Orlando

While predictive modeling won't replace knowledgeable insuranceadjusters and underwriters, it is the wave of the future for theworkers' compensation industry, a trio of experts herestressed.

Jennifer Tomilin, senior vice president at Zurich North America,said she could not foresee human underwriters ever being replacedby automation because “there are areas where we don't have enoughdata for predictive modeling.”

However, modeling will be a bigpart of every insurer's future, while companies that are “stuck inthe mud” and fail to take advantage of this evolving technology,“those folks are going to be left behind,” warned SteveLaudermilch, senior manager at Deloitte Consulting.

The three panelists offered their views and predictions during apanel session–”Tech To The Rescue: Can Predictive Analytics SaveWorkers' Comp?”–here at the annual Workers' CompensationEducational Conference. The session was part of the National Trendsprogram put together by National Underwriter inpartnership with the WCEC organizers–the Florida Workers'Compensation Institute.

The speakers outlined a variety of ways in which modeling isused, what data types it employs and what it targets.

Ms. Tomilin said modeling–sometimes called data mining–is ananalysis that finds unsuspected relationships that relate to datain novel ways.

For example, in researching a company, the modeler might look athow many employees in a particular ZIP code work from home,figuring a return to work program might be easier to employ for aninjured employee who does not have to commute to their job.

Some data that is sifted through is ultimately tossed out, shenoted, giving an example of an examination of sweet dessertpurchases in a certain ZIP code for a possible link to truckaccidents, because eating sugar can cause drowsiness.

She also noted that discriminatory data of the kind once used inredlining is rejected.

The value of predictive models, she said, is that they take “alot of the subjectivity out” of analysis and eliminate “thegut-feel practices” of underwriters.

Predictive modeling can serve insurers by, among other things,targeting fraud, identifying claimants who will benefit bytreatment from a specialist, and in helping to “triage claims,”according to Kaleb Adams, vice president of Predictive Modeling atSpecialty Risk Services.

While he said data from an insurer's underwriting department isnot always that reliable or useful in predictive modeling, soliddata is voluminous from the claims sector and can be used toprovide facts concerning a loss, a worker's claims history andco-morbidity factors such as obesity.

He said modeling can pin down cases that would benefit fromhaving a special nurse assigned, those that indicate they willinvolve a large loss, and those that need less attention.

Mr. Adams said the system can use text mining to read over notesand identify a potential fraud situation, a health problem such asmorbid obesity, or a risk management threat such as a wetfloor.

Models that are used, Mr. Laudermilch noted, are built bytesting the impact of hundreds of variables–such as lifestyle, age,employment history and pharmacy drug use–against closed claims.

The models, he said, will be based on data on injury groups,such as back strains, to identify which are the more serious casesin relation to all the others.

“This is not a replacement for adjusters,” he noted. “The ideais to give them claims they can work on, where they can make a bigimpact” accelerating best practices.

Explaining some of the areas where data is derived, Mr.Laudermilch noted that companies glean it from items such aswarrantee forms that customers fill out.

Learning that a person has interests in outdoor pursuits orrunning for fitness can lead to a conclusion that if injured, “theyare probably motivated to get healthy”–more so, perhaps, thansomeone who is a “couch potato,” he added.