User Analysis: Amazon vs. Pandora

Some of the main areas of focus we've talked about in this class concern the use of user data by big business to observe buying habits. This is obviously an important area for businesses to be able to apply what they learn by buyers and show them other products they'd be interested in while they're browsing. However, what's the best model?

Amazon uses a user's purchase and compares it to other buyers that purchased the same product to suggest other products that user may be interested in. The underlying assumption here is that those that buy a product have the same interests as others buying the same product. And while it would be difficult to predict what a user would want to buy based solely off of a single purchase, there still seems to be a better, more systematic approach that would yield a better selection.

Pandora Radio is an online service that allows users to create radio stations based on individual tracks or artists. The user can then vote a "thumbs up" or conversely when they like the track or don't. The station is known to most as extremely accurate in how good the music selection is (of course some critics disagree). So, this made me think - how can Pandora be doing it so much better? I have never been on Amazon or a similar website and seen a "suggested product" that I thought I'd like to purchase.  So, I did some research.

Pandora's underlying momentum is carried by the Music Genome Project. The project relies on the analysis of 400 musical attributes that took 30 experts 5 years to complete.  As of 2006, the library consisted of over 400,000 songs from 20,000 contemporary artists. So, we essentially have a giant optimality algorithm that takes in all of the "thumbs up" and "thumbs down" that a user inputs.

This difference presents a unique perspective into the apparel merchandising arena. Perhaps Amazon should consider attributes of the product rather than user buying habits?

To learn a more in-depth understanding of the Pandora algorithm look here