The Pandora Difference

If it was bliss to be alive back in the late 1990s heyday of world-changing Internet start-ups, then to be young and involved in a music/social networking project was like going to heaven. Napster and other file-sharing programs spawned millions of hard drive mp3 collections and such utopian slogans as “content wants to be free.” A decade later, in the harsh light of a financially desperate and hopelessly hungover new morning for the music industry, most of those ideas and projects look hopelessly naive. The sense that a revolutionary idea about music distribution could have anything but a destructive impact on the lives of musicians and artists has nearly vanished, and the visionaries of a decade ago are hunkered down protecting whatever cynical slice of the leftovers they have managed to hang onto.

Not Tim Westergren, the founder and chief evangelist of what may prove to be the most revolutionary of all free music services, Pandora Internet radio. For those few who remain unfamiliar with this remarkable product, a brief user-side description is in order. Go to Pandora.com, set up a free account, and then enter the name of a favorite artist or song. Immediately Pandora begins streaming music to your computer based on the musical identity of the new “radio station” you have just created. In amongst a handful of offerings by the artist you originally specified, you will hear a spectacularly broad range of other material that bears strong sonic, structural, and cultural resemblances to your station’s namesake artist. Your Lucinda Williams station gets you tracks by Neko Case, and your Ryan Adams station streams you songs by Jay Farrar and My Morning Jacket. If you don’t like what you hear, you can skip ahead, and if you really don’t like what you hear, you can click on a “thumbs down” button that will eliminate the artist playing at that moment from future editions of that radio station. Every so often, but quite infrequently, a short commercial will come on for 10 or 15 seconds, but then it’s back to the music, much of which is stuff you’ve never heard before, and most of which, if you have chosen your initial artist well, you like.

Like some of the other truly durable Internet products out there such as Google search, Pandora is the product of a sophisticated and labor-intensive process of describing and organizing music into a useful database. This data set, known as the Music Genome Project (MGP), lies at the core of Pandora’s capabilities. A trained musicologist listens to and analyzes every track uploaded to the system for approximately 400 different qualities, or song “genes” as they are referred to. Each song thus acquires a “vector” or numerical profile based on the way it sits with the hundreds of criteria brought to bear. Once a song has a vector, Pandora’s algorithms take over, weaving what feels like a seamless and intuitive listening experience out of millions of bits and bytes.

I spoke with Tim Westergren last week about Pandora, what it does now, and what we can expect from it in the future.

What is the Music Genome Project?

The Music Genome Project is an enormous musical taxonomy. We employ a team of musicologists and we work with 400 genes per song to describe the music in a very complete and accurate way.

Are all the genes made up of the same kind of data? Some seem drawn from internal evidence – how a song sounds, and what instruments are on it – while others, like “acclaimed work,” require context. Do you make that distinction?

Yes, there are definitely some genres that use more metadata than others – classical for instance. That’s where you are going to see genes like “acclaimed work,” because they are useful in that area. But in rock and pop, we rely almost entirely on internal evidence, on the sound and musical qualities of the song, rather than on its reputation or reception.

Some of the tags that you are using sound almost comical. Is “thin rap voice” really the musical equivalent of a gene?

The MGP tags that you see listed some places online, like Wikipedia, are what we call “roll ups” of genes, and not the genes themselves. That means they are built out of several genes so that non-musicologists will know what we’re talking about. The genes themselves might be more abstract, but we can roll them up until we are talking about a “thin rap voice,” which is something that anybody recognizes what it is. As you can see, language is a big part of what we do.

So is the Music Genome Project at the core of what Pandora is and does?

Yes, I’d say that the Music Genome Project provides the prime value of our product. But Pandora the service is a lot more than just the idea of a taxonomy of music. You also have to implement it, and you have to update it continually. The hardest thing at first was to train people to perform the analysis. We had to find a way to get a useful, consistent data set from a group of individual analysts, so there was a significant training period to get our people ready for this work.

For several years it looked as though negotiations over royalties would make Internet radio too costly to be effective. Where are we now in relation to that issue?

The big issues with song royalties have been resolved for about four months now. There’s a new law and everything. It took a lot of time and negotiation. Now there’s a fee structure that makes sense for us going forward. It’s not necessarily a fee structure that makes sense in general, because there are still some gross inequities. Satellite still pays less than Internet radio, and AM/FM broadcast, the most commercial of all the radio media, still pays nothing. But we do have a solution now, as of July 7, 2009. It’s still unfair, but it’s survivable.

Compare your free service to your Pandora One, your subscription service.

They are totally different in terms of the size of the audience. Pandora One, like all subscription music services, will only ever have a small audience relatively speaking. We view subscriptions as a capped market that will stop at a small number of millions of people.

How are things looking for the other players in the subscription music space like Real Networks?

Looking at Rhapsody I see a business that can be built so as to survive, but that will probably never get much past that breakeven point. It’s just too expensive to administer. The value is sucked out by the burden of processing all those per-play royalties.

What about Apple? Do you see them as competing with you?

If Apple made a move toward radio, obviously we would look very closely at that. But right now we are fine with iTunes. That’s a complementary business model to ours. We are not really competitors with Apple.

Is Internet radio really radio in the broadcast sense, or is it something else?

Broadcast radio and internet radio are two wholly different propositions. On the Internet, you can reach people one at a time and still make money. With broadcast, your audience must be large or you go out of business. That’s a big difference.

You are a musician, and obviously you love music. How has the work you do affected your own listening habits?

I don’t listen to music that much when I’m alone these days. I like to rest my ears for the work.

What’s the most important thing for people to understand about Pandora?

It’s all about the genome. Ultimately, the thing that makes the genome truly powerful as an idea and as a force for change in the music world is that it is not based on popularity. We are the only form of music discovery and distribution that is radically blind to popularity. When we put songs together it’s for musical rather than popularity reasons, and that really changes things.

Wow, I see. How did this core concept of being “popularity-blind” occur to you?

For me, it was a reflection of the experience I had as a musician trying to get my own music heard. If you can’t get on the radio – and what band starting out can? – then your only other choice used to be to get in a van and start touring. Not every musician can or wants to do that. Now, with Pandora, we have a huge pool of relatively unheard music; songs that are still really good, but that are just not that well-known yet. And because of our system, and the genomes, the 700,000 or so songs that we serve that would be obscure from any sales point of view make up 90 percent of our programming airtime. That ratio is an incredible, revolutionary thing for musicians, because suddenly they are out of the old radio play catch-22, where you have to already be popular to get played.

When you took this idea forward as a product, what were you thinking?

The intention was to create something – a tool for music discovery – and to have it really be pure, and not be biased by extra-musical trends. We hear all the time now from bands that are registering the impact of Pandora in various ways. They are getting iTunes sales without playing concerts because people are hearing them on Pandora. People who only know them from Pandora are booking them for shows. And this is just the beginning, because these effects will be heavily multiplied by scaling. When we have 10 or 20 times the listeners we have today, which will be soon, these effects will be much more pronounced.

A lot of the money in music these days is made on live events. Does Pandora plan to enter that arena? Will there be a Pandora music festival one day?

Probably not. When we do move into the live music space, it won’t necessarily be as festival promoters. You see we already have a really valuable role that we can assume built into our model, because not only do we know what people like, we also know where they are. We have their zip codes. When it comes time to rethink the way shows are booked, we will be there with the data to do it.

4•1•1

Tim Westergren is the keynote speaker at the first annual New Noise Santa Barbara conference beginning on Friday, October 9 at the Canary Hotel. For tickets, badges, and information, go to newnoisesb.com.