Challenging economic conditions, reduced returns on investment,in-creases in large claims, and increased competition have allcombined to drive approximately one-third of U.S. insurers out ofthe market over the last 15 years. Reductions in top-line revenueconcurrent with shrinking profitability have had their impact.Today, fewer, generally speaking larger, and definitely moreintensely focused insurers are vying for a larger share of anincreasingly well-informed and demanding market, one that hasshifted demographically, behaviorally, and intellectually.

It has been said that the insurance industry is rich in data andpoor in information. Given the constantly evolving market dynamicsand competitive landscape, a strategic imperative for surviving isthe translation of this wealth of internal and externally availabledata into meaningful decision-making criteria. Being a servicesindustry, insurers know profitability rests on understandingconsumer expectations, effectively delivering products that meetthese expectations, efficiently servicing the resulting customersin a manner consistent with their desired methods and timelines,and profitably managing the risk pools.

While many companies have attempted to cut themselves toprofitability by way of expense reductions, the real leverage inthis industry is on the marketing—or revenue—and claims sides.Unlike the third major component, underwriting, which has longbased its activities upon predicting future events based onmultiple discrete variables, both marketing and claims are recententrants into the predictive world of analytics. For the most part,the focus with these areas has been a more retrospective trendingand projection process until recently, when the more advancedcompanies have realized the potential and started investing in thepredictive aspects of analytics.

Customer segmentation and differentiation, selective leadgeneration, agent productivity, and lifetime value of customers allrepresent areas of rapid progress within the marketing world ofpredictive analytics, generating revenue from better understandingof how to profitably grow specific market shares. While theseopportunities are significant and worthwhile, it can be argued thata quicker path to profitability rests with operationalizingimprovements in claims practices and processes.

By leveraging the wealth of information available on the claimsfront, even relatively small one percent improvements in losses canresult in significant increases in profitability as well asreleasing reserves for improved investment returns. Theseimprovements are achievable on existing business with relativelyquick time to implementation. Given market and shareholderpressures, the opportunity seems both timely and material.Companies would be well served to invest in improving the use ofanalytics within their claims operations in order to realize thesebenefits.

How are predictive analytics any different from thelong-standing and well-developed practices of claims analysis? Ineffect, predictive analytics bring the added dimension of adaptivelearning from experience to the modeling process, which contraststo the more traditional trending or univariate analyses. Instead ofusing linear expressions of probability based upon reviewinghistorical data to specify risk identifying parameters, properlyconstructed predictive models massage the data in search ofgeneralizations that identify common patterns associated withhigher risks. The patterns are discovered usually via a processknown as data mining, which is a technique for working with broaddescriptive attributes of a claim to find correlations.

Here is where predictive models differentiate themselves themost, by continuously undergoing the equivalent of a learningprocess that integrates newly found correlations regardless ofwhether or not there is a causal connection. For example, thereview of a large pool of claims may indicate that a specific andnarrow age range at a particular ZIP code has a higher propensityto litigate even the smallest of claims. This information can beused as a pricing consideration, to determine initial reserves, aswell as fine-tune the process so that these claims are fast-trackedfor legal review. The relationship between age, ZIP code, and risk,or the correlation, is integrated into the model even though beingthat age and in that ZIP code may not necessarily cause higherlitigation—there is a relationship even if it is not causal.

By leveraging advanced mathematical techniques like logisticsregression analysis, decision trees, Naïve Bayes, and neuralnetworks, predictive models are able to handle a much larger numberof variables. They are able to transform seemingly random pieces ofdata into decision criteria that aids in identifying how to bestdistribute and process a new claim differently than would otherwisehave been done. As a result, the overall business value ofsuccessfully implementing predictive analytics in a claimsoperation is wide-ranging. Some of the most common benefitsinclude:

• More accurately priced risks at a finer level ofdistinction;

• Closer management of case reserves, freeing upcapital and increasing investment income;

• Lower claims-handling, supplier, and investigationcosts;

• Less claims leakage;

• More extensive insights into claims unitperformance;

• Greater claims processing efficiency, especially inthe areas of predicted fraud, subrogation, and litigation;

• Faster transaction cycle time resulting in improvedlevels of customer satisfaction.

Consistent with this wide range of possible benefits, recentsurveys of business and technical staff have indicated that ROI onpredictive analytics projects exceed 20 percent on over 75 percentof the best projects and exceed 20 percent on over 30 percent ofthe worst projects, and in all cases less than 10 percent ofrespondents claiming no ROI. According to an IDC report, the medianROI for predictive analytics initiatives is 145 percent compared toan ROI median of 89 percent for non-predictive businessintelligence initiatives. Similarly, according to a white paper bythe Aberdeen Group entitled “Predictive Analytics: The Right Toolfor Tough Times””

“Users of predictive analytics…have achieved a one percentimprovement in operating profit margins over the last year, and ayear over year increase in customer retention of six percent.Survey respondents that have not yet adopted predictivetechnologies experienced a two percent decline in profit margins,and a one percent drop in their customer retention rate.”

The value to a company, especially in today's market, should beapparent. Which begs the question, what are some of the more commonareas where predictive analytics is able to make a differencewithin a claims operation?

Not surprisingly, one of the most common and impactful areasusing analytics is identifying fraud, estimated by the InsuranceInformation Institute to cost insurers over $30 billion annually.Claims fraud adds approximately 10 percent to loss and lossadjustment expenses.

The main challenge is in identifying the true instances offraudulent activity among the millions of claims filed each year—averitable search for a needle in a haystack. This effort is mademore difficult by the fact that besides the typical types of fraud,there are the more complex issues of staged accidents and paddingor inflating the value of a claim that occur. In fact, according tothe Insurance Research Council, up to one in five claims receivedby a company may appear to be fraudulent and require investigation,with padding occurring at a 2:1 frequency over stagedaccidents.

By applying predictive analytics, a company can improve theprobability of identifying fraudulent claims for review. Thisimprovement translates directly into a more efficient use ofresources, a reduction in false positives (or Type I errors), andfewer fraudulent cases missed (or Type II errors). Type I errorsrepresent wasted resources as well as customer alienation aslegitimate claims are investigated. Type II errors result in higherclaims costs which translate into increased premiums and an erodingof competitiveness. Both put the company at a disadvantage.

Advances in the use of automobile fraud detection specificallyhave improved the fraud detection capacity of companies by as muchas 6.5 times. As a result, the more forward-thinking companies areleveraging data mining, pattern recognition, data visualization,and risk scoring to better identify high potential fraudulentcases.

Resource and claims prioritization represent another area ofpotential for analytics, and comes to fruition via the process ofevaluating best next steps based upon specific attributes of theclaim. One of the costs of loss adjustment relates to theexpediency and type of resources deployed at time of first noticeof loss. Proper determination of severity and the relevance ofexpertise and timeliness assist in selecting what resources todeploy when, whether it is an immediate dispatch of a fieldadjuster or simply the acceptance of the filed claim withsupporting documentation. In addition, an analysis of prior claimsresults across providers and suppliers can result in presentingleast cost alternatives given the characteristics and location ofthe claim. Alternatively, as part of the claims evaluation processpredictive analytics can identify claims that likely will besettled at a higher value and set them as high-priority forinternal handling, while the lower-priority claims can beoutsourced, put lower in the queue, or sent to a lower-cost skillset for handling.

As a company learns more detail about its claims payments, thatinformation can be used upstream to more accurately market, price,and underwrite coverage. Behavioral information and customerproclivities as well as utilization statistics can be used toindividualize and target products to specific market segments,identifying unique needs and offering features that address thoseneeds. The claims segments and use patterns discovered by datamining and validated by the development of a predictive model canalso offer pricing differentiations to fine-tune the profitability,competitiveness, and performance of a plan, providing the actuarieswith insights at a more granular level than the typical frequencyand severity tables.

From an underwriting perspective, greater awareness of thecharacteristics that may lead to increased claims costs or fraudprovide a framework for enhancing the risk review process, reducingexposure where possible while capturing sufficient information toallow for effective management of the risk. In all cases,predictive analytics must first be incorporated from an enterpriseand not department level so the marketing, pricing, underwriting,claims cycle is a closed loop providing continuous analyticalfeedback throughout. This closed loop approach, as opposed to thetypical departmental view, allows a company to optimize itsprofitability across the entire product lifecycle.

Despite the many benefits offered by predictive analytics, thereare a number of hurdles that have to be overcome in order tobenefit from effective application of the tools. In a recent surveyof insurance executives, three of the surprisingly most prominenthurdles were executive sponsorship, adequate ROI, and sufficientprioritization. These are organizational as opposed to operationalhurdles, and should be addressed directly as part of the managementteam's strategic planning sessions. The breadth of impact andstrategic insights offered by a predictive analytics solutionwarrants top down support and direction setting; it is not aproject to be undertaken under the radar or as a sample test in asmall corner of the business.

The low profile, low risk approaches do not provide a foundationfor the leveraged enterprise-wide potential predictive analyticscan realize. As for ROI, this was a surprise response thatindicates the need for better awareness, education, andcommunication, as the returns on analytic projects are bothsignificant and rapid for the companies that have engaged in theefforts. Priorities and resources are always a challenge in today'sbusy times; however, the leveraged impact on profits that aneffective program offers should more than offset the need to eithersupplement or reassign the necessary dedicated resources tosuccessfully implement.

Beyond the organizational hurdles, there are also operationalissues that most companies face once they decide to implement apredictive analytics solution. The most common, and most difficultto overcome, ones found across the industry typically involve thefollowing:

• Data quality is typically poor and represents one ofthe hardest issues to overcome, as the scrubbing process mustidentify and locate missing or incomplete data;

• Models requires a holistic view of the data,difficult to compile across distinct and separate legacysystems;

• Integrating with real-time transactions versushistorical data can add complications to implementation;

• Predictive models are sensitive to changes inunderlying parameters and therefore must be continuously updatedand validated against current data;

• Finding the right talent able to integrate thefunctionality of predictive analytics into an operation isdifficult, as it requires a broad view of the organization as wellas a transactional and analytical perspective;

• Organizational resistance is often a challenge asnew practices around prioritization and decision-making areinstalled;

• Transforming analytics into action requires formalprocedures and standards that are routinie within theoperation;

• Moving beyond point solutions to comprehensivedecision-informing claims analysis requires a strategic perspectiveof the data enterprise-wide starting at new business.

Even given these inherent challenges, the potential forsignificant returns on the use of predictive analytics thatdirectly translate into greater market competitiveness seem towarrant the investment. Companies that do not attempt to leveragethis technology as part of their operation may find themselves at aprice and transaction cycle disadvantage.

Famed author of the German play Faust Johann Wolfgang von Goetheonce stated: “Knowing is not enough; we must act.” This isparticularly applicable to the field of analytics, whereretrospective business intelligence like metrics, reports, anddashboards inform but do not necessarily drive action.

Predictive analytics, on the other hand, deal with future eventsand prescribe the actions to be taken—it is actionable analysis.What priority to give a claim, whether or not to audit orinvestigate, what resources to assign and when to assign them, areall actionable decisions informed by an effectively deployedpredictive analytics solution for claims. Yet beyond the specificactions taken, well developed models also offer insights intocustomer behaviors, product performance, and market risks. Theseinsights can then be factored into broader strategic decisions thatdirectionally impact a company's efforts.

Beyond the immediate returns of reduced fraud, increasedefficiency, released investment capital, and faster service lay thepotential to take a holistic view of how risk is being managedenterprise-wide from product pricing through underwriting to theclaims transaction. Companies that leverage the insights that comewith this perspective will find themselves better equipped toeffectively compete in today's challenging environment.