Predictive analytics will soon power many decisions, whether we know it or not

Can machines and systems predict the future for us?  They are already being employed to make inferences about the immediate future, especially when it comes to gauging the profitability or riskiness of new customers, the viability of new ventures, or the operational life of equipment. As reported on this site a few months back, predictive analytics is even helping to spot traffic bottlenecks before they happen.

With predictive analytics, software is employed to look at patterns in data to infer future events, from fraud to market changes to supply chain disruptions. I recently had the opportunity to speak with Jason Verlen, director of SPSS Predictive Analytics Product Strategy at IBM, about the possibilities -- as well as potential risks -- in relying on predictive analytics.

Q: Can you define “predictive analytics” for us?

Jason Verlen: Very simply, predictive analytics is looking at data that describes past events in various ways. It could be data about transactions, things that people have bought. It could be data about outreaches, whether or not somebody responded or did not respond to a marketing campaign.  It could be data about peoples' attitudes and opinions in the past. So taking all of those kinds of data, putting them together, and then making predictions about what people will do.

Q: Years ago, I saw a demonstration of predictive analytics as a way to monitor supply chain data and warn managers of potential disruptions, employing pattern matching of past bottlenecks against current data. Can it be applied to non-people situations?

Verlen: You can do predictive analytics in other domains as well. You can do it, for example, to measure risk, or to predict fraud in the future or many other such applications.

Q: Predictive analytics has been around for at least a decade. What is new, and why are more companies embracing it?

Verlen: The fundamental basis of the technology is statistical analysis that has been around for many decades – things like survival analysis, segmentation analysis, classification. But what has changed a lot is a focus not just on the techniques, but what to do with the output, how to deploy it into operational systems to directly improve and optimize outcomes.

Q:  Aren't predictive analytic-based solutions complex to implement, requiring a PhD in statistics or quantitative analysis?

Verlen: Most of the people who used our products in the past have had Ph.Ds in statistics, and really understand the depths of all these algorithms and inputs and the kinds of data that they surface. But there s a newer group of people who want to use the technology but don't have the skillset of these previous users in the earlier waves. What we’ve tried to do is allow these techniques to be used by a more business-oriented user than in the past, which has made it much simpler, and explains why predictive analytics has really taken off in recent years. It's just much more accessible to a wider group of people.

Q: Can you describe applications that would benefit from predictive analytics?

Verlen: We basically see three key pillars where this technology is applied: consumers., risk, and operations.  With consumers, predictive analytics can be applied to gain understanding the behavior of specific people to sell them things  that they'll really get value from.

In risk and fraud, if you’re a financial institution, and your offering financial products to customers such as mortgages, you can understand the probability that a mortgage will be paid back. Or, lets say you’re an insurance company. Some small percentage of claims are fraud. You use technology to figure out which ones are fraud, so you can use your investigative resources more efficiently and take a closer look at those.

In operations, let's say you’re dealing with a big machine or a tractor. The cost of a breakdown of those kinds of machines that catch you unaware is huge. There s downtime opportunity costs. There’s the time and expense to ship those kind of parts just in time to repair it. Those are not the kind of parts that you can just put in a UPS truck. What you want to do is predict when things could happen, and act proactively to prevent that kind of negative scenario.

Q: Could better employment of predictive analytic solutions at banks and financial service firms have staved off the recent subprime mortgage crisis?

Verlen: I don't want to oversell that. We now know some of those financial institutions were running on a debt to equity ratio of nearly 50 to 1. In my opinion, there is no analytical technique now or ever that will be able to stave off disaster if that's how your organization is run in terms of a debt level. Now, where the technology can be applied, and could have been applied better, is an understanding of individual cases – who is likely to be able to pay back a mortgage, or not pay back a mortgage. That is a very bona fide use of this technology that can be extremely successful.

Q: Does the rise of Big Data – enterprise information stored in the hundreds of terabytes, if not petabytes range – present new opportunities for predictive analytics?

Verlen: It's really a great new opportunity for the technology. When you’re dealing with data of that size and complexity, there is no way to manually extract value from it. You need to apply these kinds of advanced techniques across it, to lead you to the discoveries that you can make that actually help your organization perform. Look at IBM's Watson technology. All the data that has to be mined in real time for an advanced application like that to be able to handle ad-hoc questions. And there is so much data available to really get to better outcomes. Watson is an excellent example of predictive analytics.

Q: Is it likely that a lot of predictive analytics will run behind the scenes in applications, so business end users are unaware of the computations taking place?

Verlen: You are already seeing a vastly increasing number of applications, where predictive analytics is in fact being deployed, and making recommendations, but the people using the system don't even know that predictive analytics is the engine and the smarts behind it. In call centers, we have applications running behind the scenes that allow the rep on the phone dealing with the customer to offer a very good course of action for the customer. What shows up on the screen to the person on the phone is simply what the recommendation is to offer the customer, generated by the analytical technology running behind the scenes.

Q: So predictive analytics may eventually be running everywhere, whether we know it or not.

Verlen: There is absolutely no question that predictive analytics will be pervasive across a wide swath of all of applications. It will be everywhere. You will see predictive analytics running behind the scenes getting RFID tags, in call centers, and in many other cases. Predictive analytics is a new way of thinking that moves people from the old way of reacting to events, and moves them to giving them the information that they need in real time to act and improve the output of their organizations.

(Illustration Credit: US International Trade Administration.)

This post was originally published on Smartplanet.com