Next up were Carole-Ann Matignon and Andy Flint presenting on how analytic models and business rules can and should be combined. Carole-Ann and Andy are product managers for Model Builder and Blaze Advisor, Fair Isaac's products for analytic model development and deployment and for business rules management respectively.
Carole-Ann and Andy covered the value of agility in decision management, introduced analytic models and business rules and then showed how the two tools are used to manage and deploy models and rules. There are many reasons why models and rules might change (including changes in predictive characteristics of data, changes in data and changes in policies). The time it takes to get to a new model, deployed in production systems represents costs in terms of development costs (to build the model) and, more significantly, the opportunity cost of leaving an older, less efficient model in production. Many organizations take months to get new models into production and struggle to update policies that rely on those models. Part of Enterprise Decision Management, EDM, is to focus both on precision and agility. It is not enough to have a great predictive model, one must also have the ability to update them quickly and cheaply so as to retain agility.
They introduced business rules, predictive analytics and their combination (called a strategy by Fair Isaac) - key components of an EDM approach. Business rules can come from experts, legacy code or from data mining while there are a wide range of analytic techniques that can be used to produce predictive models. The big idea here is that business owners, and analytic modelers, should be able to drive the changes they want into the operational systems that run the business without waiting for IT to code the change. This requires rule management and integration of rules and analytics. To this end Blaze Advisor supports representations of decision trees, scorecards and other analytic structures as business rules. Indeed, as they point out, this is how Fair Isaac builds its EDM applications for originations, fraud, customer management and so on.
They got pretty specific, showing how models from Model Builder could be deployed as "black boxes" in Blaze Advisor and then demonstrating new functionality allowing models to be loaded into Blaze Advisor as "white boxes" using PMML. Not only does this allow the models to be managed as rules in the Blaze Advisor repository, it also means that the models can be deployed to all the platforms supported by Blaze Advisor (.NET, Java, COBOL). Here's a slideshow of some of their slides. The slides discuss:
- The lifecycle of change in rules and models
- The cost of delay in the implementation of models
- Turning analytic insight into value by embedding analytics into rules-based decision services
- How this works in Fair Isaac's Model Builder and Blaze Advisor
