 I'm Doc Searles. I'm a fellow with the Berkman Center here at Harvard at the law school, still at the law school, though, of the whole university. How many people here are Pandora users? I'm assuming we've got 100%. Anybody not yet? You will be very soon. My connection here is that I'm, in addition to being a Pandora user like the rest of us, I'm an old radio guy. I worked in radio back in its golden days in the 60s and 70s. And working in radio, we always wanted to fix it. And anybody who worked, and especially in programming and radio, wanted to do it differently. Everybody wanted to be the program director. Everybody wanted to do their music their way and thought that would appeal to the largest number of people because that's what radio is about. And in many ways, that's what the music business was about. And that, of course, changed with the net. But I think it changed, especially with this one thing with Pandora. So Tim Westergren is with us here today. Tim is the founder of the Music Genome Project, which he'll explain, which was going for six years before you started Pandora, and before the world changed utterly, I think. And so I'd like to hear him explain that. And I'm sure the rest of us will too, so take it away, Tim. All right, so I'm going to guess that the interests in this room are pretty broad in Pandora, and so the things related to it. So I think maybe what I'll do is kind of tell the company story to y'all. And as I'm kind of meandering through that, feel free at any point to interrupt me and have a question or something you want to sort of dive into more detail on or challenge, or if you just have some declaration you want to make, totally fine. And I'll happily pursue anything that's of particular interest to you. So don't be shy about interrupting me. I'm happy to have sort of an exchange. I want to thank John for inviting me here. I guess you go by Jay-Z here. Is that J-A-Y-Z? That's not your role. That's my role. All right, pretty hip. So just a quick background on myself. I'm a musician. I graduated from Stanford in 88. And I spent about 10 years after college trying to make a living as a musician. Actually, my first job out of college was a nanny. I was a nanny for five years. It's a great gig. Actually, manny is the proper term for it. But it was not what my parents expected of me when I got out of college, but it was a great job. I worked in the afternoons, played stick ball with kids in the street, and cooked dinner for the family every night. And the rest of the time I spent playing the piano and learning about digital technology and eventually started playing in bands. And I spent about 10 years then kind of meandering around the music world. I played in rock bands. I have my own recording studio. I did some film composing for a while too. And the idea for the music genome project, which eventually became Pandora, really grew out of all those collective experiences. And the first one was becoming intimately acquainted with the challenges of being an independent musician and how hard it is to make a living at that. And the principal barrier is really about promotion. If you're a working musician, it's very, very hard to get noticed. And most musicians did what I did, which is jump in a van and drive thousands and thousands of miles to play for very small audiences. And you feel good in your own mind, but it generally burns you out, which it did to me. So I've long had a passion for the sort of plight of the working artist. And as a film composer, my job was really to figure out the musical taste of a film director. That's how you do well in the music in the film industry. And to sort of figure that out, you develop the equivalent of a musical Myers-Briggs test, where you interview composers, rather film directors, and you play music for them and get their reactions to it. And you deduce from that interaction what you think they like about music. You essentially translate that into musicological terms. So when you go back to your recording studio, you kind of have a roadmap of what you want to write. And so I got sort of in the habit of thinking about music and music taste and musicological terms and really in a very granular way. And film composing is a really deliberate profession. I mean, you are composing music with a very specific intention in mind and a very specific audience to please. So it's a very kind of calculating kind of composition. And so that was sort of in my head. And these sort of ideas were swirling around. I was living in the Bay Area in 1999, which was a pretty heady time for entrepreneurship. If you weren't starting a company, you were in the minority in the Bay Area at that time. And online music was just blowing up. CDNOW had launched e-music, iYuma, all these websites. And they had sort of held all this promise for sort of online musicians. And I got kind of caught up in that. And I had this idea that, well, if I could take this sort of musicology profiling that I had been doing as a composer and codified and make it available over the web, it might be a really powerful recommendation tool, like connect music and help people discover all these indie bands because of the similarity to well-known bands. And so I shared the idea with a friend of mine, a college classmate, who had already started and sold a company by then. And he said, yeah, that sounds like a cool idea. Let's do it. And so we started a company, which is how you did it back then. And a couple months after that, raised some financing, about a million and a half bucks of seed money from, among other people, Guy Kawasaki, who was kind of one of these Silicon Valley luminaries back then. Little did we know that we raised that money two weeks before something hit the fan, and everything really collapsed. I mean, it wound up being essentially the worst time in the modern industrial era to launch a technology company. We hit the dot-com collapse, the Wall Street disaster, and the music industry disaster as well that sort of followed on its heels. Sort of a quick trajectory. We raised that money, built a prototype for the genome, and I'll actually describe to you how that works. And about a year after that, ran out of money and sort of slowly, through 2001, to make that money last, we paid our employees less and less, a small and small percentage of their salary until by the end of 2001, we weren't paying anybody. We called that salary deferral at a time, which is actually illegal in California. It's probably some lawyers in the room know. We didn't know because we couldn't afford an attorney, so we were perfectly happy. But went into this really insane three-year period without getting salary. About 45 people worked building this music genome project sort of full or part-time without getting paid. It was pretty insane. But we managed to survive or just keep going, really. We licensed the technology a little bit, not for very much money, so we really weren't kind of building any kind of business traction, but just kept at it until 2004, when, after literally hundreds, 350 venture pitches, actually it was number 348, I wouldn't count it for the hell of it, we raised a big amount of financing in 2004, about $9 million. And that really was the moment when the company sort of was revisioned. We changed our company in both in product and in name to Pandora in 2004 and kind of recognized that they're in those intervening years with the sort of rapid adoption of broadband that an opportunity had popped up in streaming audio because high-speed access to the web is really a great avenue for listening to music. And so we repurposed this music genome project and built a new layer of software on it and launched the consumer-facing site, which we did in November of 2005. So we're almost, we're exactly three years old now as a company. And it just took off like a rocket ship. We have about 18 and a half million registered listeners now. And we add about 35,000 new a day. And we never advertised Pandora. So it's entirely a word-of-mouth phenomenon, which has been really, for me, it's like a Cinderella experience, given where we were. I mean, at the depths, I had 11 maxed out credit cards. I had a personal obligation on a five-year commercial office lease in Oakland, which is like $5,000 a month. And I was a musician. So I started with no money. And here I was looking over a financial precipice. It was like, if this didn't work out, I think I was probably headed to Mexico, or spending the rest of my adult life in indentured servitude somewhere. But we managed to make it through. And then subsequently raised a bunch of money. We've raised a handful of pretty large rounds of financing and sort of built the radio service. I had one of my most out-of-body experiences was what was being in Las Vegas one time. It was like 2003. And I was in a penthouse suite in one of Las Vegas's fanciest hotels, which by itself is an experience. And in one hand, I had a glass of really, really expensive red wine. And in my other hand, I had a little truffle infused Kobe beef burger, which is like this big. It cost like $80. And it's really good. And it was an event that was being hosted by one of Wall Street's investment banks. It kind of sounds different these days, saying Wall Street's investment banks. Back when they were barbarians at the gate. And we were being wind and dined by this bank that really wanted to be the agent for our funding round that we were going through. And the CEO, Joe Kenney, and myself had been through this kind of three-week cycle of being taken out to very fancy places by these bankers. And I remember in the middle of the evening, I was like 10, 11 o'clock at night, just looking around going, what the am I doing in here? Like really sort of having a moment of thinking like at 12 o'clock midnight, the clock would strike and something weird would happen. This would become a Big Mac or something. But needless to say, went through this kind of the whole cycle and wound up, anyways, raising a bunch of money. And the site took off. And our objectives now as a company are, I think, pretty grandiose. We want to completely fix radio and do it not just in the US but do it globally, which means not only fixing it for listeners, but fixing it for musicians as well. In our collection, we have the music of 60,000 artists. And 70% of them are not signed to a major label. So the bulk of the music you hear on Pandora is stuff you've probably never heard before. And music that's not getting played has never been played on any form of radio. And the dream for me is someday we'll have not 18 million but a billion listeners around the world. And the day a musician's music gets added to Pandora is the day they become professional. Because that song, you know, I get goosebumps whenever I say this stuff. It's funny, but I really feel it's pretty strongly that we will be large enough that when that song drops in, it will get blasted out to hundreds of thousands of people right away who have explicitly expressed an interest in that kind of music. And that connection will essentially a surface a bunch of patrons for that artist who can then support them in whatever way people are supporting musicians over time. So we're 120 people now. We're headquartered in Oakland. We have offices in LA, Chicago, New York, and London. I just like to say that. And somebody working for us in Tokyo, too. And we'll do about a little over $20 million of revenue this year and are kind of well on our way to building a healthy sort of media advertising supported business. And I'm having a whale of a time now where I wasn't a few years ago. Yeah. Does your business depend on in any way on suing end users for copyright claims? No, we're sued more than doing suing. Pandora is a completely legal service. We operate under a federal statute. And I'll talk in some detail about that. We pay performance royalties and publishing royalties for every song that we stream. So as a user of Pandora, you are operating on the legal side of the line with music. We've never had to sue anybody. We have to, there are patent trolls that we get hit with. So we have our own legal travails, but not by suing end users. So the genome, this is a picture of what it looks like at our offices. We have about 50 musicians that work for us. And their job every day is to sit down with headphones on and analyze songs, musically. One song at a time along close to 400 musical attributes per song. So they actually manually measure and score every granular dimension of melody and harmony and rhythm and form and instrumentation and vocal performance, you name it, piece by piece until they've built up essentially the equivalent of musical DNA for a song. These are all trained musicians. They typically have a four-year degree in music, essentially like what I was when I was in my 20s. They play in bands. This is their day job. Yeah, John? Do you send the same songs to multiple people to see if they rate you the same way? We do 10% of the songs twice as our way of QA. And they go through extensive training on this methodology. It's all done on location. And the consistency of the rating is actually pretty remarkable now. It's very solid. I thought it was going to be how many seconds of a song, then you know this is good. I should probably repeat that question for people are webcasting, huh? So the question was how many passes through do you need to, does an analyst need to listen to a song? And it varies. Three-ish, typically. It depends on the complexity of the song. It's a really simple three-minute pop ditty. Probably a couple times is enough. Symphanies, a symphony can take two hours to analyze. Just one symphony. It's a completely absurd methodology. And ultimately unscalable. But for reasons I'll explain, I think that's perfectly fine. The question is do we have templates that we can kind of map on to different genres? And no, you start from scratch. So it's the most sort of methodical, deliberate way of doing it. And this is actually what they look at every day. This is one of seven pages of musical detail that the analysts have to look at for every song. So you can see one of seven pages. So this left-hand column here, those details are all designed to understand just the melody. So the idea here is that whatever melody you have, whether it's some kind of drone, very simplistic, repetitive drone, excuse me, or it's some kind of acrobatic R&B style melody, you can capture it through some combination of what are really like musical primary colors. And it's just different scores. And this creates sort of a connective tissue between songs. So when an analyst is done with their analysis, they essentially have 400 numbers that represent that song. It's kind of like the musical double helix. And when you type a song in a Pandora, what happens is a math equation, an algorithm takes over and computes the musical distance between that song and every other song in our collection. So it's actually just a mathematical calculation. And then we build playlists from that. Now, play listing is, as Doc would tell you, is not like, let's play a bunch of same songs in a row. There's actually a bunch of art to how you sequence songs. And we have a whole bunch of intellectual sort of work done on how to kind of optimize the sequencing. But fundamentally, what's connecting songs is their musical similarity. So it's a very intense mathematical equation that has to do a lot of computing in a hurry to build playlists. And that's what generates the station. Yeah? So the playlists are all built programmatically, not by... All right, the stations are built programmatically, right? Now, the one additional element, though, is the thumb feedback we get from listeners. So as you probably all know, when you listen to a station on Pandora, you can thumb up or down songs. When you do that, what you're actually doing is you're monkeying with the algorithm. And as soon as you thumb down a song, for example, Pandora will immediately recalculate the playlists. So the song that was about to come next gets shunted out. And a new sequence gets built. And what it's trying to do, really, is sort of replicate your best friend, say, who knows your music really well and has this encyclopedic knowledge of artists. And as they, he or she, gets to know you better, they get better at recommending stuff to you. Because they learn, you know what? She doesn't like chromatic harmony. Or that person's a sucker for a good antiphany in the vocal harmony. Or someone likes alto sax solos. Or someone hates profanity in the lyrics, those kind of things. And the idea behind this is to sort of be that machinery. Yeah? I'm curious, are you mostly dealing with music that's new or relatively new? Or are you trying to also improve that catalog? So the question is, are we dealing with mostly new music? And no, quite to the contrary. I would say on a daily basis, about half the stuff we do, maybe more now, makes 70% is back catalog of some kind. And we've done a century of music, basically, back to early jazz and big band stuff. We have some stuff that's been converted to digital that was not available on CDs. And keeping up with new releases is actually not that big a task in the scheme of things. So it's a lot of old stuff. We want to sort of be something of an encyclopedia, you know? OK, I'll go for it, thanks. I love your line about wanting to completely fix radio because Lord knows it's in desperate need of fixing. And when you listen to music programming on commercial radio stations, I have a tough time even using that term, music programming, because the way, and I've worked in that industry, the way that playlists are put together is a complete joke. And you're not in your head because you know how it works. I'm just wondering, there's two things. Soccer has been pitched to people in commercial radio, not like yours, but we'll say that it has some similar characteristics, and they pitched it to predict the hit-making likelihood. And you're familiar with that stuff, so I wanted to ask you about that. But I also wanted to know about what the radio industry thinks about what you're doing, because I think they would be hugely threatened by this. Yeah, so the radio conglomerates are not monolithic. So the question was, what does radio think of us? And I guess, what do we think of programming or of traditional programming? The software that would analyze music for different answers for hits, say, oh, it sounds like a Janet Jackson song, therefore it's going to be a hit. OK, so just in terms of how radio views pandora, I think that, for one, they're not monolithic. Different radio conglomerates are reacting differently to the internet radio. Some are more aggressive about it than others. I think, though, since we launched on the iPhone about four months ago, I think that's drawn a lot of attention from traditional radio because you can plug that into your car. And internet radio has been almost exclusively an at-work phenomenon, which is only about 10% of overall radio listenership. But now we're popping up in areas that have been historically the sole domain of traditional radio. And that's being reflected in how people are writing about it and how people are using it. So I think you're going to see an acceleration of moves by traditional radio to counteract that. In terms of hit song prediction, I don't believe in it. We get asked frequently by labels to help them pick singles. And we don't do it because we don't think we can. And that's where the company that's studying songs pretty systematically and in great detail. I just don't think it's, there are people who spend their whole lives doing nothing but trying to predict hits and get it wrong 99 times out of 100. So I've just been very dubious about that. And if I could build something that could predict or create hits, I'd probably trash it when I'd be responsible for unleashing that on the world. So yeah. So the question is, do listeners' taste develop by listening to Pandora? We don't have any evidence really to back that up. I do think, though, on Pandora, people experiment. And I think it's because Pandora makes things accessible that weren't before. And classical or jazz would be two prime examples where most people don't really know much about it. They're kind of some ways intimidated by the genre. And I think in the privacy of the little Pandora experience, they can try stuff out. And I think the funny thing about music is that it's a profoundly social thing, but it's also something we're very self-conscious about the same time. I had a really funny exchange with someone a little while ago who wrote in an email. And he said, Tim, I'm really pissed at you, Tim. You played a Celine Dion song on my Sarah McLaughlin station. And so I wrote back. And I said, well, OK, did it not belong musically? And he wrote back. And it was a he. He said, no, no, it fit. But it was Celine Dion. And so I wrote back. And I said, well, OK, maybe it was the playlist that was the set list, rather. We play songs in three or four song sets. Did it not kind of belong in that set? And he said, no, it belonged in the set, but it was Celine Dion. And I tried one more thing. I said, OK, maybe it was transitions, like the song before and after. Was it kind of a weird transition? He said, no, they were pretty good, but it was Celine Dion. So I went back and forth like this. And then finally he wrote me a note saying, oh my god, I like Celine Dion. And the song, and Pandora doesn't understand the joke. Because Pandora thinks they sound the same. Like, what's the problem? And I think there's all these funny prejudices and stereotypes. And it was like this confession that he made, that he liked Celine Dion. That girl can sing. I mean, I have bought songs by her on iTunes. And I can say it with pride. My Heart Will Go On is a great song. And I still listen to it frequently. But I think people have a funny disposition towards music. And I think Pandora kind of breaks that up a little bit, because it's so objective. Yeah? Vision of what you, one, change the product vision and two, whether you can really change it in a detrimental way, or whether at the end of the day you jump through the booths and it all turns out OK. OK, so the question was really kind of how do we interact with the licensing constraints and how does this sort of affect our product vision and how we adapt it to it? So just kind of a quick summary of that. We operate under a federal statute called the Digital Millennium Copyright Act. I know I'm going to misuse some legal terminology here, so be kind. It's a compulsory license, meaning if you're a musician and you have released your music for public consumption in some fashion, Pandora has the right to play it. You don't have to ask your permission. We have the right by law. And I actually think you can't prevent us. Is that right, John? Like, you can't stop us from playing it? It's compulsory license. It's compulsory. Even if you want us to take it down, we don't have to. The beauty of that is that we can play a vast amount of music without spending the administrative time it would take to get direct deals, which would be the alternative. It allows us to include so much music. So all these bands that would never make it up above the radar, we can just grab them and play them. As part of that federal license, there are a bunch of constraints that we have to abide by. So in order to qualify for it, we have to be radio-like. So there is a certain level of interactivity that we cannot surpass without violating that. Things like you can't rewind a song or you can't hear a song on demand. You can never type in a song to Pandora and hear it right away. That would be violating the DMCA. We can only play a maximum of four songs by a single artist in a three-hour period. And I think it's no more than two songs off one album consecutively. So a bunch of things that are all designed to make sure that it's really like a radio and not like listening to a sort of a customized stream or your own CD collection of some kind. Limit of, I'm sorry? Yeah, we can't pre-announce a song. That's why the song starts before the song tile slides into view. You can only skip six songs an hour, which actually isn't codified in the DMCA, but it's become to be seen as the sort of standard for it. Now, interestingly I think in regards to your question, Pandora fits just fine in that world because we don't make any of those promises to you. You don't come to Pandora, so you can hear Dark Side of the Moon from front to back. We don't offer that. What our offer is that we'll play stuff you like. You might not know what's coming next. It'll be serendipitous, but it'll be stuff you like in the style that you tell us you wanna hear it in. And because that's the promise that we make, we're not super bothered by these various constraints. So if I can focus the question a little. Was that where you, and you said, and then realized, well, this is still an acceptable product. It's pretty close to where we started. You know, even, maybe some more skips would be an improvement to the experience, but I can't think of a lot of things where like, oh my gosh, I wish we could do that, but we can't, because of the DMCA. So it hasn't been super confining for us. Thank God, you know, that would have been very difficult. Yeah. So the DMCA is the US. How do you deal with broadcasts in the worldwide through the Internet-enabled? Yes, it does. So how do we deal with global rights? And the reality is Pandora is not legal outside the US. When we first launched Pandora, so you have to be sitting in the US to use it. And when we launched Pandora, the only, on the website, we just essentially said, you have to validate that you are a US resident by typing in your zip code. And that is enough for you to sort of honor code that you are a US listener. Our most popular zip code for the first year and a half was 90210. That's a little embarrassing, but if you took, like, a demographic extrapolation from Pandora, like Beverly Hills was Tokyo, you know, based on a percent of listeners coming from Southern California. But we had to turn listeners off about a year and a half into the service because we were threatened with lawsuits because there is no centralized form of licensing like the DMCA in any other country. There are some murmurings of it, and the UK has similar licensing bodies, but they've come up with answers on the fee structure that are completely unaffordable. So there does not exist right now a sort of statutory rate that's affordable that we can take advantage of outside the US. Our hope is that once we resolve the rates in the US that we can take that as a template and spread it globally, go map it onto other territories. And we've been through a very interesting process, maybe interesting kind word in the last year and a half because a year and a half ago in March, a arbitration panel from the copyright office ruled to triple the rates that we pay, the performance fees that we pay under the DMCA. So it overnight essentially broke our business model. And Pandora, we essentially had a board meeting not long thereafter where the topic was how soon do we shut down? Because we weren't profitable yet and it takes a while to build up your advertising base and we'd just been hit with rates that we couldn't afford. And rather than the shutdown, we organized a grassroots sort of political campaign. Did anybody get an email from us in here that, all right, I hope you called? We encourage listeners to make some noise. And Capitol Hill received 400,000 faxes in three days from largely Pandora listeners. And eventually got 2 million faxes, phone calls, emails, letters from largely Pandora listeners over the course of about a six month period. And it was so much public pressure that Congress intervened and forced a negotiation, a renegotiation, which we are sort of as we speak concluding. And so to my mind, I think I consider it to be an incredibly sort of watershed moment, not just for music but for sort of media in general. The idea that there could be some sort of change in an arcane publishing or performance fee in Washington that's wrapped by a very sophisticated lobbying machinery, which is the IAAA, and that the people would rise up and say no, that's in a novel maybe, but not in the real world. But thanks to the web, which has this incredible communication medium, literally it was public voice that caused this to get restructured. If it hadn't been for that, our company would have been out of business a year and a half ago. So we have listeners to thank for that and we may need help again. But I think it's a really, it was a vital victory because among other things, it maintains the statutory system. Because if you lose that, then folks like Pandora have to direct license their music. Which means we have to get individual permission from all the rights holders for every song that we play. In other words, we're at the mercy of the Four Majors, which is where a lot of companies find themselves right now in the digital space. I think you're gonna see in the next 12, 18 months, a lot of companies who have struck direct deals are gonna start learning why that's not the place you wanna be. And it also is incredibly important for all the indie artists because when you're in the direct deal world, you don't have time for the garage band from Indiana and you have to focus on where the bulk of the material is because that's where you can allocate your legal fees. If you're a lawyer or an intellectual property, you've got a bright future ahead of you. Because there's a lot of activity gonna be going on in that space. Coming up, yeah. Do you think what you've built can transcend culture and language? How far can you stretch your arms in the direction of the global? We're doing Asian Indian music as we speak. So, yep. And we just added Cajun, we're about to add Cajun, Celtic and Hawaiian. We have Spanish and Portuguese language music, so all the Latin genres. And it works like a charm. So our long-term goal is for you to be able to type in the counting crows and find their cousins in Turkey. And to me, that's when this thing gets to be incredibly exciting. And when we turn people off internationally, when we shut down Pandora, I remember an email I got from somebody, it was in Yugoslavia somewhere. And he was really crushed. He wrote and he said, you know, I'm writing in not for myself, and not just for myself, but on behalf of the town that I live in. You see, he lived in a town of about 400 people. And for, I guess, quite a while, that town, every Saturday night, they had to collect at the local club for Pandora night. And it had become like the social event of the week. And this sort of mountainous region and everybody would get together and Pandora would DJ the party. So people would come and do the mix and Pandora would provide the music. And it really brought home to me how universal music is. And that was when Pandora was 95% English speaking. So we hadn't even begun to do stuff outside US and British catalog. And still we were so popular in these countries. I had an email from a guy from Germany who said, we hate your government of war, but we love Pandora. And not to inject politics into this, but again, like, that's another way of saying like, when it comes to music, we are one. And I think it's such an incredibly powerful ingredient and this service will not really realize its potential until it is truly international. And I think it's gonna be important, I think sociologically. So I can't wait for that day. Yeah. Two quick questions. First one is, can you envisage a quote when a threshold is crossed and you have no need for those 50 musicians to rate music by your database and your system is so good that you can just talk. And secondly, you are just focusing on music, but obviously the radio itself has a much broader power than your speech. And can you see the same kind of system or your system being applied to both of the states? So the first question was, do we see a time when we won't need music analysts? And not in the near future. One of the things that is a unique characteristic of the genome approach is it's blind to popularity. So if you are a brand new band that no one's ever heard of, we know what to do with you, which is not true of any system that's based on statistics, right? If you're an author and you wanna get your book recommended on Amazon, you're kinda out of luck unless you have some pre-existing pipeline to get your book above a certain threshold that it will wind up on the people who bought this also bought recommendation lists. The genome doesn't have that handicap. And so we always need that genomic approach to surface all this new music. Computers can't do it statistically and the machine listening algorithms that are out there so attempts to actually recreate the human ear are just primitive compared to what a human ear can actually do. So I think it's unlikely in the near-term horizon for us to do that. I also like kind of providing jobs for 15 musicians, a little employment program. And then your second question was, again, oh yeah, non-music, yeah. And non-music material, news, weather, sports, et cetera, will certainly find their way into Pandora eventually. I don't think we do that in a genome for that stuff, but sort of an obvious candidate for local localized information, which eventually will be a big part of internet radio. Yeah. Can you just talk about advertising? Sure. Especially at work, when you look at the screen and if you're going through your phone, I'm sure you'll never look at the screen. Yeah, that's good to head here. Okay, so the question is how do we monetize? So this is a typical page of Pandora as a tuner and this canvas that wraps around the tuner is what we call a skin. It's a paid advertisement. And we have a team of about 35 ad sales people at Pandora whose job is to sell that inventory. And the question you're alluding to is a good one, which is that historically for internet radio, the way people use it is they launch the radio and then they minimize it while they do something else. And they may not go back to it until they want to turn it off. And so they don't ever look at it. So you lose this visual opportunity. And actually when we launched Pandora, we thought it was going to be a subscription business. We launched it was $3 a month. And the deal was you could listen to it for 10 hours for free and then you had to give us your credit card. We learned in a hurry that that was going no place. People, you would use it for 10 hours and then figure out a way to become a new user again. Just colossal, systemic dishonesty. And so we went free. Not knowing if we will solve this problem. And what we learned is that Pandora listeners on average go back to the site between six and seven times an hour. And that's been very steady now for a while to do something, to skip a song, thumb a song, or find out more about the artists who's playing that they've never heard before. So there are these reasons on Pandora to go back that endure over time. And I think not the least of which is the fact that when you do thumbs, it makes your listening experience better. So there's a reason to go back to it. And that's the cornerstone of the business. If we didn't have that, Pandora would have audio advertising or it would have, it would be a subscription business. But we get a lot of inventory because people are always clicking to do on the site. And in the long run, we believe that for the web, you need to be a happy home for brand advertisers, which is what you're seeing here, to be successful in the long run. And I think this is a good alternative. We've actually thought of a version of Pandora, building a version that looks like a spreadsheet so you can have it up at work and look like you're working. So we could show you advertising. But it appears we have no need for that. Two quick questions. One on the at work and look like you're working in front. You have that digital sweatshop there, I mean that area where the artists are listening and having a great time. But it seems like a perfect telecommuting job because you need the metrics to know if they were placing the ballots. Yeah, so senior analysts can do a little bit of remote stuff. And when a senior analyst, when they're on tour, they can do some of that remote analysis. But it's important for them to be in the office. They consult with each other. There's a lot of kind of informal QA that happens in the office. And the other thing is, are you ready to do this with movies and TV? Like, this is a sci-fi reimagining between an Australian actor with religious overtones and political connotations. I think that will be done. It won't be us who will do it. But I think you'll see this approach applied to a lot of different areas. We were asked if we could do it for husbands once by a woman who was considering investing. I guess E-Harpony already had that music. Exactly, exactly. And you know that it's all solving the same problem, which is what do you do with the unknown product, which is most of what's on the web? And that, I think, lends itself to that kind of approach. Two things. One is I just want to make sure you do get through your whole talk without being erupted. Let's just perfectly find it. I'll be right here if I'm going to be sure. The other is lyrics. How do you pay attention to lyrics? One observation a lot of us have made is that finding lyrics on the web means you go to a very scammed site and they get along. And I'm wondering if publishing lyrics has been part of your charter, possibly? So the question is, what are our thoughts sort of on lyrics? What role will they play in Pandora? And we've been working on incorporating lyrics for a while. It's a tricky business because of rights and because of catalog and who has the inventory we need and so on. But it certainly makes sense. You can imagine how complimentary it would be to be able to read the lyrics while you listen. And it'll come. And we do do some lyrical analysis. So we actually do. The musicians do analyze lyrics. It'll make its way on there eventually, for sure. I'm interested in the user experience. And I think sometimes when you play music by choice, you're sort of trying to fit a mood, a breakup song, or rainy day, or so and so. Is there sort of a mechanism for doing that? Are you thinking about that? And also, how do you find the independent fans that just get up on MySpace or whatever? So the first question is about user experience. How do we accommodate for moods? Our mood is your song selection, fundamentally. And the thing about moods is that my happy song may be your sad song. It's like Fire and Rain, James Taylor. That's a happy song for me. It's a song about somebody who dies in a plane crash. Yeah, I like to clean the on, so. I have very little credibility. But so we really base mood on song selection. And in fact, when we first built the genome, we had mood genes. But the problem is we could never analyze them consistently. And that was one of the criteria for being retained as a gene was, could we all agree on what it meant? And moods are notoriously difficult in that regard. And then your second question was? How do you find the independent bands? Independent bands. A band that has their own website or MySpace or whatever. Well, does anybody here have a CD to leave for me today? No? Well, that would have been one way. We take independent submissions. They get sent in all the time. We get boxed fulls of them. And there's no prerequisite for getting in. It just has to be good quality. So you don't have to be an active band. You can be a hobbyist. We get CDs that come in that are freshly burned CDRs with the track titles written in Sharpie on them. And if it's great, we'll find a way to get that in. Listeners recommend a ton of music to us. We get about 10,000 emails a month from listeners. And a lot of them are about, I just saw this great band yesterday. They should be in the genome. Well, it's a physical product. You're not scraping MySpace and things like that? No, no, no. Yeah, I mean, MySpace has, what, seven, eight million bands? And the vast majority of them are not ready for prime time. And I can say that as a musician. Our goal in Pandora is not to have everything. It's to have the best stuff in each genre. And to have a lot of stuff you'd never heard before. Yeah. How are you making those judgments? It's a subjective editorial judgment. But it's made by people that are trained musicians. And the judgment is not, do I like it? The judgment is, OK, I may not be a fan of goth metal, but I'm a trained musician. I know what a well-done goth metal song is. And I actually, I spend some time periodically just sitting in on the reviews. And there really is, I come away always pretty confident that there actually is a pretty solid basis for judging songs. It doesn't mean that we'd always agree. And I hope it becomes an ever more controversial topic, because that means the stakes are high. I hope the stakes are huge, eventually, for getting in or not. But I actually feel pretty good about how we do that. Yeah. Will, do you see that the software at some point will be able to take a theme? And then it will play itself with the variables, and then at some point start creating something completely, basically improvise with continuous variation of things. So do you think that we can take the genome and compose music from it, essentially? Pretty much. So you start with, let's say, help by the Beatles, and then progressively the genome starts playing with your own variables or other people's variables. And then after five minutes, it's an independent genome creation? Probably not. I think it's an interesting idea. We just, even with 400 attributes, you are so far from composing something. Even with that much detail, there's a huge, huge distance to travel from there to writing a piece of music. And so it may seem like we have the ingredients for composition, but we really are a long ways away from that. So the same way, I don't think we can pick hits. We don't have the raw materials to do that kind of stuff. Yeah? So you said that my hits or misses affect me immediately, affect my list. How do you treat other people's lists? Will my hits affect other people's lists? Or do you try to cluster people with similar tastes to generate a list on this and that? Yes, the question is how do we, how do thumbs affect not just you but other listeners? And so they do affect your own station, but they also are collected on aggregate. So we have about 2 billion thumbs so far on Pandora. And we know for every song, its Q score thumb up to thumb down ratio in every context for which it plays. So we know how well Celine Dion does on Sarah McLaughlin. And if people who launch the Sarah McLaughlin station systematically thumb her up or down, we'll increase or suppress her frequency for everybody based on the wisdom of the crowd. So we do incorporate that as a layer onto Pandora. And that helps us particularly with people like Frank Zappa. Like we cannot deal with him. The genome is confused. Or same thing with Beck. Just very iconic, just pain in the ass for us basically. The genome can't capture them. The audience weighs in and can be pretty effective at building stations over time. Yeah. I know people that you go to your page and see what kind of station you're listening to. Do you ever put any thought into having, say, a celebrity, a Rick Rubin come in and make his own Pandora page and people can access that? So the question is, have you ever thought of creating celebrity pages for them to? You could go, not that I care anything. Like Rick Rubin. Well, Rick Rubin's a big fan of Pandora actually. Prince of a guy, too. But Pandora is very explicitly not about celebrity. So you've never seen Pandora say, we think this is a good band generally. Like this is our top 10. We try to avoid that at all costs. And we don't think that what Rick Rubin likes has anything to do with you. And we really want to, our whole mantra as a company is, let's figure out what you want. And try to sort of, we want you to feel that when you are a Pandora, Pandora is all about you. Not trying to kind of turn you into something else or tell you what's cool and what's not. But really sort of respond to you. So very, very unlikely, unless that was in the context of an advertisement, different rules when it's explicitly within an ad campaign. But not on the site, probably. Yeah? I choose you a genius feature. No, we have no relationship to genius. I'm just curious. Hi, how many attributes, pretty interesting that you display if you're like wondering why this song was chosen? How many of those attributes do you think reveal maximum? And it seems like you had a whole page when you were talking about full height a little bit. How many can that as a user see someday in the future? So the question is, how much of this genome could we reveal on the website? And those actually aren't genes. We call them focus traits. They're actually combinations of genes. And we have thousands of them. So we can actually put sort of gritty male vocalists. That's obviously multiple genes rolled together. And we can create, if you think about it, 400 to the 400th power or something like that of those by combining and recombining. So those are really more than anything to be explanatory of how the system works. Because we actually did some user testing of Pandora in the very beginning. We had about a dozen people use it. And we watched them from behind a one-way mirror. It's a really fascinating thing to do. And they would use Pandora and then narrate out loud what they were thinking as they were using it. And you watch them interact with it. And they were all asked at the end of that, one of the questions they were asked was how does Pandora create playlists? And they all said they play the songs because record labels pay them to play them. And the thumb feedback is sold to record labels. That's because people's legacy of radio is so much around Paola. And so we put those pains in for that purpose. They'll say, no, no, no, no, no, it's because of this stuff. We actually had a sentence on top of the tuner saying, this is how we make playlists. But people don't read it. That's the first thing you learn about UI design. They don't read stuff. So it's got to be intrinsically obvious. So that's what we did then. You mentioned before the advent of new business models where you have online services doing direct deals with the labels. What do you see as the threat from interactive guys, like I mean MySpace music, where they don't have the playlist restrictions that you have and couldn't hear you overlay an IP recommendation music discovery model? So the question is, how do we view competition from sites that have direct licensing and maybe some more flexibility to offer customization and whatnot? I think the best way to answer that is to look historically at how people consume music. And, excuse me, average American listens to about 20 hours of musical week. And historically, about 17 of that has been radio. And only three hours of it is music you own. And I think the reason is because people don't like to spend time curating their listening experience. And I think that is going to generally remain true. Even with all the free, even highly interactive on-demand services where you can go in and really kind of cut and paste and build things, that I don't think people really over time are going to have an appetite for that. So I actually think that in the long run, the site that will develop a massive audience is one that does not need to have that stuff. So if you believe that, and you may not, but if you believe that, then it goes to business model. And the cost of direct licensing is prohibited. It is way higher than it is for webcasting. And it's hard enough to build a business, even with the rates that we'll wind up with here, with webcasting rates. And it's just hard to imagine you can do it on-demand stuff. And I think between those two things, it feels like we're in a good spot. We don't think of those companies as our competition. We think of Clear Channel as our principal competition, or all the broadcasters, or anything. Yeah? How does what happen at Wall Street with the subprime MD, call of blood, Wall Street, and the credit crisis protection, consumer demand, affect advertising with the transaction? So the question is, how do we think Wall Street's going to affect our business? It will certainly constrict advertising money. I think it'll mean a slowing of the growth of online advertising. But I think there's going to be a flight to quality. So I think if you have a website that can create interesting ad campaigns that can show results, I think you're going to be somewhat immune, not immune, but buffered pretty substantially from the downturn. Because people will still have to advertise. They still have to sell products. And they'll be more picky about how they spend their money. And I think what you're going to see in the next 12 or 18 months is a winnowing out of businesses whose ad systems are actually not effective. I think a lot of companies are going to disappear, because it turns out people aren't looking at the ads. They aren't interacting with them. We actually don't. We're very bullish. Pandora is growing quarter by quarter substantially, even the Q4 this year. And I don't actually see us being hurt as much as you might think by it because of that. And also, do you see Pandora as a competition to, or as an alternative to record music? Do you really look down on it? So the answer to your first question is, how does licensing work with cover tunes? That's the domain of the performer. So provide they've got the licenses, they've done it properly. That's not Pandora's and play in that particular arena. Would it have to be a published song? Yes, yes. And then, did I say that right, John? I hope I think, yeah. The second thing is, do we see ourselves as competition to record music? On the contrary, Pandora is one of the top affiliates on iTunes and Amazon. So we sell a ton of music. 40% of Pandora listeners buy more music since they started using the service, and only 1% buy less. And there's actually been a bit of a positioning done by the record labels about this saying it's cannibalistic of CD sales. That's not true. It's hugely promotional. So I think it's vital. And I think that if web radio were to disappear because of higher rates, which I don't think it's not going to anymore, but if it did, I think you'd have seen a pretty substantial drop in CD sales. I think it's really beginning to have an impact. And it is absolutely promotional. Yeah. So the questions are the disposition of record labels. They're schizophrenic with us. They send us the new releases the day they come out, or before, and then they try to drive rates that would put us out of business. And I think, you know, if you go to a record label, off the record, there's not a single person in a label that would tell you that Pandora's not valuable to them. I think the problem is that they're trying to figure out how they survive in this, you know, in the next sort of digital evolution. And they see web radio as potentially a place they can either grab a lot of money or they want to control it, at least. And so you've got, I think it's a bit of a disconnect. There's lawyers and business affairs people who are driving negotiations but aren't as closely connected as they should be, people who are actually sourcing and promoting bans. So they kind of operate. We have some very strong promotional relationships, meaning we're sharing data, like helping labels with stuff, and that same label is being the most difficult in a negotiation. So, you know, it's yin and yang. Yeah, Wayne, back. So you've got an ad product. So your opening comes from Brad Advertising. You've got an ad product that's a little non-traditional in that, you know, you've got this model where, you know, you only change the ad that's shown when someone switches back and interacts with a player. You've got an ad product where the agency has to do a lot of, you know, specific work to design this whole gorgeous experience that's seen in front of you. How difficult was it to convince agencies to pick this up and push it to the advertisers? And how do you major success when you're talking about that flight to quality? So those are really good questions. The, we have, I would say the last, the first couple of years were primarily about evangelism in terms of the agency community. And going out and saying, no, no, no, people do look at it. They do look at it. They do look at it six, seven times an hour, you know. We actually do all this design ourselves. So we have a team of seven designers who do all the ad work. So we actually provide a solution for agencies. And I think that's hugely beneficial to them. But there's been a lot, you know, there's been a lot of education that's been necessary to do that. And I think we're just now hitting kind of a hockey stick in terms of awareness of Pandora, you know, confidence that Pandora will be around and critical mass of audience. I think those things are all going to start churning. And sorry, the second question was, how do you, you know, demonstrate or major what kind of metrics can you, so we have, we track like mad people on Pandora. So every click, every view, it's all tracked. It's all delivered as reports back to agencies. So they really know what's happened. They do ran recall studies openly over the web. And so there's lots of ways to measure how effective it is. And they do, we do very, very well. We have 80% repeat business this year already from last year's advertisers. So, is that proofs in the pudding? Yeah. Do advertisers ask to be displayed next to music that has certain... The question is, do advertisers ask to be displayed next to specific music? And yeah. These are targeted by age, gender, zip code, and the kind of music you're listening to. So a car company can come to Pandora and say, I want to advertise this new vehicle to women in their 40s who like punk music and live in Minnesota. And what kind of car that would be? A Volvo. And they can do that. And that's part of the promise, the future promise of this radio. One is the incredible targeting. And the second thing is that you're actionable. You know, when you're listening to radio, you can't actually do anything. You can't pick up the phone and call, but here you can actually sign up or, you know, buy something, have an action. And I think between that and the targeting, it could do very well economically compared to broadcast radio over time. So you've got at least three major contributions you're making to the world. One is you have a taxonomy, or as fancy people would call it an oncology, that is the structure of the genome that lets you slice and dice music in various ways. And I don't know if you would paint a bunch of property connections in that way. Then you've got the ratings done in the sweat... I mean, in the... by the good musicians. They have full health insurance, flexible hours. We do chain for the chairs, but those come off in five. Exactly. So then you have the ratings, which I imagine is a trade secret. We want that leaking out because that maybe genius would get smart. And then you have the matching engine that on the basis of those ratings, and the thumbs up, thumbs down, produces the playlist. You could see producing open APIs to some combination of that stuff. For example, I might think I've got a really good formula that if I knew the oncology, even if I don't know the end of the formatings, I could say when X of tanges do Y, this kind of song should be linked with that kind of song. And wouldn't it be interesting to have a Pandora-powered open source API so others can try their hand at doing the alchemy that leads to the playlist? The short answer to that is yes. And we do have an API now. It's available only to business partners, and it's a lot more shallow than what you've described. That's just the way of skidding Pandora. Yeah, I mean, no. Hardware companies can build it into their home devices, but it's essentially... It's not letting you deep into the guts of the genome. But I can certainly imagine that becoming a much richer and more open API than it is right now, because you have to be a company with serious distribution plans to get access to it right now. One of the challenges for us is we have a lot of legal sort of liabilities around how Pandora is distributed, things like the sequencing of songs, which you're allowed to do and not allowed to do. Plus it costs a lot of money to stream that music, and we've got to be able to monetize it. And so if we have thousands of people delivering free versions of Pandora that can't support advertising or periodically violate the DMCA license, it's our door that's going to get the knock, you know? And we... I think eventually we'll hopefully can figure out a way to allow that sort of great developer community to go in and harvest this, but it's not going to happen right away. There's some, you know, exigencies for a young company like ours, we just have to be careful about at this stage. But I'd love to make it available for educational purposes, you know? I said exigencies because you said ontology. Yeah. Did you initially get the word out of the service? The word out? How do we get the word out? I would say blogs were the... We've never advertised it, and there were two bloggers, Robert Scoble and Michael Errington, who are two very, very prominent tech bloggers that wrote very glowing reviews of Pandora. And they each probably had at the time 300,000 or 400,000 people that subscribed to their blogs. Like, that's all you need. I mean, the extraordinary thing about products these days is if you build something that people really want and it works well, you know, there's a demand for it. Like, it will out, you know? People will try it, people will tell each other about it. The average Pandora listener tells eight people about the service. And you've got so many cool ways to do it. You know, blogs, emails, Twitter, Facebook, MySpace, like all these things, just... And I've been traveling around the country, you know, this is one way of telling people about it. You already know about it, but I've given about probably 250 town hall meetings in the last three years in everything from Biloxi to Boston for local Pandora listeners. And what you meet there are, you know, people who are really evangelists about the service. And they have incredibly powerful tools to do that with. So, you know, it just flows. And it's incredibly fun to watch that. I mean, when we... Remember that ad for IBM where that company launches their website and they're watching the ticker? I'm like, yeah, yeah, yeah, oh, shit. You know, and then it's like... It was for FedEx or something. Like, we had that experience. It just blew up. We had a... We were at FedExing Harbor and trying to keep up, almost went down a few times. And it's all just completely viral. You don't spend a penny. So, yeah, it's very exciting. Yeah. This is a very practical question. So, I teach first-year law students, and they come to me and they say, I'm really interested in music. I love it. I want to be involved in music industry, but not in a record label. I want to do something in law and music. You're one of the lawyers who serve your needs best. And what advice do you have to young lawyers or one of the lawyers who would want to work with or on behalf of a company like yours? So, you're asking what type of lawyer? Not names of individual lawyers, right? Yes, not names. So, I'd say we have two kinds of lawyers that we work with. One are sort of straight-up contract lawyers because we do a fair number of contracts with business partners and... The best lawyers in that domain are people who have very strong conceptual skills. Because you're dealing with new fangled, weird arrangements, and you just have to have the ability to kind of synthesize the kind of... Your needs, your capabilities with the needs and capabilities and sort of agendas of two parties. And that's just about being kind of... It's about conceptual knowledge. And I don't think it necessarily means tons of experience. I think we've had some young lawyers working on stuff for us. So, I think contract folks. And then the other one I would say would be lawyers who have a foot... who have had a foot in the music industry in some capacity. So, understand the psychology and the philosophy, I guess, of artists, record labels, managers, the ecosystem of music. But also have a strong appetite for technology. And I think that marrying art and technology to me is the nexus of legal work for the next 20 years. I mean, it's a rare combination. But that is where artists will live or die. And I think if you weren't a lawyer and you wanted to work in the music business, I think the best job you can get is grab a handful of bands and manage them. Because you can do that now from a desktop computer and you can do a lot of damage with just a laptop. If you know how to navigate online communities and build word of mouth and so on, and bands are gonna learn that, too. And bands are gonna realize they need that... that I call the digital fifth beetle. You need someone whose job in your band is not to play an instrument, but it's to amplify what you're doing through the web. And an incredible possibility. And we sent out little cardboard flyers. And we were putting posters on telephone poles. That's not scalable. So what it is now, you know, this kind of stuff. How are you doing? Okay, way in the back. Sure. One is just if you're tracking numbers, are you seeing people directly buying musical tracks on the iPhone because you've got that click-through on it to the application and so what kind of numbers you're seeing from that? Is Yahoo approached you or a similar curator, say, Picasso, Google, about applying the same rating model for professional photographers to images? So the first question is, have we noticed an uptick in sales directly from the iPhone? I don't know if we've teased out that particular number. I'm sure that it is, though. 10% of our listening is on the iPhone now. We have almost 2 million iPhone listeners in just four months. And they listen on average an hour and 40 minutes a day. That's, like, absurd levels. And it's one-click purchasing, so add that all up and there's going to be purchasing happening on the iPhone. And then in terms of applying this to photography, I think photo is actually a perfect application for this approach. You need to hire a different set of experts, but I bet you if we spent three hours together in a room, it would be like an axonomy, just like ours, but photography. And it would be involved with images and colors and focus and dimensions. You could all probably do something like that. And it will be done. It's probably already being worked on right now by a dozen teams around the world. So I think you'll see more and more of that in the next couple of years. Okay, well, thank you very much. Sure, my pleasure. Thanks.