Understanding Predictive Analytics: A Spotlight Q&A with Eric Siegel, author of Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die

Originally published August 14, 2013

This BeyeNETWORK spotlight features Ron Powell's interview with Eric Siegel, founder of Predictive Analytics World and author of Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die.
Eric, you’ve been working in the field of predictive analytics for many years now. Can you tell us why you chose to focus on predictive analytics?

Eric Siegel: I am a former academic. I was a professor at Columbia University in computer science, where machine learning is housed. I am originally from the research and development academic side, and that was my interest in terms of core science and technology: the ability for a machine to learn from example. That is the core technology that drives predictive analytics because what it is learning is how to predict for each individual whether he or she will click, buy, lie or die, as in the title of my book, or lots of other types of consumer behavior – such as that of voters, healthcare patients and law enforcement suspects. It involves many types of individuals and their behaviors and how those predictions can help. Data is a record of things that have happened. It’s not just ones and zeroes. It is the experience of an organization, and the organization literally learns from that experience how to make per person predictions. For me, academically and intellectually, it was extremely interesting, but I decided that research and development wasn’t my focus. I wanted to go into commercial deployment, using predictive analytics and applying it in the real world, which I’ve been doing for 10 years.

You’ve recently written a fascinating book, Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die . Why did you feel it was time to write this book?

Eric Siegel: Well, a more limited version of this book could have come out a few years ago, but there's much in the book – almost half of it – that couldn't have at that time. So much has happened and that is illustrated by the sheer number of brand-name case studies. The book includes a central compendium, a color insert with 147 mini-case studies and cross-industry examples – and also covers certain advanced topics. The last chapter is on what's called persuasion modeling or uplift modeling that, for example, the Obama campaign used to win additional votes in swing states.  The timing did work well. In fact, in a way I can't believe that I was fortunate enough to release the very first book accessible to nontechnical readers focused squarely on predictive analytics. It's very much an introduction for newcomers. The book serves as an introductory textbook at some universities, but it's written in the more fun, pop science mode.

Who is the audience for this book? Are we talking about executives or business and technical people as well?

Eric Siegel: In fact, technical people are more often than not looking for a how-to book that’s very hands-on, technical and mathematical. There are not a lot of formulas in this book. This book is really accessible for any newcomer with any background. However, I do see that technologists are very interested because there are a lot of new case studies they haven’t seen before. And, there is treatment of advanced topics that even the most senior practitioners actually haven’t really looked at yet, including, as I just mentioned, persuasion modeling. There is a chapter on what is called an ensemble model – how to sort of amp up the predictive performance of a predictive model. Given that it is an industry overview, it does tend to have lots of different sub-audiences. There is also a chapter on ethics – privacy and civil liberty concerns that come up consistently with predictive analytics. I would venture to say that nobody is an expert on that topic. That chapter can be of interest to anybody.

Security and ethics are really hot topics right now.

Eric Siegel: Yes, there is a lot of attention. In general, there is a lot of concern about what data exists, what’s being captured and how it is being used. Predictive analytics ups the ante on the privacy challenges and concerns, at least potentially ups the ante, because it is a way to discern new data from data that may be considered innocuous or not so sensitive. It actually makes a prediction about you, such as whether you are pregnant as the retailer Target does, or whether you’re likely to quit your job as lots of large organizations, including Hewlett-Packard, do. And, law enforcement predicts whether you are going to commit a crime again, and that predictive computer directly informs decisions made by parole officers and sentencing judges. So literally, the predictive model influences how long you stay in prison.

We all like to think that we are individuals who live, learn and make purchase decisions in a unique way. Doesn’t that make it difficult to predict what we will do?

Eric Siegel: I think that when people feel that they’re individuals and that they’re unique and unpredictable, they are correct because we’re not predictable. The idea of predicting the weather is extremely difficult, and predicting human behavior is no different. However, what you’re doing in most uses of predictive analytics is you’re not relying on accuracy. You don’t need to predict accurately. You need to predict significantly better than guessing, and that’s what makes it valuable. So if there’s a needle in the haystack issue for law enforcement, for fraud, for customers who are going to be extremely valuable or for a rare disease in healthcare, what you’re doing is you’re making the haystack much smaller. Business is a numbers game, and you play that numbers game much more effectively by tipping the odds in your favor by saying, “This customer is three times more likely than average to be an extremely poor credit risk.” And if being a certain kind of credit risk is rare, well, this person still is unlikely to be that kind of credit risk, but he or she is more likely than average. And that can be very valuable to drive all the sort of micro decisions per individual in marketing, credit risk, fraud detection, etc., that organizations make en masse.

It seems that we can’t talk about anything in the realm of predictive analytics or business intelligence without including big data. What’s your take on big data and how does it fit into predictive analytics?

Eric Siegel: Big data, the buzzword, is more of a public relations endeavor than a technology. Like the broad terms analytics and business intelligence, big data is more about a mind-set, a subculture of people doing smart things with data. It doesn’t necessarily refer to specific endeavors or methodologies. And basically, big data doesn’t mean anything more than a lot of data. In fact, it’s a grammatically incorrect way to say “a lot of data.” It’s like saying “big water.” But it is very much a good thing to bring people’s attention to the potential of data. At the core of it, what makes data so potent and valuable is that it is experience from which to learn, and learning to predict in particular is what can guide the operations of organizations. It provides value by learning to predict. The predictions for each individual drive millions of operational decisions made by organizations every day – who to contact, call, mail, incarcerate, investigate, or set up on a date on a dating site.

Can you share a couple of specific examples of how companies are using predictive analytics?

Eric Siegel: It’s hard to pick because, as I mentioned, there are 147 mini-case studies included in the book. There are several case studies that are brought out in more detail, and we’re seeing so many use cases at the Predictive Analytics World Conference that, I should mention, is held several times a year. Chase Bank, for example, predicts which mortgage they’re at risk of losing. A mortgage holder can defect just like any other customer. Customers defect all the time. They cancel, leave, attrite, or churn or whatever term you want to use. In the case of a mortgage, they’re probably leaving by way of refinancing with a competing bank, and this can help drive decisions about who to reach out to and attempt to retain. In the case of a mortgage, it would be by proactively saying, “We’re going to lower your payments. We’re going to refinance with you if you’ll just sign this commitment.” That would be intended to intervene and decrease the chance of losing the customer. For other businesses like cell phone providers, it might be giving a customer a free cell phone or a discount. In the case of Chase Bank, there's a secondary usage that we know they look at. If they’re at risk of losing a customer from a marketing standpoint , they can reach out and try to keep the customer.  But there is another activity that banks do, which is to sell mortgages to one another. So these decisions can be driven by the value of the mortgage, which is informed by that prediction of whether that customer is going to leave. That example is  within marketing, but there are literally hundreds of other examples in industries such as healthcare, law enforcement, government and even non-profits.

Predictive analytics could really be everywhere, correct?

Eric Siegel: Yes, it is basically for any organization that has access to data from which to learn and that can make value of the predictions. It’s one thing to do the analysis. It’s another to actually use all these individual, hundreds of thousands or more individual predictions and to actually integrate them into operations. There needs to be a value proposition there. But where there is one, the number of possible ways to apply the technology is great, and we keep seeing new and surprising ways in which organizations do so.

Eric, there is so much information and so many examples in the book. What was your favorite part of the book to write.

Eric Siegel: Every chapter had a fun part to it for me. I am, of course, a technology person so there is a part of me that is just in love with the technology. I am a huge fan of what IBM’s Watson computer did in defeating the all-time human champions on the TV quiz show Jeopardy! And, it turns out that that is driven by predictive modeling, and it is such an amazing flagship success story that I devoted an entire chapter describing how it works. It literally predicts, for a given question and candidate answer, whether that is the correct answer. There are hundreds of candidate answers for each question. Looking at one candidate answer at a  time, the predictive model predicts whether this is the right answer to a given question and uses those predictions to rank all the candidate answers. At the core, it’s the same thing – learning from the history of decades of the TV quiz show Jeopardy! questions and their correct answers as well as some example incorrect answers to create that model.  That’s the second to the last chapter.

The last chapter on persuasion modeling, also known as uplift modeling, is something I’m a huge fan of. But, in general, I got a big kick out of finding new ways and new illustrations, analogies and case studies across all of these concepts and making the concepts of predictive analytics illustrative and accessible, even for complete newcomers.

That’s great. How can readers get the book?

Eric Siegel: We have a website for the book. It’s ThePredictionBook.com.
You can click over from there to Amazon, Barnes & Noble and all the other retailers. That website includes excerpts, reviews, a whole lot of press coverage, and some videos of yours truly being interviewed on the topic of predictive analytics.

Thank you Eric. It has been a pleasure talking with you today about your book, Predictive Analytics: The Power to Predict Who Will Click, Buy, Lie, or Die. I especially enjoyed reading the many predictive analytics use cases in your cross-industry compendium, and I certainly recommend this book for anyone who wants to know more about predictive analytics.

  • Ron PowellRon Powell
    Ron is an independent analyst, consultant and editorial expert with extensive knowledge and experience in business intelligence, big data, analytics and data warehousing. Currently president of Powell Interactive Media, which specializes in consulting and podcast services, he is also Executive Producer of The World Transformed Fast Forward Show. In 2004, Ron founded the BeyeNETWORK, which was acquired by Tech Target in 2010.  Prior to the founding of the BeyeNETWORK, Ron was cofounder, publisher and editorial director of DM Review (now Information Management). He maintains an expert channel and blog on the BeyeNETWORK and may be contacted by email at rpowell@powellinteractivemedia.com. 

    More articles and Ron's blog can be found in his BeyeNETWORK expert channel. Be sure to visit today!

Recent articles by Ron Powell



Want to post a comment? Login or become a member today!

Be the first to comment!