Much to my chagrin, I had to return to wintry Chicago late Wednesday night from a delightful visit to sunny and 60ish San Francisco for Predictive Analytics World, February 2010. And, like last year, I was pleasantly surprised by the overall quality of the meeting. But whereas then the conference was version one with low expectations, my fear this year was for simply a repetition of last. Not to worry: the several hundred participants were treated to a splendid show. Kudos again to conference chair Eric Siegel and producers Prediction Impact, Inc. and Rising Media Ltd.
Siegel's keynote: Eureka! Seven Innovative Applications of Predictive Analytics, set the tone that this conference would be different, focusing on applications like text mining, online adds quality, call center optimization, and reliability modeling rather than last year's risk scoring, retention and fraud detection. The business thinking driving the work is similar, however. Decide what to predict and then how to use the prediction to optimize a business process. Siegel confided that text mining and social networks would be key recurring conference themes.
I'm usually skeptical of sponsoring software vendor presentations, expecting at least a soft sell of capabilities and products. Not so with IBM/SPSS Erik Brethenoux's Beyond the Four Walls of your Enterprise: What your Customers Tell You. Brethenoux offered compelling anecdotes of the power of social networking, citing also IBM Research's Social Media Analytics that evaluates blog buzz factors and sentiments. IBM's Smarter Planet Manifesto: 1) Instrument the world's systems; 2) Interconnect them; 3) Make them intelligent.
James Taylor corroborated IBM's obsession with analytic business optimization in his case study talk: IBM Analytic Journeys. In research sponsored by IBM, Taylor chronicled successful “smart” customers in various industries, looking for commonalities of success. Among his initial findings are that such companies obsess early on the “End Game” – pervasive, actionable and predictive decision opportunities, as well as “The Journey”, which includes the analytics platform and delivery mechanisms. As Taylor outlined the success profile, I couldn't help but feel deja vu to a Harvard Business Review webinar I recently participated in entitled Analytics at Work, by Tom Davenport, that came to very similar conclusions. Taylor acknowledged discovering the similarity with the Davenport methodology a bit later. That validation, though, is quite comforting. Indeed, John Elders Text Mining: Lessons Learned, outlined yet a third similar methodological recipe. The overall case control design would probably be strengthened if a “control group” of analytics failures were juxtaposed to the success stories.
Delightful Andreas Weigend, former Chief Scientist of Amazon, entertained the audience with his Predictive Power Part II: Advanced Analytics in the New Data Economy, continuing his theme of the enveloping social data revolution introduced last year. Weigend notes the data world has changed significantly in just one year, with Facebook now at over 100M users entering 15B pieces of data per month. For Weigend, it's now C to C business, an evolution from eBusiness to meBusiness, to weBusiness. He showed a YouTube clip of an iPhone GPS app in practice that links potential dating prospects real time. If only I'd had that capability 30 years ago!
Sponsors Neteeza and SAS spoke of a partnership where big data meets big math in their Next Generation Insight presentation. Their joint take is that the current separation of database and analytics platforms is inefficient for companies with large data that compete on analytics. The answer: In-Database Analytics that embed statistical engines into the database appliance, exploiting limitless massively parallel processing capabilities. As a statistician that likes to play in a sandbox, I'm not sure I agree with the vendors.
Neil Raden sort of agreed with them in his In-Database vs In-Cloud Analytics: Implications for Development. Ever the skeptic, Raden argues that success in analytics doesn't augur for success in business, speciously citing the poor short-term stock performance of Competing on Analytics' Analytical Competitors. Raden's not generally a big fan of Cloud Analytics, arguing the economics don't fly, offering a financial analysis where cloud deployment is cost-prohibitive even at year one. That may be true for companies with huge data, though OpenBI's experience with much smaller cloud BI deployments suggests automobile lease-like thinking: the longer the projected life of the program, the better purchase looks in contrast to lease.
I'd been warned that the conference site, the Palace, an historic landmark San Francisco hotel, was not the most tech friendly. And indeed I was in AT&T zero-bar, 3G hell for two days, having to skip out of the beautiful meeting rooms repeatedly to handle important OpenBI correspondence. Finding an outlet to recharge my thirsty iPhone was no less challenging. And producers, how about adding fruit, cereal and granola for breakfast next time, like last year? Not all predictive modelers are fueled by pastry at 8 AM!
Predictive Analytics World – Take 2, highlights additional presentations I found enlightening. Click here to read.