Eric Siegel, PhD, is the president of Prediction Impact, Inc., executive editor of the Predictive Analytics Times and founder of the Predictive Analytics World and Text Analytics World conferences. Formerly a professor of computer science at Columbia University, Siegel has published more than 20 papers and articles in data mining research and computer science education. His new book, Predictive Analytics, is addressed to a broader audience – the general public. He will be speaking at an imminent series of complimentary dinner events for marketers, sponsored by IBM: Business Analytics: Solutions for Results-Driven Marketing (coming to Chicago August 6, Boston August 8 and other cities in North America). He recently responded to a series of questions about his book and its impact on marketers.
Q: Your book, Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die, has been compared to Freakonomics and Moneyball – two best-selling books that made specialized fields interesting to non-technical readers. Your reaction?
A. I’m certainly pleased by the comparison. Indeed, my motivation was to take predictive analytics out of the academic realm and make it accessible to people who could benefit from it, both in their personal lives and on the job.
Q. You’ll be speaking to marketers at IBM’s “Results-Driven Marketing” dinner series. What benefits does predictive analytics offer marketers?
A. Predictive analytics is the most actionable form of data analysis. It predicts consumer behavior on an individual basis to drive, in detail, business decision-making for each individual -- for instance, whether to contact a customer or not, or which of several recommendations to make in a given situation. What could be more valuable than to see each individual's propensity to buy, or their likelihood to defect to a competitor?
Q. A recent IBM survey of CMOs worldwide* showed that two-thirds felt that return on marketing investment was going to be the primary measure of their effectiveness by 2015. If that’s so, how does predictive analytics contribute?
A. If you'd like proof of how predictive analytics improves that return, just look at the case studies! For example, my book provides 147 examples of predictive analytics at work – more than 20 of them in marketing. Take First Tennessee Bank, for instance, which cut direct mail costs by 20 percent yet increased response rates, for an ROI of 600 percent; or Target, which increased revenue by 15 to 30 percent. And you don’t need to be big to get that kind of boost: Vermont Country Store more precisely targeted their catalog mailings and, as a result, earned 11 times higher than the investment it took to apply predictive analytics in this way.
Q. What do you see as the hottest trends in predictive analytics today?
A. Two hot trends are uplift modeling and ensemble modeling. With uplift modeling, you’re predicting not what people will do but your ability to influence what they’ll do. That is, instead of predicting if a customer will respond, you're predicting whether they will respond if you contact them -- and only if you contact them. It's a subtle difference, but it can make a huge difference for certain business scenarios. The 2012 Obama presidential campaign used uplift modeling very effectively. There are many other case studies – from banking, cell phone companies, and other industries – that demonstrate the value of uplift modeling.
Ensemble modeling may be the most important predictive modeling advancement of the last decade. It brings the “wisdom of crowds” approach to predictive modeling. Ensemble modeling played a key role in the competition for the $1 million Netflix prize several years ago – which I describe in my book. Research shows that ensembles often boost a single model’s performance in the general range of 5 to 30 percent.
Q. What do you hope marketers will take away from your presentation, and your book?
A. The confidence to move ahead with predictive analytics. The knowledge marketers need is not on the rocket science level. They do need to know how to prepare their data for a modeling tool -- that's the most time-consuming, hands-on portion of the project. It’s substantial work, but well worth the great results that are achievable.
Q. Did your publisher use predictive analytics in marketing your book?
A. No, we didn’t have a large enough data set, since this was my first book for a general audience. But I did test titles for the book by placing Google AdWords. “Predictive Analytics” was clicked on about twice as often as, for example, “Geek Prophecies,” and beat out other options as well. So it became the title, by popular acclaim.
Register to hear Dr. Siegel at a complimentary IBM dinner event near you.
*From Stretched to Strengthened: Insights from the Global Chief Marketing Officer Study. IBM Institute for Business Value, 2011.