Sunday, March 2, 2008

Pandora at Sloan

I want to say thanks to Tim Westergren, Co-founder of Pandora for speaking at Sloan this past Wednesday. Tim gave a great talk on the founding of Pandora, the future of the music industry, and the Music Genome project.

One of the most interesting aspects of Tim's talk was his explanation of Pandora as part of the music discovery process. Because Pandora uses over 400 objective criteria to rate each song, the music discovery process is not driven by the latest fad or ratings that trend toward the mean. Pandora spends between 15 - 45 min rating each song. In addition over 70% of their music is not on a major label. For my money Pandora clearly has nailed one piece of music discovery on the internet. Ultimately they still need to dramatically improve the ease of purchase. Tim mentioned that many of their artists are not on iTunes or Amazon.

For the first three weeks of Pandora's life the company attempted a subscription model. When this was a dismal failure they moved to a free service. Currently, Pandora is supported by ads and a revenue sharing agreement with merchants like Next time you want to buy a CD (or other item), help Pandora out by clicking through their site.

Pandora recently sent me an email with some tips for improving my station:
Want to create a station with a very specific sound? Consider giving a thumbs up to a song only when you like (almost) every aspect of a song, because each thumbs up will add more songs to your station, perhaps in a range that is wider than you had hoped.
Thumbs down has a less dramatic impact than thumbs up – it simply bans that song from that station, so consider using it liberally for any song you don’t like, or for a song that just doesn’t ‘fit’ on that station.

1 comment:

Anonymous said...

That's an interesting tip about the thumbs down feature - thanks for posting it. It would be cool if you could change how it worked. Two options would be (a) don't play this song (like it currently works) or (b) don't play any songs like this (how I wish it worked!)

I also wonder how scalable their human labeling approach is., in comparison, does recommendations automatically through collaborative filtering of user tastes, and I've found their 'Neighborhood Radio' to be pretty close to my tastes (and free to listen to).