Tuesday, December 05, 2006

Pandora Town Hall, UC Berkely, Dec. 4, 2006

Last night I attended a second Pandora Town Hall over at Cal in the Heller Lounge at the Student Union. I’m fairly sure that that room did not exist when I attended Cal – it’s one of a very few rooms in that building where I did not play music at some point. I played on the same bill as Sheila E., of all people, upstairs in the Ballroom once. Not that surprising really: she’s a local girl, and the director of UC Jazz at the time was well acquainted with her dad, Pete Escovedo. I also played a private goodbye get-together with my favorite group of street musicians up on the second floor as well. Ah, the good old days…

But back to the Town Hall. It was a warm winter evening in Berkeley and the room was hot and stuffy. About 150 people showed up, and there was still room for more. The crowd seemed to be about half grad students and half community geezers like me. This town hall had a much smaller presence from Pandora than the San Francisco one did and was, I suspect, a much more typical version of the Tim show. We started about fifteen minutes late.

Tim greeted the audience, took a couple of immediate questions, but the second, “What is Pandora?” launched him into a abbreviated version of the history of the company and a description of the player. The crowd seemed to be somewhat more conversant with the player than the earlier one.

Since July the number of analysts at Pandora has increased from 42 to 45. They are now analyzing tracks at about 15,000 per month compared to 12,000 then. Registered users have increased from 2.5 million to 4 million. My notes say that they now have registered 3 million feedback thumbs, but, given that that number in July was 100 million, I suspect that that number is really 300 million. There are now over 500,000 songs analyzed and available to play on Pandora. Pandora is still hiring which remains a good sign.

Tim outlined the current top 3 priorities of the company to be:

  • A mobile version of the player.

He noted in answering a question later that the infrastructure to support a mobile version of the player is not really available yet. It would really help to have a cheap cell phone connection to the internet or municipal wifi in place to get Pandora to be mobile.

I’d love to have a mobile version of Pandora. However, I seriously might never buy any music ever again were Pandora available everywhere. Thus, I’m not really convinced that Tim’s vision for a $100 billion annual market for music is served by mobile Pandora. It is, however, a service I would be willing to pay, say, $20 per month to have, and while that might help Pandora, I don’t see how that money would get efficiently back to the musicians.

  • International Service

Pandora continues to explore how to legally offer their service to people outside the US, and how to incorporate music which has not had a US release.

This goal is probably the one that I’d most like to see happen. I suspect, however, that progress in this area will be slow and incremental.

  • Listener to listener interactivity

I’m becoming more cynical about this goal. I love to talk about Pandora, but I’m finding that very few people do. Pandora-forum.com peaked at a total of four regular posters before it died a cold, cold death. There’s a lot of enthusiasm when people initially discover Pandora, but I think for most people that once they’ve incorporated the player into their lives, it becomes the music spigot, and they’re about as interested in talking about it as any other utility. There is probably more interest and activity at the P,G and E forums.

Tim, as usual, opened the floor to questions.

The desire for Classical music was once again expressed. Tim still agrees that it is something that he wants to see on Pandora, but he noted this time that it requires a different musical training for the analysts.

Someone raised the issue of the traits listed for the songs. Tim polled the audience on how many would like to be able to build stations based on traits, and the response was quite strong even when he pushed back and asked if we would do so regularly.

Tim noted that the guy in charge of designing the interface had a background in designing interfaces for children’s toy, and so was always pushing to keep things simple.

I believe that the simplicity of the player’s interface is the single most important key to Pandora’s success. The first thing that anyone should see of the player is that if they type in an artist or song that out will come an unending stream of related music. Yes, I want to get under the hood, and I want more tools to design stations than we currently have. But those tools should be relatively hidden. The first version of the player that people see must remain simple and elegant.

Tim also noted that the traits that we see on Backstage and which have been scraped onto the wikipedia page are not the traits in the genome. He called them “focus traits” and said that they were aggregate descriptions of clusters of values of the underlying genome traits. This revelation helped me to understand Pandora a bit more. The focus traits are probably being assigned to clusters on an ad hoc basis, and they are probably floating all the available resulting focus traits up to Backstage which is why there are so many more focus traits for the standard catalogue than for the Holiday songs. They probably just have not assigned that many focus traits for the Holiday genre.

The editorial process was addressed when one audience member mentioned that several bands he knew had sent Pandora their CD, but the music has never appeared. Tim noted that the process takes about five weeks, and so if the music had not been posted up by that point then it had been rejected.

I’m actually fine with the foraging process and the editorial decisions except for Pandora’s dirty little secret: they don’t analyze all the tracks on a CD, and when you successfully search for a track with the player that has not been analyzed the player does not tell you that fact, but, apparently, instead generates music as if the corresponding artist had been selected. One of the regulars at the now defunct pandora-forum found at least one CD for which only a single track had been analyzed. I am particularly frustrated when I discover that tracks that I particularly like have not been covered on a CD when tracks I like less have been. There seems to be no hope that such tracks will ever be incorporated into Pandora unless many people search for them, and how likely is that going to happen for pieces like The Bobs’ “Share A Load”?

In response to a question about musical genres, I noted how much more focused the Holiday station based on all the available version of Ave Maria is compared to the same station based on the non-Holiday versions. Part of that is the fact that there is very little Classical music encoded on Pandora, but it also suggests that meta-information tags like “Holiday” can make a Pandora station much more focused and enjoyable. Tim noted for a related question that they might be open to letting people import feedback from other music services.

A couple of different questions arose around the interaction of people’s music collection and Pandora. One gentleman wished Pandora could read his iPod track-list and not play any tracks he already had. Modulo the usual copyright issues, that would seem to be a reasonable request. Another guy wished the tracks he owned could be played they way Pandora does. This one is much trickier. If Pandora has not analyzed the tracks you own then it can’t really assess which tracks are related to each other. It might be possible to match your list of tracks and indicate when a track already own is playing, however. Which reminds me, I’d love to have flags indicating when a track is “fresh” (analyzed in the past three months) and “new” (released in the US in the last twelve months), and then be able to emphasize or de-emphasize fresh, new or owned songs for each station.

Another audience member inquired whether Pandora would ever be interested in starting a record label. Tim said no, because the principle to which they aspired was one in which no artist or company could buy preferential treatment from Pandora. The audience member then ranted about iTunes, how Apple undercut fair returns to the artists in favor of sweet deals for the record companies in order to sell iPods. Tim offered an alterntive vision of a distribution system in which people were connected directly to artists, paid, say, 25 cents per song but the artist received 15 cents of that rather than the 2 or 3 cents that they receive under the iTunes deal. Tim did not say how or who would engineer such a distribution system.

Someone asked if Pandora was going to create a genome for movies. Tim did not answer the question, but did note that someone did recently call him asking for guidance and help in creating a genome for wine. He said that it was likely that we would be seeing several other genome approaches to identifying products in the coming years.

The crowd began to get restless at 8:30, and so Tim brought the meeting to a close and released the swag. I spotted two hat designed and three new t-shirt designs.

1 comment:

TimofSuburbia said...


Great post! Thanks for the details regarding the meeting. A lot of the same topic were discussed at the Durham, NC meeting and I think you are right on the money regarding folk's inclination to discuss Pandora after their first introduction.

The future is wide open for Pandora if they can figure out the legal issues and make it a profitable business.

Seems they need to just go ahead and hire some expertise in classical and other genres instead of talking about getting additional training for their current staff.

Pandora.com seems to be approaching the Web 2.0 phenomenon sideways, but there focus on growing the database instead of the interface or the community building aspect seems to be working.