The Black Box Changing Life Underwriting
ANAHEIM, Calif. – “Predictive modeling” is coming to life underwriting, and while it may help reduce underwriting costs, the approach may be confusing to clients or even a touch scary, says a long time underwriter. Advisors will therefore need to be ready to educate on this.
Predictive modeling refers to systematic analysis of data, including historical information. In the life underwriting context, this data is about consumers.
Analysts use the information to create predictions, or models. Underwriters can apply the information to life insurance applicants in order to determine their risk classification for the insurance coverage they are seeking.
Karen Phelan, a veteran underwriter from Springfield, Mass., outlined the various types of models that are emerging during a wide-ranging talk about underwriting trends at the 2012 Million Dollar Round Table Annual Meeting.
Advisors may be familiar with the approach in other, non-insurance contexts. For instance, predictive analytics have been used in credit risk scoring, medical research, consumer marketing and even meteorology, the underwriter said. “iTunes uses it through its Genius application to provide song recommendations; Netflix does the same for movies.”
In the life insurance world, the models have the potential to be faster, better and cheaper than traditional underwriting methods, and thus a better client experience, Phelan said.
But the models have raised concerns, too, since they “are a kind of black-box approach and do not provide the type of transparency important to producers and clients when determining the risk class,” she said. There are questions about model credibility, actuarial soundness and regulatory considerations, too.
What to Know
Predictive models fall into different categories. Phelan considers one of them — the consumer data model — as having “a little scary” side to it.
The consumer data model involves applying information obtained about an applicant from various databases and information resources. The data might show things such as the applicant’s online purchases, magazine prescriptions, TV watching behavior and leisure activities, the underwriter said.
“It’s pretty scary how much someone can know about you with just a name and an address ? and even scarier if you provide your Social Security number!”
She said the underwriters could use analyses of the data and then draw correlations that point to increased risk behavior or the development of adverse health issues.
The scary part concerns the conclusions that might be reached in this manner. For instance, Phelan said, “If my kids left the TV on all day, am I deemed to have a more sedentary lifestyle and therefore a higher risk of becoming morbidly obese or developing diabetes? Or, what if I decide to buy some running shoes, a treadmill and some fitness magazines?”
The consumer data model is already being used as a screening tool for those good risks, the underwriter pointed out. It is not, she said, purported to be the approach to use in situations requiring an adverse decision — or at least not yet.
If the data were used for adverse underwriting decisions, she quipped, “what would the adverse letter to the applicant say? ‘We are unable to offer you coverage as applied for due to your couch-potato tendencies and where that might lead you?’”
Other predictive models include:
The laboratory/physical measurement scoring model. This takes the applicant’s lab data and physical measurements, build and blood pressure and runs them through the model to create a holistic picture of the insured rather than looking at each factor exclusively. It uses data that the life insurance business has traditionally viewed as having significant protective and predictive value, but Phelan foresees challenges in explaining to clients a resulting decision that may not be “best class,” even though all the components independently look favorable and within normal ranges.
The biometric model. “You could argue that the lab/physical measurement model is a true biometric model, and you’d be right,” Phelan said. But this model also includes lifestyle factors and incorporates clinical studies on just about every biomedical marker (smoking, cholesterol and body mass index), plus information such as driving history and lifestyle, she said. Then it blends the data to determine if an applicant is standard, preferred, or super-preferred.
Carriers are trying to determine the approach to use with this modeling and the extent to which they will rely on it as part of the assessment process and for predicting mortality, Phelan said.
But in the next decade, she predicted, most carriers will incorporate predictive modeling into their process in some way. This will likely focus, at first, on simplified issue or low-risk products, on identifying very good risks quickly, or on augmenting more traditional underwriting.
“Producers and consumers alike will need to be educated and aware of the approach to risk assessment,” she said.
By Linda Koco InsuranceNewsNet. Linda Koco, MBA, is a contributing editor to InsuranceNewsNet, specializing in life insurance, annuities and income planning.