Fulfilling AI's promise in workers' comp

Register now

The term “artificial intelligence” (AI) is weaving its way into the mainstream of insurance companies of all sizes. It’s a hot topic: Consider that at the recent Dig | In event in Austin, Texas, at least half a dozen exhibitors classified themselves as “AI” or “AI/Fraud” companies.

For many smaller carriers, however, especially those in workers compensation, the notion of artificial intelligence—and the expectations related to its application, i.e., where it could provide the greatest value—can be overwhelming. The improvements we’ve seen around risk calculation and pricing have, in some part, been fueled by data troves being sliced and diced to derive new ways of looking at risk, and enhance the data-driven insights that can be then integrated with other state-based rates.

Another area where AI is positioned to drive value is in claims. Unlike the initial success being seen by property and casualty insurers in applying AI to auto physical damage claims in order to predict such things as frequency and severity, workers comp insurers are at the starting gates here, too.

But the promise is there: Consider how researchers from the National Institute for Occupational Safety and Health (NIOSH), together with colleagues from the Ohio Bureau of Workers’ Compensation, used artificial intelligence (AI) or machine learning methods to successfully auto-code more than 1 million workers compensation claims. Following a machine learning approach, researchers ”taught” a computer to use an extensive and complicated data set to answer the research question “What caused this injury?” The claims were placed into one of three categories: (1) ergonomic related; (2) slips, trips, and falls; and (3) all other categories combined. The Journal of Occupational and Environmental Medicine noted that what took the revised computer program less than 3 hours to finish would have taken 4.5 years to manually code at an average manual coding rate of 2.2 claims per minute.

So, knowing what is possible, where are we as an industry on this journey? Not surprising, even the largest carriers are still evaluating the troves of structured and unstructured data related to claimants’ injuries, health and return-to-work status in order to make better claims decisions.

In my view, the science of artificial intelligence is still more of a promise than a reality. At the heart of AI is the core of data science—the ability to take data from a repository in order to design, build and then successfully test advanced models based on machine learning algorithms.

In spite of data scientists working tirelessly to gain a better understand of the data they are working with, the reality, according to Bill Inmon, is that they spend 95% of their time actually just accessing, cleansing, and preparing data. In a recent blog, “Data Science—a Sign of the Times,” Inmon, the famous “father of the data warehouse,” pointed out some of the inadequacies of the understanding of the data scientist when it comes to data-driven insights. Here are just a couple:

First, the assumption is that the amount of business value present in data must be linear… the more data, the greater the business value. “The reality is much different,” says Inmon. “The truth is that some data has tremendous business value and other data has very little business value. Trying to squeeze business value out of data when it isn’t there in the first place is a fool’s game.”

Second, the output from the data science effort must directly and clearly relate to the improvement of business in the eyes of the C level executive, says Inmon. “The data scientist can create the most elaborate and most sophisticated output in the world, but if the C level executive cannot understand it, and if the C level executive cannot use it to answer at least one of these questions (How can I get more customers? How can I generate more revenue? How can I reduce expenses?) then the data science effort has been a waste.”

In other words, all the fancy footwork and academic algorithms in the world won’t make the business run better unless the end result contributes to a material improvement, Inmon says. And based on the fact that the workers comp industry is known for its methodical and measured response to the application of the latest technologies, this may actually be an advantage.

So, for now, perhaps the term “artificial intelligence” is more about marketing than about reality for the average workers’ comp insurer. But the improvements workers comp insurers are seeking in terms of data-driven decision making can’t be denied.

For reprint and licensing requests for this article, click here.
Artificial intelligence Machine learning Workers' compensation