Highlights from my fifth Predictive Analytics World
Oct 21, 2011 12:04 PM
Having been an early supporter of PAW, I am happy to see this conference flourish. It was a pleasure to see so many friends and colleagues who are such an enthusiastic and loyal part of this community of hands-on practitioners. Eric Siegel, Program Chair, kicked it off with the hot topic of net-lift modeling (aka up-lift modeling, true-lift modeling, differential response modeling…). I hope we will see greater adoption of more sophisticated experimental design to help answer this question: Which half of my marketing budget is effective versus not, or which of my customers would've responded/bought anyway? It's good that he is pressing us to think more about how practical and valuable it is to provide better answers to these questions.
John Sall's keynote on Speeding Up Discovery was especially relevant and visually appealing:
Faster, easier, richer graphics tell you more things letting you interactively dig deeper.
With smart automation/design, you can minimize the drudgery and be "in the flow" of solving the problem.
Many positive comments on John's keynote followed throughout, including a few mentions in a very interesting and entertaining talk at the Predictive Analytics Meetup group Wednesday evening by Professor John W. Emerson of Yale: "Bayesians, Frequentists, and Big Data: Musings on Statistics in the 21st Century." He made a great point that statistics is perceived as hard because it's poorly taught (by many) — not everyone needs the heavy artillery of mathematics, and that statistics is easier.
I had the pleasure of hearing Dean Abbot twice — once for the pre-conference workshop and once with Bill Seer on some interesting survey work they've done for the YMCA. Manu Sharm, Principal Research Scientist at LinkedIn, shared some very interesting perspectives and activities at LinkedIn. His view on what makes a data scientist: Curiosity + intuition + data gathering + standardization + statistics + modeling + visualization. To that, I would add creativity. Another important point he made that also came up in a few other talks: "You need visualization to influence decision makers; without it, you are just talking to other data scientists."
Though I didn't get to hear Matt Flynn of The Travelers present on JMP, SAS and R, I had seen him do a fine job at the JMP Discovery Summit last month — very cool to be able to take advantage of the best of such powerful paradigms. Don Cozine of Analytici shared a fresh approach on market mix modeling — an econometric approach (VARMAX) coupled with simulation. You can always tell when a really technical topic is well received by how long the speaker is kept after his talk, which in this case, was a good while.
Antonia de Medinaceli's talk on fighting fraud in the US Postal Service made pretty clear the challenges and complexity in dealing with so many potential types of fraud in such a huge organization.
John Elder's special plenary session drew several laughs while making some important points. "To have maximum impact, you can't just carry the ball across the goal line — you have to carry it up into the stands." He made the point that we have to have the perspective of attaining business success, not just technical success. The ability to live in both research and business is like being a shore creature living in both worlds of water and air. Many amazing things happen where two worlds collide.
Tom Davenport, describing himself as an "analytics motivational speaker," gave a very well-received keynote on Everyday Analytics. Favorite quote: "It's about the relationships (people)," attributed to Carl Kempf, Chief Mathematician at Intel.
Sadly, Usama Fayyad, who was scheduled to give a special plenary on futures in targeting and online marketing was unable to make it — we missed him and wish him well.
The only complaint, which is really a compliment to Rising Media and Eric Siegel, is that it's impossible to attend all the talks and hard choices have to be made. Thankfully, some of these speakers will also be presenting at SAS' Analytics 2011 conference in Orlando next week (though not necessarily on the same topics), among them are Matt Flynn of The Travelers, Don Cozine of Analytici and Anthony Goldbloom, CEO of Kaggle.
Many good takeaways from PAW about speeding up discovery and attaining faster value. I'm looking forward to more at Analytics 2011!