Hard Data + Higher Math = Better Outcomes.
You hear it all the time: history repeats itself. In insurance, history holds outsized importance, in part because it is what governs the future. How claims have been handled in the past largely indicates how they will be handled today. Yet the industry’s longstanding best practices for analyzing historical data—combining some basic rules featuring only a handful of variables, supported by intuition and hunches—has invariably led to less-than-optimal outcomes and wasteful expenditures. Conventional wisdom is undoubtedly fallible.
The arrival of predictive modeling in commercial insurance lines promises to go a long way toward improving that situation. It brings greater mathematical muscle to the table. Industry leaders such as Zurich are putting vast resources behind applying disciplined, statistical techniques to large quantities of data, including data beyond internal historical data, such as industry and census data. By considering many variables simultaneously and sorting out the underlying relationships, customers gain greater insight into claims cost drivers and can focus on interventions that provide the greatest benefit. Preexisting ideas don’t need to dictate as the data speaks for itself, producing greater insights.
The earliest applications of predictive modeling in claims were in fraud detection—identifying characteristics that suggested a claim might be suspicious and need further investigation. The upfront work required for predictive modeling—assembling, collecting and “cleaning” vast quantities of data—required a large investment. Once the foundation is in place, the process gets easier; smaller, incremental investments are needed to move on to models for other issues like severity and propensity. Severity refers to the likely final cost of the claim after all the factors have been considered: in workers’ compensation, for example, severity is broken down into medical cost and the associated wage, or indemnity, cost. Propensity identifies the likelihood that a claim will be a lost-time claim as opposed to a medical-only claim.
“There are almost as many applications for predictive modeling as there are business problems,” says Tom Barger, Director, Zurich Services Corporation, Zurich in North America Claims.
Predictive modeling has helped clarify the notion, for example, that the same injury can have different values in different industries in different states. Such modeling helps put the right claim professional on the claim from the beginning, in terms of the skill level needed. From an efficiency standpoint, the insured gains better, more cost-effective outcomes by paying only for needed expertise and getting proactively managed—not over-handled—claims.
“You want to use appropriately skilled resources when it’s a simple claim,” says George Hansen, Chief Financial Officer, Zurich in North America Claims. “Similarly, when you have something severe and complex, you want to make sure that you have more experienced claim professionals on the case so they can be proactive in identifying issues. For example, in workers’ compensation, we might find characteristics that a claim may benefit from having medical oversight in place sooner. The claimant can get appropriate treatment earlier in the process or make sure there isn’t overtreatment. This allows us to better manage the loss costs.”
“The experienced claim professional has the skill to identify the potential issues with the claim early enough to say, ‘I can see where this is headed—we need to intervene with a proactive return to work plan,’ or, ‘I need to get a nurse involved immediately,’” Barger says.
In workers’ compensation, many customers focus on the severity of a claim. Barger notes that the true bottom line is the total cost of risk, and says that predictive modeling has helped to show the complicated dance between severity and frequency. Its ability to shed light on this relationship will only increase moving forward, to the benefit of both insurer and insured.
“Simply because you have ‘lower severity’ doesn’t necessarily mean you’re getting the best outcome from a total cost of risk standpoint,” Barger says. “The industry hasn’t woken up to that yet. It will start weaving the severity and frequency models together, along with the expenses, to develop a total cost of risk approach for customers.”