Thursday, July 20, 2006

Pandora Townhall Meeting, San Francisco, July 19, 2006

Last night I attended a town hall meeting led by Tim Westergren, CSO of Pandora. I took notes at the request of the folks over at Pandora-forum. I'll include my reaction to the various issues that came up in italics.

It was an beautiful evening in San Francisco. The temperature was downright balmy with a good breeze keeping things cool without the usual fog chill. The event was held in the small Zeum theater. By the time the meeting started twenty-odd minutes late every seat in the roughly 200 seat venue was full.

Tim introduced himself by giving a brief history of these meetings. Starting in January with a visit to Portland where two fans showed up, and including one stop in Texas where no one showed up at all, these events have become increasingly popular with 150 coming to one at NYU, and, of course, now the San Francisco event was so popular that they're having an additional event next week to accommodate the overflow.

Tim started by setting the agenda for the meeting. He planned to give a history of the company and the website. Since the whole purpose of these meetings is creating dialogue with the listeners, he encouraged people to ask questions and to feel free to depart from that agenda.

He remarked that Pandora was actively hiring, and pointed out the HR person in the back to talk to if anyone was interested in working at Pandora.

It's an extremely encouraging sign for the health of Pandora that, like, the third thing out of Tim's mouth was that they really need people. As we shall see, this last seven months as been a period of explosive growth at the company.

Tim mentioned briefly that the company is a bit over seven years old now. They landed their first bit of capital in March of 2000 immediately before the bubble burst.

He then turned to his personal history. He had tried to make it in a rock band. They toured throughout the West in van, often passing other bands on the road between gigs. It was clear that one of the major problems for musical artists in general is connecting the artists to their audience.

Tim then went into film composition where directors were often looking for Tim to create a song like one particular one the director had in mind for a scene, but that would be cheaper than acquiring the rights.

In the Internet culture of that time there was a hope and belief that the music industry would be turned upside-down: the digital technology would allow artists to produce and distribute their music without becoming beholden to the major labels. Amazon, for instance, launched a program called "Amazon Advantage" in which the company offered to warehouse any music (say, 5 cds from a band) and then sell them from the site. The problem remained, however, that the artists could not find their audience.

Tim was living Palo Alto when he came up with the idea for the genome. Pandora now has 42 music analysts encoding that genome for songs. They analysts spend an average of 20 minutes on each track. The investors complain that the process is not scalable, but Pandora maintains that in order for the genome to work the listeners must be trained.

We'll come back to this point in the Q & A, but I believe that Pandora is exactly right on this point.

After the first infusion of capital in 2000 they ran out of money after one year. The employees did not take salary for two years, contributing time as they were able. In March, 2004 they landed an additional $8 Million. During that period they had licensed the technology to a few companies, but the big thing that happened during that period was that broadband penetration in the US had grown dramatically so that a critically large market had become available for digitally streaming radio.

Further, the DMCRA (the audience boos) cleared up licensing issues. Under the act there was a spectrum of royalties from free for broadcast to specified amounts for digital sales of tracks with digital radio somewhere in between. In became clear that a good use for the Pandora database would be the generation of play-lists.

And so they changed the name of the company and spent a year and a half building the Pandora device. They tested the product by trying it out on 200 of their friends and family in October, 2005 telling the testers not to spread it around yet. Within a week there were 5,000 people listening. The genie was escaping the bottle, and so they massively accelerated the development of the software, and started charging a fee for the service to stem the tide. People just re-upped for the free first ten hours by cleaning their cookies etc.

Thus, they launched the free service in November. At that point they were Slashdotted at which pointed 250,000 visited the site in a single day, and bandwidth had to be expanded dramatically. At this point 2.5 million people have created an account. Since the site is supposed to be limited to US listeners, they require the listeners to supply a zip code. The most popular zip is "90210". Tim: "That's deeply depressing."

Now the company is focused on building out the corporate infrastructure, creating a mobile product and figuring out how to work out the licensing issues for moving legally into International distribution. (Tim is completely aware of the popularity of Pandora in other countries, and he really wants to make the site global, but it's a tricky issue since there is no international equivalent to the DMCRA for Pandora to leverage.)

Considering the industry problems from the musician's side, there are 270,000 SKU's generated in the US for recordings each year but only 5,000 of those recording account for 90% of the sales which are currently $10 billion. Tim believes that if the artists can be connected with their audience more efficiently that the sales should be more like $100 billion. His vision is that with Pandora and other internet sites facilitating the connection between artist and audience that the average musician will be able to have a good, middle-class career doing what he or she loves.

Tim points out that the cost of producing an album has dropped to next to nothing in comparison to prior decades. Quality mixing software is essentially free on modern computers. Furthermore, distribution costs have plummeted as well. Thus, the remaing production costs are purely about promotion which is where Pandora is working to fit in.

Tim then opened the floor to questions.

The first questioner referred to an essay by film director Jaron Lanier. (I've tried to find the essay in question. It may be this one.) The questioner summarized Jaron's thesis as saying the music is becoming more homogenized because the available tools limit what can be produced. (Maybe his fifth point in the link?) Tim disagreed, and promoted the position that Pandora should work against such trends.

Okay. Jaron is a friend of a friend, and a couple of decades a ago I got to hang out with him a bit. Jaron is WILDLY eclectic when it comes to music. I'm eclectic: he's orders of magnitude more so than I. Back when he was living in Palo Alto, his livng room was filled with the entire inventory of a typical Lark In the Morning. You'd ask him to identify and instrument and he'd pick it up and start playing, say, a nose-flute. All I'm saying is that Jaron's standard for diversity in music may just be a wee bit wider than the average Pandora listener.

Tim mentioned that over 100 million feedback thumbs have been clicked at this point.

That's a whopping 40 per account. It's pretty clear that most Pandora listeners are not concerned with evolving their stations.

Tim mentioned that they did run a test allowing people to give any answer as to why they gave a thumbs up or down. He indicated that they will be incorporating the results of that test into future features.

Tim dodged answering when they were going international (he deftly dodged all "when" questions). He did mention that they have two full time employees working an the problem at this point.

Someone asked how they plan to make money. Tim stated that advertising should drive the business. They would like to keep the advertising entirely graphical, but even there they've been choosey. They refuse to accept poker ads, for instance. However, the question left hanging unasked in the air was whether they would ever have to admit ads into the audio stream.

Tim was asked whether Pandora was pursuing business-to-business opportunities. Tim responding that they do not wish to lose their current focus. He did mention, however, that the software is currently being used by directors to find similar songs for their films, and bookers for musical venues are similarly using the software to find similar bands for engagements.

Someone asked about getting Pandora for their home system. Tim pointed out the Squeezebox principal in the audience and mentioned that Pandora accounts for 50% of Squeezebox listening at this point.

Tim promised that there will be more features for power users.


Tim surveyed the audience if they felt their Pandora stations were too repetitious. There was a strong minority of hands up on that one.

I raised my hand. I can build a station with a wide variety of music that is not repetitious; however, I still hear the same track on the same station within a six hour listening session. I doubt many are doing six-hour listening session as am I. And so I'd just like the option to increase the time between repeats beyond the industry mandated minimums.

He pointed out that they've made a conscious effort to keep the interface simple. He talked about interface testing and the frustration of watching people trying to use the program from behind a one-way mirror. "The button's right there!"

A DJ from Sweden pointed out that LastFM has much better features for musicians. Pandora wants to provide similar features for artists. Currently, they'll provide some information for artists when requested, but it's not automated yet. Tim made the case that Pandora does not require an initial tagging by an audience for a song to go into rotation in a station.

Tim surveyed the audience about iPod ownership. He pointed out that studies have shown that people basically go through a burst of ripping when they first get one, and their radio listening drops. But after a few months they are listening to more radio than they did before the purchase.

There were a couple of A&R representatives from small labels in the audience. The first asked if Pandora would ever take money from labels to get their artists included in the database. Tim did say, "We're never going to slot music because someone is paying us." (the audience cheered.)

I interpret Tim's response to mean that they won't force songs to be selected more frequently by stations, but that response doesn't quite address the question about how tracks are selected to be encoded. Would they take money to get a release encoded? Or, worse, prevent other labels tracks from be being included? I doubt the latter, but what if BMG said, "Look, we'll pay the salaries for all analysts necessary to encode our entire output?"

Tim was asked about the process for selecting which tracks to encode. Pandora currently has two employees whose entire job is foraging for new music. They receiving about 30,000 to 40,000 suggestions for tracks each month but only have the capacity to encode 12,000 currently. Tim did say that they do not envision Pandora striving to be completist. They see the decision for encoding a particular track as an editorial decision. He did not address how such editorial decisions are being made in general. They see themselves as the conservators of a music library.

If so, why The Shaggs, Tim? For the love of the Goddess, why The Shaggs?

The other A&R rep asked why Pandora required the product encoded to have an SKU. Tim responded that they structured the database to use SKU as the key identification number, and told the rep to get UPC's for their product: it's apparently cheap to do so.

Someone asked if they considered scaling up the process of encoding the music by setting up something like a wiki process. Tim responded that they had tested the idea, but that without the four days of training that they put their analysts through, the average listeners do not agree on what the various genes mean in practice. He can have two trained analysts analyze the same track and get almost exactly the same encoding which is not the case for untrained listeners. He mentioned that 10% of tracks are reviewed twice to confirm this consistency.

As I mentioned above, I think that Pandora has got this part of the process exactly right. That is exactly the kind of control that one would like to have in place.

Tim mentioned that they do get a commission on sales through the site, but they did not see that as a large source of potential income.

Tim closed on a non-question from a listener who was pretty much overwhelmed (positively) by all the new gospel music she had discovered through Pandora. That audience cheered the shared experience.

Pandora provided soft drinks for a mixer after, and Tim introduced several others form Pandora that people might like to talk to. They offered free t-shirts and caps to the people who came to the meeting.


Anonymous said...

Great report! I hope that the other people attending this meeting might put up a few of their opinions here or at the forum.


~TVS said...

Wow, maybe I should have taken actual notes (instead of mental ones) at the Orlando meet-up.