Features

Gregory PS: Your next move was to Blue Martini (now Escalate Retail). What did you learn there ?

Ronny Kohavi:
At Blue Martini we built an e-commerce platform with tightly integrated data mining. We described the architecture at http://ai.stanford.edu/~ronnyk/integratingEcom.pdf, and it was novel and I believe ahead of anything that existed, and probably ahead of most systems today. Most web platforms are built for content and commerce, and data mining is a later bolt-on, where data comes from weblogs and javascript tags. At Blue Martini, we built data mining concepts into the core, so that data was collected reliably and then transformed into a classical warehouse star-schema structure automatically. The data for the KDD Cup in 2000 came from a Blue Martini customer (http://www.ecn.purdue.edu/KDDCUP/).

Many of the lessons and challenges we learned at the time are summarized at http://ai.stanford.edu/~ronnyk/lessonsInDM.pdf. It is at this time that I began to appreciate the cultural and organizational challenges, and the fact that data mining is the vitamin for healthier living and only well run sites get to the point of improving their offerings through mining; most sites had a hard time keeping the site up and running, and were focusing on fighting fires, so they needed more aspirins to handle their headaches, not more vitamins. Jim Sterne at emetrics 2007 in DC (http://www.emetrics.org/2007/washingtondc/agenda.php) recently reiterated that point for the web analytics space.