Although predictive modeling has been utilized by P&Cinsurers for years, the focus has primarily been on pricing andunderwriting risks. We now are seeing increased interest andutilization of that process to manage claims. Current exampleapplications of predictive modeling for claims include:

• Identifying potentially fraudulentclaims
• Predicting which claims may benefit fromhaving a specialist assigned to them
• Developing logic to more scientificallytriage claims to the right level of expertise in theorganization
• Determining what differences in the book ofbusiness are impacting claim severity

Process Can Reduce Errors

At P&C insurers and third party administrators, the claimsmanagement task is handled by teams of professionals consisting ofclaim adjusters, claim managers, nurse case managers, client riskmangers, and other claim and medical experts. While theseprofessionals generally have years of experience and can identifyclaims that may need intervention, predictive modeling can help adddiscipline to this process and reduce the potential for human errorby processing multivariate patterns that humans are incapable ofrecognizing.

At SRS, our current predictive modeling solutions rely on datafrom several areas: claimant demographics, employment data, injurydetails, medical billing information, and claim financials. We alsouse natural language processing to text mine and extract keyinformation from claim handler notes, and then integrate this datainto our models.

Predictive modeling can be especially useful in identifyinglarge losses. While large losses account for only five to tenpercent of open claims across the workers' compensation market,they account for 60 to 75 percent of the outstanding dollars forworkers' compensation claims.

Early identification of large losses is difficult, as claims cantake many different courses based on the claimant, the injury, andthe treatment. When seemingly innocuous claims deteriorate, itoften takes more than a year before their true costs become known.Failure to recognize the magnitude of these claims early can put aninjured worker down the wrong path and lead to spiraling claimcosts.

Assist in Early Identification

Predictive modeling can facilitate early identification ofclaims that would benefit from nurse case management. The processalso can flag claims that may not currently have large losses butdo have a high likelihood of adverse development.

At SRS, our nurse case management model estimates financial riskand evaluates claims shortly after intake and then performs anothersweep at 25 days when additional data is available. We have theflexibility to utilize either a score-based threshold or avolume-based threshold to determine which claims are likely to needintervention from a nurse. The goal is to reduce overall claimcosts through targeted early intervention and to get the injuredworker back to work as soon as possible.

Industry estimates indicate that fewer than 10 percent of largelosses are recognized at 90 days. At SRS, by using predictivemodeling, we can identify up to two-thirds of large losses byfocusing on those claims that have scored the highest.

It must be noted that no predictive model is perfectly accurateand that accuracy depends heavily on the manner of deployment.Nevertheless, those companies striving to lead the pack in claimsmanagement and mitigating losses would be wise to incorporatepredictive modeling into their strategies and workflows.

Kaleb Adams is vice president of predictivemodeling at Specialty Risk Services, Hartford, Conn. On Monday,Aug. 16, he and other modeling experts will examine predictiveanalytics at the FWCI 65th Annual Workers' Compensation EducationalConference and 22nd Annual Safety and Health Conference in Orlando.The discussion will center on how insurers and their riskmanagement clients can get a better handle on their respectiveworkers' compensation exposures using these latest technology toolsand techniques. In addition to Adams, panel members includeJennifer Tomilin, senior vice president, Zurich North America,Schaumburg, Ill.; and Steven W. Laudermilch, senior manager,Deloitte Consulting, Philadelphia. Sam Friedman, editor-in-chief ofNational Underwriter P&C, Hoboken, N.J., will moderate.Conference information is available at www.fwciweb.org.