Just reading the words “predictive analytics” can make executives break out in a cold sweat. Unless you have an actuarial background or are a Mensa-level data modeler, the concept of predictive analytics can be extremely intimidating.

While the P&C industry was an early adopter of predictive analytics—using credit scoring to predict auto claims rates—their life counterparts have been slower to embrace the opportunity. Yet one could argue that life insurance companies were among the first to rely on predictive analytics. The first actuaries began collecting and analyzing data that produced remarkably accurate estimates of life expectancy. Combining these aggregate mortality tables with underwriting techniques that assess the individual risk has produced a reliable risk selection process. Although this process is costly and time-consuming, the industry’s reliance on medical underwriting is the biggest impediment to the adoption of predictive analytics.

Predictive analytics offers hope for the future. As companies compete for producers, a simplified, streamlined underwriting process will be a differentiator in the marketplace. Enhancing underwriting efficiency with predictive modeling will help insurers generate and place more applications. In addition, underwriting resources will be better used as more routine work is completed by the model. Initiating a predictive analytic project requires strategic thought.

Target → Data → Model → Refine

  • Target (Start with focus): Many organizations initiate an analytic endeavor by gathering all available data before beginning their analysis. This often leads to an overemphasis on collecting, cleaning, and converting data rather than understanding its potential uses. Companies embarking on predictive modeling should first define the insights and questions needed to meet the key business objective. Only then can they adequately identify the data needed.
  • Data: By defining the desired insights first, companies can focus on specific subject areas and use readily available data in the initial models. Studying the insights delivered through the initial models will identify gaps in the data infrastructure and business processes. Efforts can then be turned toward collecting the necessary data and making process improvements identified by the insights, thereby improving the model with each iteration.
  • Model and Refine: The predictive analytic model measures the quantitative effect of multiple simultaneous variables and associated outcomes in order to identify strong correlations. The best predictive models depend on rich sets of data from which factor variables can be mined. The model can then be built, refined, and fitted. Models may range from simple linear regressions to advanced techniques, such as decision trees, neural networks, generalized linear models, and generalized additive models. The deployment of predictive analytics is both art and science because the models must be chosen to get the best fit of data and factor variables to produce a strong predictive target outcome. The most common model variables include:
    • Application data – any piece of data submitted by an applicant. The easiest data to work with is in a format such as multiple choice, yes/no, or numerical.
    • MIB – Virtually all insurance companies belong to the Medical Information Bureau and screen all applications for MIB data that provides details on insurance applications submitted to other companies.
    • MVR – The Motor Vehicle Report provides details of an applicant’s driving history.
    • Third-party marketing data – Several firms have started collecting consumer consumption data. The most relevant example for life underwriting is the prescription databases. Due to the protective value of this data, it is becoming a standard underwriting requirement.
    • Lab tests – Traditional underwriting requirements (blood, urine, EKGs, etc.) provide significant data for risk evaluation but are costly to incorporate into predictive models. The ultimate solution will most likely involve developing a scoring algorithm to drive the correct decision criteria.

Once a model is developed, it must be validated and calibrated. In the life underwriting environment, this is generally done by comparing the modeled decisions with underwriting decisions for cases that went through full medical underwriting. Therefore, we can state that predictive analytics is not a replacement for underwriters. Underwriters are indispensable contributors to the model—development through refinement.

An effective way to use predictive analytics is to run all applications through the model to get a “risk score.” Low scores (high face amounts, older ages, MIB codes, etc.) will be routed for full underwriting, with targeted requirements identified. Moderate scores may be routed to a junior underwriter or senior case manager for validation (i.e., labs are clear; otherwise, the case is routed for full underwriting). High scores are jet-issued without human intervention.

The goal of predictive analytics in life underwriting is to reduce cycle time and underwriting costs. Applications that pass through untouched can be issued in one to four days, depending on outstanding requirements (MVR, MIB, Rx profiles). This can also have a significant impact on underwriting costs. Using typical requirement costs, predictive analytics can reduce underwriting cost between $125 and $200 per application.