Can AI Democratize Life Insurance?

The United States is seriously under-insured. Forty percent of the populace has no life insurance, and among those who have it, one out of five say they don’t have enough.

Why should one care? If you die without life insurance you leave others to pay your debts. If you head a family, you put your spouse and children at risk.

The gap persists partly because fully underwritten life insurance is difficult to purchase. It typically requires a mini physical that includes the drawing of fluids (e.g., blood and urine). Underwriters, for their part, often avoid applicants with serious diseases like cancer because those cases require an arduous — and expensive — medical assessment process.

Several new startups, however, are using AI (artificial intelligence) to address these problems. In the process, they just may democratize life insurance.

In July, Haven Life, a Massachusetts Mutual subsidiary, announced that it was expanding its fully underwritten term life insurance policy to non-U.S. citizens and individuals with chronic illnesses. Haven Life use artificial intelligence to match applicants with plan options with face amounts up to $3 million. A process that had taken a month(s) now takes 12 days or less. (Policies can be issued in minutes for lower face amounts.)

Atidot, a Tel Aviv-based InsurTech startup, is using artificial intelligence, machine learning (ML), and predictive analytics to help life insurance companies optimize their books of business. Recently, it assisted a top-10 U.S. life insurer in identifying underinsured policy-holders to the tune $1 billion.

AI and machine learning can have an “exponential impact” on the life insurance industry, says Atidot CEO Dror Katzav. In the past, carriers may have utilized a handful of predictor variables to identify insurance prospects, e.g., age, gender, smoking, etc. His firm uses 500-600 variables.

One variable, for example, indicates whether your household income is more or less than your neighborhood’s median household income. This could be associated with a propensity to buy life insurance.

In making predictions, Atidot employs sophisticated ML algorithms like Deep Neural Networks (DNN) and Random Forests, but it avoids overly techy language when interacting with insurance executives or regulators. Discussions are focused on premium levels and missing payments  — not output neurons or sigmoid functions.

“Black box” type ML methods often don’t work well in a the insurance business context, adds John Lucker, Principal – Deloitte Advisory, Deloitte & Touche LLP.  They are often not understood by C-suite executives, and they are the ones that have to support a  new project. For that reason, many insurers still favor more interpretable algorithms like logistic regression.

Everyone wants to know why an underwriting decision was made,”  explains Laura Boylan, Haven Life’s Product Owner for Algorithmic Underwriting, so model interpretability matters.

To the extent that AI, or machine learning, can simplify the underwriting process — make it less intrusive, faster — more people may be encouraged to apply for coverage, says Boylan. In the case of the chronically ill — people with diabetes or depression, say — Haven Life uses machine learning algorithms like Random Forests to cross-check public records (e.g., motor vehicle records) to verify the information that applicants have entered on the company’s website. The system may prompt additional questions, like details of an applicant’s prescription history. The algorithms don’t reject applicants, however, they simply raise a red flag–at which point human underwriters are summoned.

The entire life insurance industry recognizes that machine learning is important, and will play a critical role in the industry’s future, Boylan says, but it remains difficult to integrate these new processes into legacy infrastructure systems. Many life underwriters are working with systems infrastructure that is very old–going back to the 1970s in some cases.

Katzav expects insurance companies implementing AI to begin with the “easy pieces,” like enhancing the customer experience, which can have an almost immediate financial return, and eventually apply AI into risk management, and product development.

“Fraud prevention is huge,” adds Boylan. Machine learning is already used by 22% of life carriers to identify potential fraud, according to RGA’s 2017 Global Claims Fraud Survey.
It is most common in the Asia Pacific region where its use has reduced end-to-end time for cases involving fraud by 15 days, or about 20 percent.

With AI, you are changing the paradigm, says Katzav. “You can interact with the customer in real time. You can analyze risk in real time, when the interest rates are changing, for instance, not months afterward. It creates a better customer experience.”

About the Author

Andrew W. Singer, principal, Singer Publications, has been a professional business writer and editor for more than 30 years, including two decades as editor-in-chief of a financial services magazine. He received an M.A. in statistics from Columbia University in 2017 and has been an associate instructor in Machine Learning within Columbia’s M.S. Program in Applied Analytics.

Sign up for the free insideBIGDATA newsletter.