page, and promoting songs or artist popularity goes against everything Pandora believes in, he says.


But what about co-listening features like those on “We’re extremely dedicated to a personalized experience, so if we were ever to implement co-listening features–which is fundamentally about groups coming together–we’d have to strike just the right balance with our core personalized experience,” Conrad says.

[Image: Flickr user _ambrown]

About the author

When Kevin isn't writing for Fast Company, The New York Times (2X Page One, Dalai Lama interview), Vanity Fair or The Economist, he's consulting ( turning client brands into fast businesses. Kevin's advised leadership at a top space exploration company, Google and Harvard University on branding and research.