Eric Siegel, Ph.D., is president of Prediction Impact, a consultancy and conference organization that centers on the much hyped but still developing field of predictive analytics. A former machine learning professor at Columbia University, Siegel jokingly considers himself a “recovering academic” during his last nine years as an outside voice to enterprises looking for widespread implementation and training with their predictive models. Ahead of his keynote speech and chairman duties at the Predictive Analytics World conference in Toronto later this month, Information Management caught up with him on the lack of business-side readiness for analytics students, and the “paradigm shift” coming for predictive models.
Give me an idea of the changes in the use of predictive analytics you’ve seen since you starting leading these conferences a few years ago.
The idea of predicting influence rather than behavior is a paradigm shift. Normally what you’re doing with a predictive model is you’re scoring each individual customer. Are they likely to buy this product? Be admitted to the hospital? Commit an act of fraud? Whatever behavior the organization wants to predict. But really what organizations want to do is make change, have an impact and persuade. To do that, the more direct prediction goal isn’t what people are going to do, but the prediction of your persuasion on the customer. It’s analytically a much more challenging thing to take on and senior practitioners barely have this method on their radar, but there are some amazing case studies showing that if you’re getting a lot of value from predicting behavior and then you go to this method, you’re in comparison doing even better.
So, anecdotally, where do enterprises stand in terms of getting more business users in touch with predictive information?
In a weaker definition, we hit mainstream use [of predictive analytics] years ago, because all of the larger organizations use analytics in one way or another. All of the cell phone carriers can predict which customers are at risk for defection, for example. In terms of mainstream in that most businesspeople really know what PA is and the potential has been realized, I would say that the way things are going, we’re a few years away from the basics being general knowledge. But the usage is so open ended. There are so many corners of the organization that can tap into this potentially great boon of enacting prediction.
Marketing, sales and customer retention have some obvious implementations with predictive data. But is there another area in the enterprise, or specific industry, where you see an untapped benefit with a useful predictive model?
One of the biggest sources of potential comes from the kind of data that you are learning from. In terms of how you are applying the model for business benefit, the well-treaded paths are in marketing, in sales, in fraud detection - I don’t see those are going anywhere. But the effectiveness of these predictive projects and their returns are growing, from textual or unstructured data or social data … An enterprise is a large, complex thing, and if they want to learn from their data, they need to transfer that into a large, simple thing. That is, they collect all of these bits of information known about each person and make sure it has uniformity and regularity that they can get out of that for all of the internal systems they have.
Click here for a DM Radio discussion with Eric Siegel and others on the use of predictive technologies with data mining.
What is the difference you see in attaining an education in predictive analytics and what is needed in the field?
Universities are starting to introduce predictive analytics all over the place, mostly in their business schools, but there is a big distance between learning about core technologies and academically published learning versus learning at an enterprise-level initiative in the deployment of analytics. If you’re just a technology person and you’ve been trained on the core technology but not how to really use it, you haven’t been taught how to take a business perspective into account. This speaks to the major challenge with enterprise deployment: It’s one thing to make a predictive model over the data that’s cool and statistically significant, but to make a model that’s actually useful and will pass all the hurdles and be actionable within the framework of real-life enterprise constraints is completely different. That goes outside of core technology.
Register or login for access to this item and much more
All Information Management content is archived after seven days.
Community members receive:
- All recent and archived articles
- Conference offers and updates
- A full menu of enewsletter options
- Web seminars, white papers, ebooks
Already have an account? Log In
Don't have an account? Register for Free Unlimited Access