As part of business analytics, the Predictive Modeling process is a search for explanatory values. It starts with the identification of a problem that needs to be solved: real estate appraisers want to predict selling price; executives want to predict sales; life insurance carriers want to predict mortality. Uncovering relevant relationships between the variable you are interested in predicting, and independent variables that influence its result, starts with statistical analysis. A mathematical model is then developed using a modeling technique like regression analysis that defines the relationship between the dependent variable and many independent (explanatory) variables. This can then be used to predict the value of the dependent variable based on changes in the independent variables.
Modeling has seen expanded use outside of the domain of actuarial departments. Underwriting departments and industry laboratories have also used modeling to help improve underwriting decisions and improve efficiency.
Modeling and “triage”
Emergency rooms triage incoming patients by sorting out treatment priority based on severity of health condition. In underwriting departments this means figuring out which applications need closer attention and which may be spared additional requirements and may move more quickly to approval and issue. Finding ways to do this with reasonable accuracy requires the analysis of data from past experience and the application of what has been learned to future underwriting. Modeling can help sort this out. Doing this well ensures better placement of resources and better efficiency.
Modeling and quick issue
Data analysis and modeling techniques have helped underwriting departments develop quick issue programs that avoid APS and/or labs altogether. Using business rules and statistical information gathered through decision science methods like regression analysis, carriers have been able to issue some policies without traditional underwriting. The rules will define what can be allowed in the program and what cannot. The application is then scored based on numerous data points (“influencers”) and may be issued without ever going through the traditional underwriting channel. Better risks get fewer touches; time and money are saved.
Modeling and lab scoring
Insurance laboratories have used modeling techniques to study the impact on mortality of various labs and lab combinations. Learning what individual lab tests, or combination of lab tests, may impact mortality can help improve pricing. These previously unidentified “influencers” of mortality can now be modeled and used in underwriting. Laboratories have done fluid testing for decades and have delved into this data for insight. With years of lab tests and information on mortality from the Social Security Death Master File, laboratories have been able to identify statistically significant relationships between lab tests and all-cause mortality and thereby create a scoring methodology that can be used to better assess risk during underwriting. The individual lab scores are then summed and used in the risk assessment, similar to the way credit scores have been used in the mortgage business. An approval that may have otherwise been Preferred Best may now only be Standard Plus due to the final lab score. Traditionally, one could expect intuitively meaningful explanations of underwriting assessments, which is important in order to sell the value in the approval to the end consumer. The lab scoring process, however, challenges this expectation, and this is one downside.
Carriers must underwrite the person. The best thing an independent agent can do is collect information on their client that lowers the perceived risk of below average longevity. Any characteristic, activity or fact that indicates a healthy lifestyle, or implies the likelihood of good, or unusually good longevity should be added to the file. This can be done either by cover letter or in the remarks section on an application. “Bring yourself to life on paper and sell yourself as a good risk,” should be the mantra.