Claims Indemnity Leakage--Estimated at an annual rate between six to 10 percent of NWP.

Customer Satisfaction--Carriers achieving high levels of satisfaction retain customers and enjoy lower customer acquisition costs. Among customers who indicate high levels of satisfaction with their carrier overall, 65 percent said they "definitely would" renew their policy with their auto insurance provider. Conversely, only 43 percent of customers who report low levels of satisfaction said they would definitely renew their policy. Claim service can dramatically impact customer satisfaction driving 44 percent of the overall insurer impression by customers who filed a recent auto claim.

Combined Ratios

As combined ratios remain high and investment portfolios continue to yield low returns, the need to drive maximum efficiency in the claims handling organization becomes much more important. As has always been true, the adjuster is the first line of defense.

Imagine if you could have your best adjusters handling each of your claims and that those adjusters were guaranteed to be with you for years to come. Each of the key concerns listed above would be greatly reduced. Your best adjusters by definition are highly efficient, do not make inappropriate claim or expense payments and deliver the highest levels of customer satisfaction.

Unfortunately it isn't possible to have your best adjusters assigned to every claim. Not only are carriers losing their top talent each year due to the normal employee transitional flow of a free market society, but the majority of experienced adjusters are heading for retirement--70 percent are over the age of 45 and 33 percent are over the age of 55.

To make this situation worse, new talent is not joining the ranks of carrier's claims departments--only four percent are under the age of 35. In fact, Deloitte Consulting predicts a shortage 84,000 claims adjusters in the United States by 2014. It becomes necessary to preserve these highly experienced and talented claims handling resources for the claims (or pieces of the claim, such as features or individual coverages) that require their extensive expertise, where they can have the most impact while having less experienced resources handle more routine tasks. Carriers, however, can struggle to consistently separate those claims that can benefit from early intervention of expert adjusting resources from the low risk claims. In addition, too often certain claims remain under the radar until they morph in to the complex or high cost entities that they really are.

Helping Hand

The adjuster needs a hand. What is needed is a fundamentally more advanced, more sophisticated approach--one that combines modern predictive-analytics and data-driven claims workflow solutions.

Predictive analytics is a branch of advanced statistical practice that specializes in data pattern recognition. Predictive analytics examines the data elements surrounding a specific known event and then examines the historical changes in those data elements to determine when and how they changed in a statistically significant way.

Models are then built that accurately and consistently identify unknown events that fit the pattern, allowing prevention and/or mitigation techniques to prevent or lessen the impact of the predicted event. Statistical predictive modeling can automatically detect extremely complex, subtle patterns in the data that humans alone cannot. Predictive models can also adapt over time to changing conditions.

Two Decades of Service

Predictive analytics has been applied effectively within the financial services community for over 20 years. One of the best known applications is around credit card fraud identification where combined data driven and technology supported efforts in the credit card industry has shown very impressive results with models screening 85 percent of U.S. credit card transactions for fraud, resulting in a 50 percent reduction in industry losses. There is not a credit card company operating today that would want to risk doing business without predictive-modeling-based fraud identification tools.

Property/casualty insurers also routinely use predictive analytics in their underwriting organizations. One key area is in rate segmentation. P&C carriers have also started to use predictive analytics in their claims areas with most efforts to date focused on claim fraud with impressive results. Deloitte Consulting reports, "Although these tools need to be used carefully and fairly, they can have a significant positive impact.... (companies) typically achieve reductions in claims costs of 3 percent to 5 percent and some attain savings as much as 5 percent to 10 percent."

But the application of predictive modeling in P&C claims has not yet realized its full potential to transform the way all claims are handled. The application of predictive modeling to the claims process more broadly offers tremendous potential to improve outcomes for an insurer's entire book of claims.

Data and Workflow

However applying predictive models to real world claims processing is not without challenges.

Two key challenges are around data and workflow.

Essential to highly effective predictive models is the scope of data included. Predictive models benefit from lots of data. Since they can evaluate complex data patterns, generally the more disparate data that can be brought to bear on a business issue, the more accurate the decisioning can potentially be. A fundamental quandary in claims handling is that the more data that is made available to the decision making process, the better decisions that can be made; but humans become quickly overwhelmed by more and more data.

The solution to this quandary is not to restrict the amount of data used to examine the business issue but rather to automatically analyze these large quantities of data into results that can be acted upon by humans. This is what predictive modeling does.

The real opportunity to maximize the impact of predictive modeling lies in inclusion of more data. In addition to a carrier's own data, predictive models that incorporate public records data and other sources of data external to the insurance company can be responsive to conditions relevant to insurance claims but external to the normal data collected by insurance companies. For example, tapping into claims and policy data across companies gives the models a broader perspective of the claimant or provider's history. Today the opportunity exists to leverage sources of data not previously available to predictive models for claims handling.

But as important as the quantity of data presented to predictive models is the quality of the data. High quality data is an absolute necessity to obtain the maximum results from any predictive modeling effort. Missing, poor or inconsistent data has historically hampered the success of predictive modeling efforts in the P&C claims space, slowed the progress of individual carrier's internal efforts and greatly increased their costs.

Another difficulty with working with insurance claims is that much of the most valuable data resides in the freeform text of the claim notes or loss description fields. Traditionally this data has been difficult for automated systems to utilize. Techniques like text mining and freeform text searching overcome this limitation. Text mining uses statistical methods to extract "meaning" from freeform text and present it to predictive models in a way that it can be processed automatically.

Good Models

But having well-designed predictive models and lots of data only solves part of the problem. To obtain the full benefits of a predictive modeling solution, results must be actionable and must be delivered to the resource best able to influence the predicted result in time to exert appropriate influence. The results generated by the analytic system need to drive a change in business processes to be fully effective, and that means getting the right information to the right person at the right time.

The best way to ensure that the full value of predictive modeling is realized is to seamlessly integrate the output of the model into a robust, modern and flexible claims decision support application. The way this works in practice is that model results trigger automated alert guidance to the adjuster--or even initiate the automated transfer of the claim file to more specialized claims handling resources. It is important to incorporate the analytic result into the user's workflow as seamlessly as possibly. The less the adjuster is forced to deviate from the normal claims handling process the better.

Because claims handling is a dynamic process, with each claim evolving and changing over its life, the analytic process must be continuous and dynamic as well, with each claim rescored every time data changes. Any change in score initiates further adjustment and refinement of the handling of the specific claim. The goal is that each individual claim would be handled in the optimal manner--with low risk claims paid quickly (thus achieving best opportunities for reduced indemnity and loss adjustment expenses and increasing customer satisfaction) and more complex claims or features handled by the adjusters best equipped and trained to handle them.

One complex and high impact example of where the potential of such a system can be clearly seen is in the application of modern, integrated predictive modeling and workflow solutions to the early recognition and closer management of high severity auto injury claims. Though these claims may not be the most frequent type of auto injury claim, they are very important because of their disproportionate cost to the company. And, though BI and PIP frequency has been falling in recent years, medical severity has been rising dramatically, which makes addressing this problem even more critical.
This is an area of value where Mitchell and LexisNexis Risk Solutions have combined efforts to assist insurers. Mitchell has high quality data within its DecisionPoint solution, LexisNexis has predictive modeling expertise and extensive public records and other specialized databases and Mitchell DecisionPoint has the Sentry rules workflow engine--the three essential pieces for a successful predictive-analytics-driven claims optimization solution.

The solution provides insurers:

o Early identification within Mitchell DecisionPoint of high severity medical claims

o Immediate routing of identified claims to specialized claims resources via Mitchell Sentry rules engine

o Ability to make maximum efficient and cost effective use of major injury units; nurse case management, medical specialty networks and other severe injury care management skill sets

o Increased opportunities for straight through processing of low severity cases

o Improved opportunity to achieve early ultimate loss reserve accuracy as well as early loss adjustment expense and indemnity control via immediate engagement of appropriate claims handling resources

Catastrophic Injuries

Catastrophically severe injuries, such as significant burns or spinal cord injury are easy to identify as having high medical exposure from the start and so reserves for such claims are usually initially set appropriately high along with a highly detailed claims handling plan executed by top claims handling experts. Conversely there are injury claims where the vehicle damage does not support high medical exposure, and so these claims are also generally reserved and handled correctly.

But in some cases the nature and extent of the injury is more ambiguous. Sometimes these claims progress for some time before the true extent of the medical exposure is recognized and accounted for. Generally there are two reasons for unrecognized excessive medical exposure. In some cases the indications that the claim is more severe than initially apparent are very subtle and easy to miss until later in the claim. In other cases evidence of increasing exposure may be evident but not recognized leading the claim to not be handled appropriately.

For example, the fact that a simple claim has no activity for a period of time is often not a good sign. It may mean that the claimant has engaged an attorney and is accumulating bills. However in the press of day to day claims adjusting the adjuster may have a tendency to not notice this silence and ignore such a claim for too long. In both cases applying predictive modeling to the data can properly identify these claims and help to make sure they get the attention (and reserves) that they need.

In work with several clients Mitchell and LexisNexis Risk Solutions have demonstrated potential gains in applying a predictive-modeling and data-driven approach to the identification of unidentified high medical exposure claims. Models were first built on several years of historic data. This data included the carriers' own claim and policy data together with medical bill data and other external data such as public records data. The models were designed to identify the trajectory of a claim from the data present early in the claim. When tested against actual data, these models showed the ability to identify 60 percent of the total medical indemnity for the group of test claims 12 months later within the top 10 percent of highest scoring claims. These results were achieved within 10 days of FNOL. In other words, these models were able to predict with uncanny accuracy what the exposure of the claims would be 12 months later.

In an operational situation, the predictions generated by the model are compared to the current reserve and claim handling plan early in the life of the claim. If the claim appears to be adequately reserved and handled at the appropriate claims resource level, alerting is suppressed. However if there seemed to be a discrepancy between the existing reserve and the projected exposure and/or the claims resources assigned, the claim is brought to the attention of the adjuster or manager (or both). The claims continue to be rescored throughout the life of the claim so that as conditions change, the models are able to alert the adjuster at the earliest possible moment that the data suggests an unexpected trend.

Summary Judgment
Addressing medical exposure issues is only one example of new opportunities to apply modern predictive-analytics and data-driven claims workflow solutions. Application to claims fraud still offers tremendous opportunity as does the recovery area and, of course, further striation of the book of claims by claims special needs (or lack of special needs) to obtain maximum efficiency and effectiveness and value from your claims professional team.

Greater application of advanced predictive analytics, expanded access to robust data sources and tight integration with workflow offer the opportunity to revolutionize the claims handling process. By generating precise results that allow for each claim to be directed throughout its lifecycle in the most optimal manner, a predictive modeling and data-driven approach to claims handling leverages the claim handler and specialist skills in the most efficient manner, realizing the potential for significant reductions in loss and loss adjustment expense.

(Mike Mahoney is senior director of product marketing for Mitchell International, Inc. and

John Lorimer is vice president product development for LexisNexis Insurance Claims Solutions.)