The two kinds of predictive analytics

I saw this article in InformationWeek today - Businesses Mine Data To Predict What Happens Next - and it made me realize that people can be confused as to what predictive analytics means in the context of decision management. In fact it made it clear that there are two uses of the term predictive analytics.

Firstly there is the use of predictive analytics as a kind of analysis done with BI-like tools. This is typically offline and used to inform a knowledge worker.  Essentially this is building a report not on what has happened but on what will probably happen in the future. Many of the examples in the story are like this. Clearly this is an improvement over the usual approach of using reporting and visualization tools to simply understand the past. However, it still assumes a relatively low volume of decisions and that a person with some analytic skills is the best consumer of the prediction.

Second there is the use of predictive analytics as a way to make operational systems "smarter". This is typically both offline - the creation of the models is done offline - and in line - the execution of the models is done during a transaction. The references in the article to how the financial services industry uses predictive analytics mostly refer to this kind of analytic. For instance, fraud detection involves neural networks for predicting the likelihood of fraud. The attraction of this use of predictive analytics is that it applies the "smarts" to every transaction as it happens even if the person involved is not analytically sophisticated (think call center representative) or even if there is no person involved at all (think website or ATM).

When I talk about decision management using predictive analytics and business rules it is largely in this second context. By applying predictive analytics to actual transactions you can get a great return.