Our Wired World

Is Machine Learning Ready For The Big Leagues?

How Legal & General is stepping up to the plate in the life sector

by Fred Tavan

Mr. Tavan is the SVP and Chief Pricing Officer at Legal & General America where he leads actuarial, underwriting, and data science innovations. Visit www.lgamerica.com.

The insurance industry has been humming along nicely with a system that works and is well-understood, if at times a bit clunky, analog and cumbersome. While spending some time in the Property & Casualty league, machine learning is beginning to step up to the plate more frequently in the life insurance sector… and making quite the impact. But with hot shot rookies come inevitable challenges along with the need to find the right balance of shaking up the system and valuing what got the industry to this place already.

Spring Training

P&C has always had a faster advancement toward digital transformation given the types of data it has at its disposal, but the lessons and applications that the industry uses can act as a playbook for the next iteration of machine learning in insurance as it finds ways to integrate into the complexities and long-term realm of life insurance. But it’s not as simple as dropping in new algorithms and models for carriers or brokers.

Pre-seasons exist for a reason, to try out the rookies and see where they are best deployed, see where they can go while acknowledging where they are currently best suited. When calling up machine learning to the life insurance industry, Legal & General America (LGA) looked first at what key performance indicators (KPIs) would be both the most trackable and allow for the most seamless integration.

Looking at the fact that LGA receives more than 100,000 life insurance applications each year, it has a massive bank of data going back decades on how different conditions have direct causation to life expectancies. The underwriting manuals, which dictate application statuses and rules as to who is or isn’t insurable, are dense and can be complex. But when organizations work to automate their manuals first, they are able to then create if/then statements for those rules, and let machine learning handle the simpler cases. This allows the complex cases with impaired risks to be passed along to the seasoned underwriters for that special touch and expertise.

Regular Season

Automating manuals, whether for underwriting or otherwise, is the perfect warm-up for organizations to introduce automation to their systems and get the existing teams comfortable with the digital transformation process. From there, it’s looking back to the KPIs and where along the underwriting and insurance timeline this technology can be most useful.

As most advisors know, a common complaint around getting life insurance policies has long been the poking and prodding of additional exams and tests. So, setting a KPI on reducing the number of applicants that need to undergo those additional tests was the logical next step for LGA. Once the systems have learned the underwriting manuals, it can look at the causation models from years of underwriting data around health-related conditions to see if there is anything in the applicant’s chart that would warrant follow up exams or attending physician statements (APSs).

Having historical and real-time data is crucial for the machine learning to work – it can’t develop a lot of science or expectations around two people in a cell. The historical data is where life insurance does all the underwriting insights and analytics to see where it went right or what it got wrong. The new processes take all of this to make more accurate predictions in real time.

Automating manuals, whether for underwriting or otherwise, is the perfect warm-up for organizations to introduce automation to their systems and get the existing teams comfortable with the digital transformation process...

These systems assist in taking the simpler applications off the underwriting’s team to-do- list by merging the historical knowledge and real-time data that’s available. If an application comes in with 40 different data points, the model can see that maybe eight are driving the most protective value and those eight also line up with real-time data sources that are available now – electronic health records for example. When there is solid alignment, that’s when the wealth of underwriting acumen and today’s data can come together. What’s even more intriguing of this new player in the insurance game is that it can also share a percentage of how often it can get it right or wrong based on different data sources.

AI is built in a way that it can also raise the flag when it needs an underwriter to pinch hit – something humans often have a harder time doing. Knowing and asking for assistance without the accompanying shame that the human ego brings, is an unexpected perk of the digital transformation, and something we should try to learn from machines even as we keep teaching machines new rules and models. After all, teams are built on the collaboration of all resources available.

The dramatization of both baseball teams and AI can have some imaginations running wild on potential for growth. The MLB may have been able to bring Field of Dreams to life (minus Shoeless Joe) but the uncertainty and the fear that goes with it around AI and machine learning can illicit more skepticism from advisors and applicants. That’s where the umpires come into play – the possibilities are endless with the new technologies.

It seems daily there’s a new way that P&C is taking satellite images of homes or cars to push innovation forward. The regulations in place around life insurance rein in the propensity to have a wild thing in the roster. Data is not perfect, it can be flawed and biased just like humans, which is why the data scientists setting up these models must be aware of and understand potential biases. Sticking to the health related, historical wealth of causal data vs. correlating data helps everyone stay safe. Carriers are simply able to get the same data in a more efficient manner with the latest adjustment to the batting order.

The Playoffs

The life insurance industry is still in the early innings with digital transformations and machine learning is still getting that batting average, but it’s making significant strides. As it continues fine tuning the easy applications, freeing up underwriters to be able to dig in and write more impaired risks to help more people find the comfort of having a life insurance policy, it’s impossible to not think about what could come next.

Building the skills of machine learning to develop into a solid utility player, the next phase would be introducing more natural language processing. The ability to be able to read the complex medical documents with a high level of accuracy, list all the impairments found and intrinsically know what to do with that information will bring machine learning to the all-star game. If these models can step in and start taking control of 85-90% of applications – the healthy cases – by pulling and reviewing prescriptions and what question the EHRs answered already, underwriters can take on the task of getting creative to write impaired risks and offering coverage to many who thought it wasn’t in the cards.

The life insurance industry is in the thick of its rebuilding years, challenging the stereotypical notions of the application process and all that underwriting entails. By combining the old school and new school ways of thinking, everybody wins.

The industry is able to enhance the ability to write more policies leading to more families and individuals having the coverage they need and more advisors making those brighter tomorrows. Who’s ready to play ball?