Predicting user behavior still not an exact science

The benefits of predictive analytics boil down to the decisions of the people involved

A recent article in the Philadelphia Inquirer on the use of predictive analytics to determine which of Philadelphia’s parolees were likely to commit murder caught my attention. A broad definition of predictive analytics would be the process of matching statistics with historical data in order to predict future events, mainly human behavior.

The news out of Philly reminded me of the movie Minority Report, in which the government’s Pre-Crime Department could predict a crime before it was committed. In the movie, police are sent out to arrest the future perpetrator prior to the commission of the crime.

Philadelphia, the City of Brotherly Love, is reviewing the work done by University of Pennsylvania criminologist Richard Beck, a statistician who believes he can use data on parolees from the probation department to determine who is most likely to attempt murder.

Whatever the outcome in Philadelphia, this got me to thinking about the use of predictive analytics in more benign areas such as business and IT.­

CopperKey, a company that offers hosted software for predictive marketing campaigns, represents a new trend in predictive analytics. CopperKey CEO David Castillo calls it “leapfrogging,” in which the analysts with the Ph.D.s are bypassed and the tools are put in the hands of end-users.

CopperKey’s solution takes a black-box approach to predictive analytics. A company uploads its customer list, saying, in essence, “Here are the people responsible for putting bread on my table, find me more like these.” Within minutes, a list of potential customers similar to existing customers is spit out.

Josh Mellberg, president of Senior Advisors Wealth Management, says that the CopperKey black box increased the response to his company’s direct-mail by nearly 40 percent.

IT is also using predictive analytics to combine network-related data -- measurements you take off the network to gauge its activity and health -- with business information to improve the bottom line.

For example, data about customers dialing in to a call center to complain about poor cell phone coverage can be correlated with what the network equipment is doing and will do at a certain location, under specific weather conditions, at a particular time of year. By putting these sets of data together, the carrier can take practical steps to improve quality of service before the customer decides to try another vendor.

IT is also using analytics to understand and predict usage patterns across thousands of servers in order to ease capacity planning. With predictive analytics in place, IT can optimize its assets, notes Anne Milley, director of technology product marketing at SAS.

Of course, all is not rosy. What is still needed, Milley says, is a way to put analytics to work faster -- or, as Milley puts it, to reduce the difference between data latency and decision latency.

Part of the problem is that, although getting data out of systems is fairly straightforward, putting the new data back in can be tough. How do you feed the forecasting information back into your inventory replenishment and planning systems? In most cases, there are no standard APIs to make use of, which means more integration work for IT.

Milley also believes that although predictive analytics may support decision-making, there is no guarantee that it will lead to the best decisions.

“Somebody, somewhere has to wallow in the data and validate it,” Milley says.

And that is true whether it is a marketing campaign or a parolee’s future.

Mobile Security Insider: iOS vs. Android vs. BlackBerry vs. Windows Phone
Recommended
Join the discussion
Be the first to comment on this article. Our Commenting Policies