We are absolutely working with the iLog team because there's a lot of affinity between the two technologies. In order to make the best possible decisions, companies are combining rules engines, which interpret what you already know in if-then-else sorts of decisions, with predictive analytics, which explore what you don't know but need to interpret.
What's the place for the statisticians and data modelers in this new decision-management paradigm?
These skilled people will always be critically important to SPSS, but over the next few months and years, you'll see a greater focus on the business analysts who understand the business processes and who are responsible for things like customer retention and higher lifetime customer profitability.
Are we talking about embedded analytics here, whereby the decision support is integrated with applications or business processes? Or to ask it another way, who is the intended customer?
The entire business intelligence and performance management team is going to be selling and implementing analytics right alongside Cognos technology. In that sense we're moving up the value chain in what would otherwise be a conventional BI sale. But I think SPSS will also start to get deployed through other routes. If someone is doing business process management or optimization with iLog technology, for example, there's a natural affinity with SPSS. We're also working with the InfoSphere data warehouse team because SPSS is used in data mining. One of the key missions that I have is to try to reach out to different quarters within IBM -- whether it's Business Services, Cognos, InfoSpere or iLog -- so that whenever they are working with customers, they are aware of the affinities with SPSS and predictive modeling.
You didn't say anything about applications in that rundown.
From an applications point of view we also have analytic applications for financial performance management, and there's another logical connection there.
Will this decision-management engine integrate with applications from the likes of Oracle and SAP?
You could make decisions just in the context of information available from ERP or CRM systems, but IBM believes relevant data resides in a lot of different places. Social networking and Web 2.0-style applications, for example, are leaving tons of clues behind as to what people are thinking. You can also do surveys to develop attitudinal data. In the consumer package goods space they do focus groups and scientific research on emotional reactions to products. IBM can pull data together from a lot of sources -- structured and unstructured, within the enterprise and outside the enterprise -- to make the model even more accurate. That offers advantages that are greater than what you can learn by focusing on data within a particular application.
Can you share an example in which SPSS predictive analytics are used within a mainstream business application or process?
At Infinity Insurance they needed to process claims in a timely way, yet they also have to watch out for fraud, which is something that costs the industry tens of billions of dollars each year. The company built a predictive model, based on past history, to profile which claims turned out to be fraudulent and which ones turned out to be okay. They didn't just want to raise red flags on claims that might be fraudulent, they also wanted to know which claims were least likely to be fraudulent so they could fast-track them with no need to investigate. Not only does that save expense, it also improves customer satisfaction. And for those claims that are flagged, you're kicking off an investigation right away, so you're resolving the case more quickly. It used to take Infinity 14 days to complete a claims investigation, but that has been knocked down to 24 hours.
To what extent can predictive models like that be prebuilt for industry-specific applications?
Well, you can't just take a model that you built for Infinity and plop it into State Farm. You have to look at the customer's data and then build multiple models using multiple statistical techniques. You also need an ongoing feedback loop to ensure the accuracy of your model. That means you keep track of accuracy using champion and challenger models. If the champion is becoming less accurate over time, it's time to switch over to one of the challenger models that might be more accurate.
It sounds like you might need PhD-level analytics experts behind the scenes continuously assessing and refining the models.
To the contrary, business analysts would very much be able to handle the kinds of modeling and analysis I'm talking about. They would need to have an analytic bent, but the tools are designed for business users.
Your description also doesn't make it sound like SPSS will be going after prebuilt models and applications for specific industries. Is that the case?
No, we are headed in that direction, and we believe that you can have accelerators and prebuilt models that can significantly reduce the time to value. I see models as something that companies will view as the source of their competitive advantage. There are certain things that are common to all companies within a specific industry, but the company-specific insights that you can add to a model can be a differentiator.