If the summarizers have entered a new case, it is extracted from the database, the appropriate FLRE is instantiated and the case is evaluated.
The FLRE was designed and developed by four engineers over a period of six months.
A new program was written to query the database automatically for new cases at a timed interval and to instantiate appropriate instances of the FLRE as required.
The FLRE was developed using a design for six sigma (DFSS) approach including optimization based on evolutionary algorithms.
Regular underwriter auditing of the FLRE is also critical to ensure that the engine is correctly classifying policies over time.
When we deployed the FLRE for term life insurance underwriting, we created an offline quality assurance (QA) process to support the auditing process.
This includes the data entered into the engine, the individual rules that are fired for each instance of the FLRE, and each FLRE decision.
If a change is made to the underwriting guidelines, the maintenance team can also deploy changes to the FLRE between generations.
Perhaps one of the most interesting aspects of the FLRE design and maintenance was its error-cost-based derivation.
It also had an additional benefit: as the decision complexity increased, new capabilities of the FLRE had to be added, and a deployment process created.
At this point, we can use the same EA-based optimization tools employed during the initial tuning to find a parametric configuration that structures the FLRE to better approximate the new SRD.
In the first two generations, if an application was sent to an underwriter, he or she had to start on the application from scratch with no visibility into what the FLREs had suggested.