Predictive Analytics: Interview with Eric Siegel, Ph.D.

Seems like everyone’s trying to understand the future with respect to their customers. We see companies like Google, LinkedIn, and Facebook using predictive analytics to predict user user behavior. Even Stephen Few is giving predictive analytics air time, in classic Fewian cut-right-to-the-core style. So, when we recently had the opportunity to get some predictive analytics insights from one of the industry’s thought leaders, we just couldn’t pass it up.

Eric Siegel, Ph.D., is an expert in predictive analytics and data mining, and is the Conference Chair at the Predictive Analytics World 2010 conference. This is the premier predictive analytics conference and is the "business-focused event for predictive analytics professionals, managers and commercial practitioners." We asked Eric some questions about the trends he’s seeing in this field and wanted to share them with our community.

(Also, don’t miss out on the Predictive Analytics World discount code at the end.)

Juice: BI visualization has certainly started to become more mainstream in the past few years. Where is predictive analytics on this maturation/adoption curve?

Eric: Predictive analytics has crossed the chasm and hit mainstream in many sectors, such as credit scoring for financial institutions, response modeling for large direct mail houses, fraud detection, and others. And it is mature in that most large and many mid-tier businesses have employed it in one way or another, if only in a first-stage fashion. All industry verticals are replete with success stories.

Juice: Would you say predictive analytics is used more for understanding or for action?

Eric: I’ve always had the impression that predictive analytics is employed with action more the central objective than understanding, although understanding is usually also enjoyed, at least as a "side effect." A predictive model’s scores drive operational decisions for each customer - that’s the action for which it’s designed. But by taking a gander into the rules or patterns embedded in the predictive model, strategic insights are often also gained.

On the other hand, the results of the Predictive Analytics Survey put the two benefits as a near tie. This may be because, while fewer projects put insights ahead of action, those with action first also typically include insights as well (the pertinent survey question was a check-all-that-apply).

Juice: What are some of the best examples you’ve seen of predictive analytics applications that are designed for the "non-analysts"?

Eric: Well, there are two sorts of "action" that can be driven by predictive analytics: decision automation and decision support. In almost all cases of the latter, where staff "in the field" are provided additional information in order to make more informed their decisions - such as customer service agents providing cross-sell offers based in part on system recommendations, or consumer banking branch managers greeting their clients most at risk of churn - it is a non-analyst who "consumes" the predictive scores output by the analytics system.

Juice: How important is real-time to predictive analytics results and resulting actions?

Eric: This depends entirely on the application: what actions or decisions are being driven by predictive scores? So, no knowledge of analytics is required in order to answer this question. The good news is, when the predictive scores output by a predictive model are required in real-time - such as for selecting the optimal ad to serve to a user based on her profile and behavior - predictive models themselves operate quite quickly. They may involve sophisticated math, but they almost never have any iterative/repetitive "loops" in their programming, so they can turn a customer’s data into that customer’s predictive score very very quickly. It is the derivation of the predictive model in the first place, the application of predictive modeling over historical customer data, that may take hours or days, depending on the analytical method and analyst’s process employed; once you have the model, you are ready to fly.

Juice: How does scenario analysis fit into predictive analytics? What are some of the best practices around scenario analysis?

Eric: Predictive analytics generally works at a lower level than standard scenario analysis. It is doing such an analysis at the individual customer level, predicting the probability the customer will exhibit a certain behavior, such as a response, purchase, or defection. So, when considering a prospective predictive analytics initiative, its potential benefits could be put into a scenario analysis. For example, if predictive analytics is to be used to target a retention campaign, its target benefit of decreasing churn by, say, 10% more than current retention efforts could be plugged into a scenario analysis in order to calculate project ROI and gain further traction for the project.

For more information about predictive analytics, see the Predictive Analytics Guide

More information about the upcoming Predictive Analytics World Conference, Feb 16-17 in San Francisco.

(And finally, here’s the 15% off discount code for the upcoming conference: JUICE010.)