Divining Bodily Injury Claims

Predictive modeling—the process of using internal and publicly available external data to create a statistical model that projects future events—can be used to analyze a claim at FNOL and throughout the lifecycle. An effective model design process considers multiple internal and external data sources, identifies claim characteristics with predictive power, and creates an algorithm that optimally weights claim factors. The model is then validated against an independent data set and by business users.

An effective model provides output that includes a model score and explanatory messages that offer objective insights and serve as a call to action for claim handlers. Effective predictive models are dynamic, allowing emerging claim or claimant-specific information to be incorporated into the model as key events unfold. Claim models can also be designed to predict a variety of potential outcomes, including claim-loss and expense severity, litigation propensity, and fraud. This provides even more value to the enterprise.

Insurance companies have used predictive modeling to improve the selection and pricing of risks for well over a decade. It has been used by insurers to fundamentally transform the traditional underwriting process, resulting in significant loss ratio improvement. In fact, the success of underwriting predictive models has become so widespread that the technology is now considered a de facto core industry solution. Some would even call it "table stakes."

With claim predictive models less mature in the insurance industry, it may be surprising to learn the technical performance or "predictive power" of early claim models is actually superior to its underwriting brethren. Due in part to a richer, more diverse data set, effective claim models do a better job of prospectively identifying high exposure claims than underwriting models do at pinpointing poor risks. Based on internal data analysis performed by Deloitte Consulting, the following provides a side-by-side comparison of the model performance of an underwriting predictive model versus a claim model designed for comparable companies and lines of business.

Consider the 46 percent result in Table 1 for the projected "worst" underwriting risks. This score indicates the actual loss ratio for the worst 10 percent of risks identified by the underwriting model is 46 percent higher than the average loss ratio (represented by "0″ in the table). Similarly, the 115-percent result indicates the actual claim severity of the worst 10 percent of claims identified by the claim predictive model is more than twice the average closed claim severity. The claim predictive model insights are achieved at FNOL, so claim organizations can effectively recognize high-severity cases and take early action.

Achieving the underwriting segmentation illustrated here is universally considered an impressive result and such tools have yielded significant financial and strategic benefits for insurers. Similarly, the potential financial and operational opportunity relative to a successfully implemented claim predictive model is extraordinary.

Overcoming "Roadblocks"

Despite the rising tide of true believers among claim organizations, some nevertheless remain skeptical about the business value produced by predictive analytics. Two common areas of doubt exist: concern about the integrity of one's own claim data, and concern about whether a claim model is truly needed to recognize high severity bodily injury claims.

First, let's address the data. Many companies question the credibility and usefulness of their claim data because incompatible legacy claim systems and inconsistent data capture has stripped away the analytic value. Fortunately, clean data is not required to build a strong claim predictive model, as model development involves a rigorous cleansing process that sifts through available data and isolates the usable historical claim information. The exercise may also involve text mining of unstructured data, such as claim notes, for predictive characteristics. Finally, added insights are also delivered through publically available external data sources that provide lifestyle and behavioral context to individual claims. The ultimate result is surprisingly strong claim segmentation from seemingly incomplete or "dirty" data.

The second objection typically encountered is the perception that high-severity bodily injury claims—the critical 10 percent as shown in Table 2—are easily recognized, thereby minimizing the advantage of a tool that prospectively assesses claim exposure.

While it is true that bodily injury claims with very serious injuries may be simple to spot, many large losses often lack overt, high-severity characteristics when the claim first hits an adjuster's work queue. In fact, our research indicates that the results of closed file audits performed by experienced claim handlers reveals that approximately 55 percent of high severity cases go unrecognized at FNOL. These cases, which had high adjuster reassignment rates and were often litigated later in the claim cycle, represented more than one third of loss and expense dollars. The conclusion was that earlier recognition of this critical 10-percent subset would have helped mitigate the explosive nature of many of these bodily injury cases.

Converting Insight Into Action

Technology tools, like gym memberships, only pay dividends if you use them. The creation of a claim predictive model is only a launching point to creating business benefits. The tool and the information it provides is intended to enhance (not replace) the critical thinking skills and claim practices required to improve outcomes. In many cases, claim and legal resources may perform the same claim management tasks, but much earlier in the claim life. Presented with model outputs enriched with non-traditional internal and external claim characteristics, vital information is front and center for claim handlers to understand severity drivers, formulate a solid strategy, and accelerate fair resolution. In particular, effective predictive models positively impact several critical decision points along the claims value chain:

Claim routing and claim handler assignment. There's no substitute for getting it right the first time with claim assignment. The unfortunate reality is "right claim, right resource," often takes multiple tries. Based on our research, auto and general liability claims are, in fact, reassigned about one-third of the time, thereby resulting in average loss and expense severity that is almost 3.5 times higher than average.

The well-known pitfalls of claim reassignment—resource inefficiencies, costly learning curves, and disruptions with the claimant or attorney—can be mitigated through accurate upfront claim segmentation. Even in claim operations where routing and assignment are rotational, based on geography or client-dedicated, early warning of a claim's potential explosiveness allows the organization to lend a hand to potentially overmatched resources. The end result is a more systematic matching of a claim to the resources better equipped to quickly and efficiently settle the loss.

Claim settlement strategies. Speed in assessing and settling a claim for fair value signifies success. Yet the lack of complete and reliable FNOL information often stalls the adjuster's ability to evaluate exposure and bring the bodily injury claim to resolution.

Insight about claim severity and litigation propensity, including the data points and reasons behind it, can provide claim and legal resources with a line of sight rarely available through traditional claim management methods. Although the picture may not be complete, the predictive analytics advantage can lead to improved claim investigations, fact-based resolution strategies, and settlement negotiations aided by new data insights.

Fraud management. An assessment of fraud propensity can contribute to more accurate fraud referrals and help deter fraud through early intervention. Traditional fraud management relies on the claim handler for primary detection, which often results in an inconsistent special investigation unit (SIU) referral process and a reactive, scatter-shot approach. Referrals tend to focus on prior claim history and longstanding red flags to identify potential issues.

With an analytical tool providing a consistent, multi-dimensional view to assess fraud propensity, claim organizations can improve the quality of SIU referrals, reduce manual efforts, accelerate referral times, and shift the focus of SIU responsibilities toward fraud deterrence.

Litigation management. Despite insurers' best efforts, litigated bodily injury claims are a reality. Also not surprising is that litigated cases cost an average of eight times more than non-litigated claims, according to internal data analysis. Though many cases are well developed once litigated, new data points related to the legal matter typically boost the model's predictive power. With improved segmentation and enhanced model outputs at first notice of suit, more effective litigation management can be applied. The options in assigning counsel are many—in-house versus outside, firm selection, partner versus associate, and specialized resources—and yet choosing the defense attorney whose skills and experience more closely match the needs of the legal matter are essential to achieving good results at a fair cost.

Typically if defense counsel (staff or outside) is overmatched, then the outcome tends to be undesirable. If one is overqualified, the result may be good but at an excessive cost. Armed with the insight and outputs of an effective litigation model, the legal referral can hit the mark, setting the stage for an accurate assessment of settlement value, the need for experts, and solid resolution strategies. The ultimate result is a better outcome at the appropriate level of expense.

A Final Thought

After more than a decade of observing underwriters using predictive models to improve results, many claim organizations have now embraced the notion that data can tell a very insightful story. By combining statistical analysis with the deep knowledge of claim experts, leading-edge claim operations have made data analytics the engine that powers their claim segmentation strategies. Early and accurate assessment of bodily injury claim severity—and specifically the unrecognized critical 10 percent—delivers a game-changing capability that can help focus resources on explosive cases and accelerate fair claim resolution.

Most importantly, those that have designed and implemented effective claim predictive models are realizing four-to-six percent in annual loss cost savings (Deloitte Consulting Insurance Clients). In a difficult environment of rising combined ratios, effective predictive modeling offers claim operations the opportunity to deliver real, sustainable results to the enterprise. It's clear that the leaders and fast followers are off and running, so will you join the race?

Steve Laudermilch, senior manager at Deloitte LLP, and Jim O'Connor, associate general counsel, Deloitte LLP, contributed to this byline.