Scorecards have long been used in business to provide easily understood measures of events and situations. A scorecard is literally a list of scores which can be summed to provide an overall rating. This rating is then used to inform a decision – whether to approve a loan or not for example. In fact credit rating is one of the more established uses for this technique and typical inputs might include age, salary, sex, marital status and so on. Each of these characteristics might then be scored and an overall rating generated, which is then used in a business rule to determine credit worthiness.
FICO has decades of experience in helping organizations build predictive scorecards, where historical data are used to determine how inputs should be scored and new data subsequently rated based on the patterns that have been found. Depending on industry it is often essential that these patterns are easily understood, and this means ‘black box’ techniques such as neural networks and support vector machines cannot be used in pattern creation. FICO has pioneered techniques which use sophisticated statistical techniques such as Generalized Additive Models and non-linear constrained programming. Some techniques from machine learning are also available, and specifically random trees and the application of bagging techniques (borrowing from ensemble learning methods).
All-in-all FICO has managed to navigate a difficult path between rule transparency and incorporation of contemporary techniques, and not surprisingly the technology is heavily used in finance applications where decisions must be transparent and easily interrogated.
The Scorecard Module is just one element in the FICO Model Builder platform.