Following an investigation, for example, an adjuster may conclude that the loss suffered by an insured was caused by a hit-and-run driver, and he or she may therefore intend to find out if the police have identified a suspect. Because of the workflow pressures mentioned earlier, however, that well-intentioned adjuster may forget to add a "diary" for this action, and the follow-up may never actually happen. Similarly, even if the adjuster discovers that the police have identified the suspect driver, the adjuster may miss subrogation on the claim. Finally, in the event the claims adjuster determines that subrogation exists, he or she may very well miss sending the case to the recovery unit because of the workflow pressures.

Eliminating Oversights
Text mining can help alleviate such oversights by automatically analyzing the adjuster's notes and sending alerts about opportunities for action. As a simple example, a text mining system may recognize the phrase "hit-and-run driver" and automatically add a diary entry for the adjuster to "please identify hit-and-run driver." Another text mining engine might closely follow hit-and-run cases and seek to detect the concept of "adverse driver identified" in the ensuing adjuster notes. Once a detection has been made, the claims adjuster could automatically be prompted to "please evaluate subrogation opportunity." Such text mining alerts can significantly improve the efficiency of the adjusting process while drastically reducing the number of missed opportunities for adjuster action.

Although setting up a text mining system can be challenging given the unstructured flow, typos, and abbreviations in adjuster notes, the eventual business results are well worth the effort. Furthermore, once a core text mining engine is established, it can be quickly and easily extended to produce insights across a variety of claims concepts, from identifying the at-fault party to fraud red flags, dissatisfied claimants and insureds, and details about the location and the cause of a given loss.

Mining for Fraud Mitigation
Fraud is a challenge that pervades all industries, especially P&C insurance. Estimates from the Insurance Information Institute (I.I.I.) indicate that fraud accounts for approximately 10 percent of the P&C insurance industry's incurred losses and loss adjustment expenses (LAE)—or about $30 billion each year. Recent indications also show that fraud is a growing menace in personal injury protection (PIP) states, with direct impact on company results and insured premiums. Recognizing, investigating, and denying suspicious claims are thus key priorities for carriers.

A common example of automobile insurance fraud is the "swoop and squat." This refers to a scheme in which several potential fraudsters use an older car to cause an accident with a newer luxury vehicle, based on the assumption that the owner has high insurance coverage limits. The scammers pull in front of their intended target vehicle—which is usually near a highway exit—and stop suddenly, thereby causing a collision. They then report damage to the car and fake numerous soft-tissue injuries, leading to a big payday for them and a big loss to the victim's insurance carrier.

While experienced, alert adjusters can recognize such schemes and get the Special Investigative Unit Services (SIU) involved in a timely manner, workflow pressures, lack of experience, and inadequate training can often lead to missed opportunities. Furthermore, fraud schemes are constantly evolving, and patterns may not be as easy to detect as in the above example.

Text mining can be leveraged to distill insights from adjuster notes to systematically create a multitude of fraud concepts, such as questionable injury, excessive treatment, low-speed accident, sudden stop, and an accident near a highway exit. Predictive models can then be established using these concepts—along with other structured data from the claims system—to generate highly accurate and actionable referrals for IME or SIU intervention.

Improving Workflow Routing
In addition to alleviating flaws in the claims process, text mining can help reduce blemishes in workflow. Many insurers continue to use legacy claims-handling systems. An adjuster may identify subrogation opportunity on a claim, but to route it to the recovery unit, the adjuster may have to navigate through a number of screens or pages to the recovery referral screen. Worse yet, at some carriers, the adjuster may have to open a separate database and cut-and-paste information into it to route the claim to the recovery unit. Owing to this obviously burdensome process, many valid referrals are often missed.

Text mining, coupled with strategic phrase training for claims adjusters, can eliminate inconsistencies and improve the workflow process. Managers can train their adjusters to enter a predetermined set of phrases into their claims narrative, and a nightly/batch text mining system can be set up to automatically identify and route the claims for appropriate business action. For instance, the adjuster might write "*** Route to Recovery ***" in his or her claim notes as the signal to the text mining system to send that particular claim to the recovery unit.

Simple applications of text mining such as this can increase adjuster productivity and potentially improve a company's effectiveness in handling previously missed subrogation opportunities.

Manage the Customer. Innovate the Process.
Customer contact areas at insurance companies typically deal with issues other than claims. The issues can range from billing concerns to policy- and coverage-related inquiries.

Traditionally, these comments are buried in CRM-like systems and seldom leveraged. With increased use of e-mail, voice-to-text applications, and social media sites, however, the volume of customer commentary that can be made available to be mined is increasing. The information derived from these contact points is rich and may be used to increase customer satisfaction and retention, improve business processes, and even identify areas for innovation.

Customer comments can be systematically text-mined to discover persistent issues and new ideas using sentiment analysis—essentially the ability to discover opinions and tones in text. For instance, customer complaints regarding the billing system may reveal a sentiment of dissatisfaction through words such as "unable, down, terrible, and slow."

Armed with the type of issue and the sentiment, an insurance carrier can take proper action. The complaint can be routed to the appropriate customer service representative to resolve the customer's issue quickly and create a positive customer experience. In the longer term, persistent patterns of specific issues can be used to make the business case for prioritizing resources and budget investments. For instance, multiple customer complaints might help focus the carrier's attention on improving the billing system, while customer suggestions for an iPhone app to handle account management and claims might trigger new product development investments.

Critical Business Insights
In sum, text mining is a versatile analytic capability that offers numerous benefits to the P&C back office, from effectively analyzing losses and routing claims to combating fraud, minimizing missed opportunities, and enhancing the customer-contact experience. These techniques are quickly transforming obscure textual data assets into insightful business-critical applications that enhance claims efficiency, customer satisfaction, and responsiveness—cornerstones of customer loyalty and retention. Insurers that successfully adopt text mining can significantly improve back-office functionality and effectively provide their customers and policyholders with excellent service and favorable returns for the company.