I just got back from the IBM Information on Demand (IOD) conference in Las Vegas.  A key message was that the future is in analytics and predictive analytics at that.  IBM has already invested $12B ($8B acquisitions, $4B organic growth) in analytics since 2005.  Its recent purchase of SPSS has enabled the company to put a stake in the ground regarding leading the analytics charge.

Predictive analytics uses historical data to try to predict what might happen in the future.  There are different technologies that can help you to do this including data mining and statistical modeling.  For example, a wireless telecommunications company might try to predict churn by analyzing the historical data associated with customers who disconnected the service vs. those that did not.  Attributes that might serve as predictors include dropped calls, calling volume (in network, out of network), demographic information, and so on.  An insurance company might try to predict future fraud using past claims that where the outcome is known.  Adam Gartenberg’s blog describes more examples of this.  IBM plans to make predictive analytics more pervasive in several ways. 

• Making models easier to build. It will make predictive modeling tools easier to use for those who build the models.  A good example of this is the SPSS PASW Modeler product that uses a visual paradigm to build various kinds of models.  I stopped by the SPSS booth at the show and saw the software at the demo area and it is nice with lots of feature/functionality built into it.  Training is available (and I would argue necessary), for example, to understand when you might want to use a certain kind of model. 
• Embedding the predictive model in a process.  Here, the predictive model would become part of a business process. For example, a predictive model might be built into a claims analysis process.  The model determines the characteristics and predictors of claims that might be classified as fraudulent.  As the claims come through the process, those that are suspicious, based on the model, would get kicked out for further examination.  

So, given these two approaches, can predictive analytics become pervasive? 

In the case of making predictive modeling tools easier to use, the question isn’t whether someone can use a tool, but whether he or she can use it correctly.   The goal of a tool like PASW is to enable business users to build advanced models. Could a BI power user who is accustomed to slicing and dicing and shaking and baking data effectively use a tool like this?  Possibly, if they have the right thought process and they pay attention to the part of the training that describes what type of technique to use for what type of problem.  It is a good goal.  Time will be the judge.

As for embedding predictive analytics in business processes; this is already starting to happen and here is where the possibility of making prediction more pervasive gets exciting.  For example, telecommunications companies can embed predictive analytics into a call center application to understand an action that a customer might take.  A call center representative can make use of the results of the model (without understanding the model or what it does).  He or she is simply fed information, from the model, (in real time) to help service a customer most effectively.   The model can be created by a skilled analytics person, but deployed in such a way that it can help a lot of other people across an organization.  One key will be the ability to integrate a model into the actual code and culture behind a business process.

Look, I don’t have a crystal ball (little predictive modeling humor there), but I am very excited about the possibilities of predictive modeling.  I did this kind of modeling for years at Bell Laboratories, way back when, and it is great to see it finally gaining traction in the marketplace.  Predictive analytics can be a truly powerful weapon in the right hands.