Momentum is building for change in the life insurance industry. In the past year especially, companies have begun to actively explore new ways to deliver their products to the marketplace. The focus has been on improving the customer experience and driving new business. Supporting this focus has been a drive to bring new data, whether that be electronic versions of existing data or completely new sources of information, into the underwriting process.
Historically, point of sale underwriting has been limited to simplified issue policies which are constrained to lower face amounts and higher premiums. However, with additional data sources and processing techniques, accelerated underwriting is now possible for new business outside the simplified issue market.
Criminal history, clinical laboratory results, electronic health record, financial background, underlying credit history components and social media content are examples of data that are either accessible now or are in development. With many of these new data elements and their combinations comes a new challenge in the form of risk scores.
The majority of these scores are created from highly credible data (solid data, lots of records, good mortality feedback, etc.), using sound statistical modeling techniques and, on a standalone basis, appear to perform well as a predictor of mortality. The challenge comes in determining if or when, where and how these scores can augment or replace traditional underwriting processes.
Legacy underwriting is significantly based on expert opinion and continuous adjustment, some by trial and error. It relies on rules and qualification ranges. Interaction between various sources of information is less strongly considered than in statistical model approaches.
Traditional underwriting has generally resulted in consistent, predictable long term mortality experience within an underwriting class. (This likely explains why the industry has been slow to introduce new data and technology to the underwriting process.) However, while mortality experience may be acceptable at an underwriting class level, a closer look shows that individual mortality within an underwriting class can vary greatly between applicants (Figure 1). In other words, legacy underwriting is good at predicting mortality experience at the underwriting class level but less so at the individual applicant level.
Figure 1 - Rule Based Underwriting Identifies Class Risk
Risk Score Based Underwriting
By contrast, mortality risk scores are typically built using statistical models designed to express the actual mortality risk of an individual. These scores take the form of underlying hazard values or some mapped structure for the hazard value such as a percentile.
Successful use and integration of a score can vary depending on the data that is used to build the model and how that data relates to the applied population. For example, a risk scoring model that uses property and casualty data (life insurance applicants would likely be a subset of this data) will need recalibration to be more applicable to typical life insurance applicants.
Likewise, a model built using a population of life insurance applicants with low policy face amounts will likely have key attachment point variation when applied to a high face amount population. The underlying mortality patterns may hold up, but at the very least an adjustment factor may be needed.
With this in mind, it is critical to understand how the score performs in a specific population and to appreciate the relative differences between each population. Scores are unforgiving. If the model determines an association exists between an input and mortality, the score will reflect it.
Figure 2 - Risk Scores Identify Individual Risk
However, correlation does not always translate to causation. For instance, if an observation is that people who eat curly fries live longer, their longevity does not necessarily mean that it is because they eat curly fries.
Another consideration when utilizing scoring models is that low incidence conditions will not be reflected well within a score, as scores can only reflect the information available. Statistical significance requires a minimum number of observations to establish a correlation. Scoring models can also be less flexible than the traditional underwriting process because adding information typically requires the entire model to be rebuilt. Models look at associations. If someone has multiple conditions, many models will tend to skew towards riskier scores due to rating the overlapping conditions and not being able to fully adjust for the overlaps.
Based on experiences to date, there are many hurdles in attempting to implement a standalone risk score. These range from supporting the producers to legal and compliance with several challenges in between. The most successful initiatives have taken a hybrid approach – layering risk scores with traditional underwriting in various combinations.
It is important to realize that both legacy underwriting approaches and new model (risk score) approaches are predictive models. Due to the different nature of each approach, combining them is not an insignificant undertaking.
Legacy, rule-based selection creates pools of individuals who together meet certain mortality expectations, but the individual mortality risk within the pool is often widely dispersed. Scores can be used to expose and mitigate the dispersion
Companies need to determine when in the underwriting process to insert the risk score – before or after legacy underwriting or somewhere in the middle. Where the score falls into the underwriting process can have significantly different impacts on the overall process and its outcomes. In some cases the resulting underwriting decision will be counter-intuitive from past approaches, causing underwriter and producer consternation and anxiety.
Figure 3 - Applying a Risk Score Before Underwriting
Figure 4 - Applying a Risk Score After Underwriting
Integrating risk scores into the underwriting process can be time consuming, complex and risky. However, in the near term it is also the most realistic path forward to gain the advantages the scores offer while not unbalancing the rest of the underwriting and sales process. Future iterations of these scores and greater familiarity with their potential and impacts will continue to allow them to take on a greater role and influence in underwriting.
Once you have determined your company is ready to move forward with investigation into one or more of these data sources or risk scores, you will need to do or obtain several things in order to be successful:
- Stakeholder consensus on overriding business goals
- A baseline for current portfolio composition, key metrics and performance. Many of these data points likely already exist and now need to be brought together.
- Knowledge of the data used to build the score and how the score relates to your applicant population
- A parallel study of legacy underwriting data including decisions to which the scores can be appended. This will allow better tuning of the selection rules.
- Key performance indicators and a method for tracking them on your block of business (e.g., if your population had 18% tobacco user self-admission previously it should remain relatively stable if the new process is successful)
- Proactive monitoring
- Adjustments and changes based on lead indicator findings
- An aggressive communication and feedback plan for all stakeholders
Just as legacy underwriting required trial and error to perfect, so will these new approaches using risk scoring. However, the legacy underwriting process has evolved over many years, even decades. Current market needs will not allow for that much time in regard to adopting risk scores. Reacting to early results and adjusting the process will have to be done in faster iterations.
Risk scores promise to be great tools if built for and assimilated appropriately into life insurance underwriting. SCOR believes they are here to stay. We are heavily invested in studying and understanding the effectiveness of risk scoring and helping clients effectively integrate risk scoring into the underwriting process.