Tom Fletcher and Karen Phelan discuss how predictive analytics, given certain important considerations, is a powerful and enabling tool for Life underwriters.
Predictive analytics has been applied to nearly every industry and discipline to date. The correlation coefficient, a statistical workhorse underpinning most of predictive analytics, was a 19th century development. Factor analysis, a technique developed to reduce measurements of human abilities to sets of concepts, such as ‘intelligence’, followed in the early 20th century. Regression was widely in use by the mid-20th century. The reason for all these developments is that the world is multivariate, complex and full of uncertainties. By the late 20th century, models were being applied to predict human behavior in ways that were both fair and unfair. Many disciplines have since emerged to address what, how, and when predictive models should be fairly applied.
Data, bits of information converted to some structured form, impact our ways of working. But data is far more effective (usable) when coupled with predictive analytics. Building statistical, predictive models from data for a specialized purpose is therefore both an art and a science.
Data is more effective (usable) when coupled with predictive analytics.
In Life underwriting, that specialized purpose could be to directly or indirectly evaluate mortality risk. Most medical underwriting manuals will incorporate more than a dozen factors (e.g., BMI, blood pressure, cholesterol and A1C), often with three or more levels (below, average and above threshold). This creates an intractable amount of information for the underwriter to absorb. For example, five factors, each with three levels, equates to 243 combinations. This is too many conditions to process in our heads. Complex rule sets become essential.
With predictive models, the underwriter has at their disposal a tool to augment decision-making consistently and efficiently. However, these models don’t build themselves, nor should model usage mean usurping the underwriter. Underwriters should be involved in the process of model development, make decisions as to how models are deployed, and ultimately be freed-up by the models to focus on more complex underwriting cases.
With predictive models, the underwriter has at their disposal a tool to augment decision-making consistently and efficiently.
Predictive models are being used for identifying individuals at high risk of specific causes of mortality (e.g., cardiovascular risk and accident risk), as well as for identifying risk factors such as smoking propensity, truthfulness and the like. The outputs of these models could be estimates of some unknown (e.g., a measure of affluence and health proclivity), probabilities of an occurrence (e.g., death, purchase, lapse and smoking) or categories/classes of risk (e.g., preferred and standard).
Outside of underwriting, predictive models are often used upstream, for example by marketing to identify target cohorts of individuals with a higher likelihood to purchase and/or qualify for Life insurance policies, and a lower likelihood to lapse. Downstream (claims) models can also feedback information to underwriting. These models may or may not be aligned with the underwriting goals of separating risk. Ultimately, the underwriter needs to understand whether and how these models impact the insurance decision-making process.
If models are developed with the ethical principles of objectivity, integrity and fairness in mind, then there’s opportunity for enhanced consumer experiences, as well as trust among underwriters. These ethical principles require that we properly align the business problem to be solved with the rigorously screened and approved-for use of the data. Many people and machines can build models, but a collaboration of skilled professionals is required to put models that instill confidence into practice.
Using predictive analytics has at times been seen as a threat to underwriters, but more recently, the industry has come around. Underwriters are responsible for assimilating vast amounts of data in order to render an appropriate decision. With more data out there today than ever before, it’s hard to determine what’s relevant to the underwriting process and what’s simply ‘noise’. Predictive analytics can help to determine what’s useful and streamline processes. It allows underwriters to ‘up their game’ by using the right data in an efficient way to offer valuable insights that might otherwise be missed. The key is to ensure that we use the right data in a responsible way.
At PartnerRe, we have extensive experience in predictive analytics for Life and Health insurance, not just for underwriting, but also for Life insurance business needs up- and downstream of underwriting.
We not only develop predictive models, but also:
If you are developing or planning to develop your own predictive models and would like an experienced partner to guide you and work with through this process, please do get in touch.
Tom Fletcher, PhD., VP, Data Analytics, North America Life
Karen Phelan, Vice President, Underwriting Strategy & Innovation, US Life