Last week Tim Westergren, the CEO of Pandora, came to speak at MIT Sloan. I have been a huge fan of Pandora ever since it launched 2,3 years ago so it was exciting to meet the man behind it.
Pandora
When I want to listen to new and good music, I go to Pandora (here’s my Pandora profile). It’s simple and it really works. Just last week, after the talk, I told Ting about the site and she wrote me later gushing about her addiction to her Kelly Clarkson channel (Kelly Clarkson? hmph). I haven’t found anybody who didn’t love the site once they’ve tried it!
From Tim’s talk, I found that after all these years Pandora’s music “engine” is still powered by the human ear. This is pretty interesting since not long ago I was just having a discussion with some one who thought Pandora must have figured out a better way to automate and scale its database by now. Well, I got it straight from the horse’s mouth: of the 120-ish employees at Pandora, 50 of them are “musicologists” who sit around all day long listening to new songs and rating them based on 400+ dimensions. It takes ~15-45 minutes to analyze each song. Tim joked about taking great pains to beat the personal bias out of musicologists to get an objective analysis - I had a vision of a poor struggling musician being locked in a tiny room listening to crappy techno all day long. He also defended Pandora on not factoring in social context and popular opinion - the phrase he used was it “eliminates social snobbery”.
He gave an anecdote of a user who emailed him - upset that on his “Enya” channel, a Celine Dion song - with a rather cheesy reputation - showed up. After a few back and forth emails, the man concluded, “Oh my gosh. I’m a Celine Dion fan!”
I suppose that’s a pretty good argument and directly contradicts some of the other recommendation analysis engines out there such as The Echo Nest, which purports using sentiment analysis of a song or artist on the blogosphere. I still think if there was a way to automate the music analysis portion of the business Pandora could scale its database a lot better.
On Music Subscription
Tim also stated that he doesn’t believe in music subscription services because 1) it’s “not the way people are use to paying for music.” This point I don’t agree with since I am the kind of person who would pay for subscription music. The radio is great but I also want control and access to a large library. 2) He also referred to the regulatory hurdle related to on-demand subscription service as a deterrent - each song has to be individually negotiated, thus making it impossible to have a large database of songs. On the other hand radio play requires just one RIAA license.
On Tim
I think above all I liked Tim’s personal story as a struggling musician turned entrepreneur. He’s a great story teller and has a ton of self-deprecating charm - taking us through Pandora’s early struggles (and even getting sued by employees by giving them IOU’s!) to its eventual popularity.