Skip to content

Know the score: For this life insurer, an algorithm changed everything


When a life insurance giant applied artificial intelligence and machine learning to its own massive application database, the result was staggering: risk pools with 6% fewer deaths. This is their story.

Here’s a challenge about research: You don’t know how it will turn out. You have a question. Or a theory. And you don’t know the answers. (That’s why you’re researching in the first place, right?)

Here’s a fact about actuarial science: There’s always an answer. You want to know the probability of something. You don’t know what that probability might be. But you do know said probability can be calculated given sufficient data.

Now combine those two. Here’s the result from actuarial research in insurance in the modern era: The answers to your questions can change everything about your business.

We know. It happened to us at LifeScore Labs, an insurtech subsidiary of insurance giant MassMutual.

The beginnings of AI and ML

Like all insurers in the early 2000s, MassMutual had questions about how artificial intelligence and machine learning might transform the industry.

Click me

The answers were out there; early work in probability and risk modeling made that clear. So MassMutual decided to do the research. The hope, of course, was that the result would be a level of improvement in something — speed, efficiency, cost, other areas — to warrant continued investment in advanced data science.

Other insurance companies tried to do this research. But MassMutual had an advantage: a data set of a million applicants spanning 15 years and containing health, behavioral, and financial attributes. That may be the largest and most comprehensive application data set in the industry.

What’s the score?

MassMutual combined that massive data set with survival modeling to develop a “life score,” deployed in an algorithmic underwriting system. And when compared to traditional underwriting methods, the results were, in a word, staggering.

The research showed that if the new AI and machine learning model’s risk-selection methodology had been applied to historical applications, it would have produced risk pools with six percent fewer deaths.

That is, the new “life score” outperformed traditional underwriting by six percent on the basis of claims. It was a remarkable reduction in projected mortality loss—and ample reason for the company to embrace the model in its operations.

That “life score” today bears the brand names MassMutual Mortality Score (M3S) and LifeScore Med360. It underpins the company’s underwriting process for most life insurance policies.

We had questions. We had the science. We had the data. And the answers changed everything.

If you’d like to see how the life score model compares to your existing underwriting methods, let’s discuss.


Find your score