ILOG today announced a Scorecard modeler as an add-on for ILOG JRules ® (which I first saw at DIALOG). As their press release says, this add-in allows customers to “incorporate statistical scorecard models directly into decision services” – a key tenet of enterprise decision mangement or EDM. ILOG is targeting financial institutions clearly but apparently also sees “a growing market for scorecard usage in other industries” as do we.
The Scorecard Modeler is an add-on to ILOG Rule Studio designed to allow users (typically business analysts) to directly create and manage predictive or additive scorecards. Scorecards are a well established way of implementing predictive analytic models for predicting risk and or customer behavior and integrating them into the rule studio allows rules and models to be used together.
The implementation in the JRules studio creates a new rules artifact that allows a business analyst or technical user to take a scoring model (built by a statistician using SAS or SPSS or something similar) and implement it on top of ILOG’s Business Action Language.
The scorecard allows scores to be assigned (scores can be formulas) based on attribute values and weighting to be specified. The scorecard allows a user to manage reason codes – they can specify a list of reason codes (name, code, description, priority) and then allocate them to scores. In addition it allows multiple reason codes and has some out of the box reasoning strategies to, for instance, handle ordering of codes, handle duplicates etc. Both the score returned and the reason codes can then be manipulated in rules subsequently. Overall most of the features you would want, including some nice nested scorecard support, is there.
There is currently no PMML import (something offered by Fair Isaac, Zementis and Pegasystems) but manual recreation of a scorecard is not a hugely time consuming step. In addition their point and click formula builder makes it easy to build calculated characteristics, a key issue in implementing scorecards. An early stage customer claims to have reduced total development time to about 1/3 of the previous time. The scorecard modeler is not currently available in their business user web interface.
While the scorecard is managed as a single artifact for versioning etc, logging and testing are done at the cell level (a cell in a scorecard typically represents a single rule).
ILOG made the observation that, even outside financial services, their customers are starting to understand the value of analytics in conjunction with rules and are excited about having scorecards where before they might have made do with rules and decision tables. As is often the case with business rules, the improved communication between groups that comes with a declarative metaphor is key.
Overall it seemed like a well thought out implementation, as you would expect with a development partner like Equifax who use analytics so extensively. As a firm believer in the use of rules and analytics together I am delighted to see another rules vendor add support for analytics, especially at a pretty sophisticated level. I am looking forward to seeing some PMML management in the future as well as some thought as to how to present analytic models to business users as part of their rules management environment.