There has been a lot to learn during the many months of the COVID pandemic. One of the greatest lessons that SCOR has learned is the value of finding creative and engaging ways, especially in a remote and virtual environment, to share insightful knowledge and information on topics of most interest to our clients. Without the benefit of in-person meetings, we have explored and used numerous digital technology tools to deliver meaningful insights and deep dive discussions from our perspective as one of the world’s largest reinsurers.
Through this process, we have also learned from our clients, who have unique experiences and understanding of their particular markets, products and clients. This mutually beneficial experience has deepened and strengthened our partnership relationships with our clients across the industry.
Most recently, we completed a series of deep dive virtual meetings with clients on the topic of predictive modeling and how it can be used in accelerated underwriting (AUW) programs. While there are many ways that predictive models can be used – from target marketing to claims analysis – these recent discussions focused specifically on AUW as part of our Underwriting Reimagined initiative.
During our client discussions, we shared the predictive modeling journey that SCOR has been on as a case study and offered suggestions on where and how clients may begin to plan or accelerate their own roadmap. We shared our insight of the value of predictive models versus rules-based AUW decision making.
While all companies use a rules-based safety net for their initial AUW program eligibility, most companies use a combination of rules and third-party predictive models for AUW triage. The case study, which was based on our own experience, demonstrates the added value of using predictive models that are built on a company’s own data (what we refer to as “proprietary models”).
We shared results from both acceleration models (models that predict where additional underwriting evidence is unlikely to change an AUW decision) and mortality stratification models (models that produce a stratified risk score to better assign cases to underwriting classes). In both cases, the power of proprietary models significantly outperformed rules-based outcomes.
One concern frequently voiced during these client discussions was regarding how regulators view predictive models. Regulatory concerns around potential discrimination and model bias is a growing issue for the industry, but we believe that these concerns can be overcome with strong governance, model decision transparency and clear communications that ensure model integrity.
With a commitment to avoid a "black box" mentality, SCOR has implemented an internal model governance framework that requires that models are free from discriminatory bias, justifiable in their outcomes, support adverse action processes and be based on causal factors where required. By proactively testing for “proxy discrimination” and self-regulating our data and models, we have been able to alleviate skepticism and accelerate predictive modeling acceptance among the clients we work with.
So what’s next for SCOR? We are currently working on expanding our capabilities to include predictive models that can be calibrated to any client company’s business. These models are expected to be available in 2022 to explore with interested client companies that lack the resources or credible data to build their own models.
We believe that predictive modeling will continue to drive the Life insurance industry toward individualized risk assessment and frictionless underwriting. While the companies we have engaged on this topic are at different points in their predictive modeling journey, all appreciated that we shared our insights and experiences and that we are leading change in the industry toward a future reimagined.