Science Project

Claims organizations continue to fight the good fight, constantly balancing loss adjustment expense with controlling claims costs while also attempting to stay focused on the needs of the customer. Complicating matters are a shrinking talent base, challenging regulatory and legal environments, and claims technology that often is inadequate to get the job done. Yet despite this difficult environment, many claims operations have effectively instituted a brand of best practices and worked hard to squeeze out the "claim leakage." Ironically, claims organizations frequently are doing the right things but just not early enough in a claim's life to produce maximum results. To tackle this problem, industry leaders are turning to data analytics as an innovative approach to claims management.

Over the past decade, many insurance companies have utilized predictive models to improve dramatically their risk selection and underwriting results. Based on the potential business impact of prospective claim segmentation, insurers, third-party administrators, and self-insured employers now are turning their sights--and predictive models--on claims.

Claims management has long been more of an art than a science. The art is the expertise and experience claims adjusters develop over time, enabling them to investigate, administer, and settle claims effectively. Adjuster experience builds sharp instincts that influence the actions and intensity applied to a given claim, and yet there often are important insights that sit beyond even the most seasoned adjuster's line of sight. An extraordinary opportunity now exists to take the value associated with adjuster experience and competency and significantly enhance it by applying the science of predictive modeling. The result is a claims organization positioned to achieve breakthrough performance.

In the traditional claims environment, the type of injury sustained often dictates how a case is managed, including the experience level of the adjuster assigned, whether medical resources are engaged, and the general level of attention the claim receives. And although medical diagnosis is essential in assessing claim exposure, experienced claims professionals have long known other medical factors such as co-morbidities and the claimant's behavior also can greatly influence the ultimate outcome. In fact, many times the person, not the injury, matters most. This view is evidenced repeatedly as claimants with severe injuries may recover quickly and return to normal activities while others who suffer relatively minor, uncomplicated injuries experience claim outcomes well beyond average.

The persistent challenge in anticipating high-exposure claims is a claimant's behavior is revealed over the course of the claim. As a result, claims developments that bubble up high-exposure cases often occur too late in the process, leaving resources at a distinct disadvantage to affect a fair and timely outcome.

Typically 20 percent of a claim population accounts for 60 percent or more of loss costs. The objective is to pinpoint this cost-driving segment early in the claim life cycle and interrupt that development. Rooted in science, predictive models have the power to analyze homogeneous-looking claims, such as lower back strains, and differentiate them based on probable outcome. Essentially, a claim predictive model applies a lens to every claim and projects the claim outcome relative to the average result for that particular injury or claim type.

Building a claim predictive model is like putting together a 75-piece jigsaw puzzle. The initial design challenge, however, is that more than 1,000 pieces are analyzed to determine the final set needed to create the ultimate picture. During the build process, data that reflects behavior, such as a claimant's distance from work, family responsibilities, and even the claimant's hobbies and interests, is complemented by medical data (from treatment patterns to co-morbidities, such as diabetes or high blood pressure) to produce the final set of variables that maximize the model's predictive power. The results are impressive: Within days of intake, those claims identified by the model as being the "worst" 10 percent typically have loss costs of well over two times the average claim in a similar injury group. The predictive power stands true in the "best" 10 percent of claims, as well, with claims costs about 50 percent below average.

To build the model, all data elements or claim characteristics undergo univariate (single variable) analysis, which is a statistical technique that assesses the relevance of each data characteristic to the ultimate claim outcome. A second level of analysis (multivariable) then is used to reveal how certain data characteristics relate to each other. For instance, assessing the relationship of two distinct variables--for example, the distance between a claimant's home and work location and prior claims history--provides much greater insight than simply evaluating variables independently. Once the optimal set of variables has been chosen, mathematical weightings are created for all of the data elements and combined into an algorithm that maximizes predictive quality.

The resulting output includes not only a score (typically 1 to 100) but also the support or "reason messages" behind the score. By understanding why a claim was assigned a certain score, claims organizations become empowered with the information they need to drive the most appropriate claims management actions.

Like the claims process itself, the predictive model is dynamic. It both scores claims in real time at the point of intake and also generates scores and reason messages throughout the claim life. As events occur during the course of the claim life cycle, the claim score is refined to reflect the potential change in exposure enabling the adjuster to take action.

Design and development of a robust claim model generates extraordinary business value for any claims organization. Loss cost reductions of five percent or more, increased productivity and efficiency, and improved customer satisfaction are all achievable results. But running claims through a scoring engine yields little impact unless a claims organization effectively "operationalizes" the model output.

The business value of the predictive model is magnified by its versatility. Unlike many off-the-shelf products that are typically focused on one aspect of claims management, more sophisticated claim predictive models impact multiple decision points throughout the life cycle. From initial routing and assignment to claim resolution, predictive model output provides valuable insight. Armed with guidance from the outset the claims professional can bring resolution more quickly and effectively to even the most challenging cases. Outlined below are several components of the claim life cycle improved through the science of predictive modeling:

Initial claim routing. Business rules commonly evaluate claim exposure and determine the claim's destination--a low-touch processing center or an adjuster's desktop. The routing rules, typically based on injury severity, have mixed results evidenced by relatively high reassignment rates. Through more advanced claim segmentation at intake, predictive model output bolsters routing accuracy and gets claims to the right place more often.

Claims assignment. A claims talent crisis is looming with a projected shortage of more than 85,000 adjusters by 2012. With the number of expert adjusters dwindling, initial assignment of claims to the right resource is more important than ever. By better understanding a claim's true exposure, explosive cases are quickly directed to the most qualified adjusters while low-exposure claims are channeled to less experienced resources or auto-adjudication.

Fraud detection. Many studies have estimated that fraud, including the exaggeration of a claim, can represent more than 20 percent of claim payments. Stemming the tide requires a combination of speed and focused effort. Many claims organizations take, on average, several months before identifying the existence of fraud and engaging specialty resources to combat the problem. Through predictive model output--both the score and the reasons behind it--the propensity for fraud is promptly detected, allowing the right resources to be engaged and take the right actions more quickly. This proactive approach not only mitigates adverse claims development but can deter it altogether.

Medical case management. Various approaches are used to identify claims that warrant the intervention of medical resources, yet claims organizations typically refer claims based almost exclusively on diagnosis. But injury type does not always equate to the need for medical management. By augmenting the diagnosis-based approach with additional dimensions that indicate medical complexity and behavior-based issues, a truer measure of exposure is determined. The result is the right mix of claims being identified and immediately assigned to medical case management resources.

Breakthrough performance is achievable by accelerating the claims time line. Science, through data analytics, is the impetus. The moment a claim is entered, a prospective view becomes available enabling swift and decisive action. Ultimately, the proper resources, armed with better information and insights, are matched to the appropriate claim exposure.

The results are very real:

o Loss cost reductions of three to five percent or more.

o Reduced loss adjustment expense.

o Improved efficiency.

o Increased adjuster productivity.

o Increased customer satisfaction.

While it's true the claims sector has long been a discipline where hindsight exposes missed signals and lost opportunities, the acceleration of the claim life cycle through predictive modeling can yield exceptional results. The most progressive claims organizations already have designed and implemented sophisticated predictive models, and others are close behind. After years of incremental gains, the world of claims is on the verge of leaping forward. The paradigm is shifting, and science is leading the way.