Reaping the Financial Rewards of End-to-End Claims Analytics

Because the claim fell in the model's "red zone"—scores representing claims with the highest severity potential for the specific injury group—the company determined that an in-person visit to the claimant's residence was appropriate. The claims manager arranged for an experienced adjuster to visit the claimant's residence without advanced notice. When the adjuster asked if the claimant was home, a family member kindly let the adjuster know the claimant was busy serving customers at the local bar down the street. As the reader can easily imagine, the insured kindly withdrew his claim. This is a high-quality example of how predictive models can help adjusters identify, confirm, and properly address non-meritorious claims activity. A straightforward, yet effective in-person visit, as the result of a high model score, led to the claimant withdrawing the claim. This, in turn, drove the cost to zero.

Another interesting story involved a workers' compensation claim that initially received a below average score. All characteristics of the claim were suggestive of a fairly straightforward low-cost claim. However, as the pharmacy bills were incorporated into the models, the score increased significantly after three months. [2] Upon further investigation and review of the reason messages explaining the major drivers of the high score, the insurer was able to determine the claimant was aggressively re-filling prescriptions—that is, medications consistently refilled prior to due dates, which is indicative of unusual and suspect behavior. As a result of this finding, the insurer put into place a tight pharmacy and medical management plan in order to curtail the duration of drug intake and contain costs overall.

The Role of Data and Science

The ability of claims predictive models to effectively identify the highest risk claims can be attributed to the diversity of internal and external data sources utilized; the number of predictive variables used to develop the predictive models (more than 1,000 variables studied); unique modeling aspects for analysis (claimant, employer, medical and pharmacy bills, and so on); and the mining of unstructured data through text analytics of free form case notes and other reports. 

As Chris Anderson, the former editor-in-chief of Wired Magazine, explained, "perhaps the most important cultural trend today: the explosion of data about every aspect of our world and the rise of applied math gurus who know how to use it." 

In the above examples, the higher-than-average expected claims severity was determined by applied math gurus looking at claims-related data. Some examples of claims-related data include:

• The day and time the injury was reported.
• The accident location and surrounding circumstances.
• Historical workers' compensation claims and claims from other lines of business (a "frequent flyer" indicator).
• Claimant lifestyle data—for example, fitness interest and existing co-morbidities.

Over time, medical and pharmacy data was also incorporated into the models as a way to identify unique treatment patterns, outlier behaviors, and so on. 

Unlike the human mind, which can only process a handful of data points at any given time, mathematical models are capable of detecting and processing hundreds of such complex patterns consistently over time in order to return an unbiased prediction. As Richard Nisbett and Lee Ross noted in their book titled Human Interference: Strategies and Shortcomings of Social Judgment, "human judges are not merely worse than optimal regression equations; they are worse than almost any regression equation."  

Business Implementation

It is important to note the development of claims predictive models is only one piece of an effective analytics journey. Even the most accurate predictive model has minimal business value if it cannot be successfully incorporated into the day-to-day workflow of the insurance company's claims professionals. As such, a carefully crafted business plan, acceptance and buy-in from end-users, and seamless technology integration are crucial components of successful analytics deployment.

One of the hallmarks of success is the presence of a well-defined business implementation plan that dictates how the model will be used, how the model output will be presented to claims professionals, and how the claims process will be enhanced to incorporate the model output and insights. Defining a claims predictive model strategy that identifies how the model will be used is critically important. As such, it is often considered one of the first steps in the predictive modeling journey. Some questions to ponder are:

• Will the model be used to triage claims and direct them to the appropriately skilled resources? If so, then how?
• Will the model be used to indicate when certain claims should be escalated for referrals to the special investigative unit (SIU)? If so, when?
• Will the model be used to identify claims where early medical intervention is appropriate? If "yes," then what are the criteria to be used?
• Will the model be used to determine which claims might require more focused oversight? 

Knowing the answers to questions such as these will help define the focus of the model. Moreover, it will assist the insurance company in selecting the right financial, operational, and model key performance indicators (KPIs) to monitor both pre- and post-production. In addition to the traditional KPIs focused on loss & ALE, fraud & recoveries, claims volume and quality, it is crucial that insurers monitor and measure detailed model performance KPIs, such as score distribution by decile, reason code distribution by decile, claims score migration through time, and so forth.

Effective Communications

When it comes to end-user adoption, obtaining user buy-in through effective communications and leveraging key claims advocates is critical. If the end users feel as if they have a stake in the predictive model roll out, then the company is more likely to realize the financial benefits. To foster a strong sense of ownership, it is important to include key claims professionals early on in the business implementation process. This begins with their participation in the development of the claims predictive model strategy that will dictate how the model will be used and how the model output will be presented to their fellow colleagues. Involving key stakeholders in discussions as to how the model will be used on a day-to-day basis and what process changes might be required to successfully leverage the model output truly imparts a sense of ownership to everyone involved. 

The use of an effective communications plan is also important when rolling out the claims predictive models to the claims community. Common concerns include the perception that the introduction of a claims predictive model will replace or reduce a claims handler's ability to leverage his or her own experience and expert judgment when settling claims. However, this is simply not the case. Rather, a claims predictive model is a tool that enhances the claims professional's ability to perform their job, by allowing them to work more efficiently and make more informed decisions early on in the life of a claim by leveraging 50 to 100 predictive variables. 

Much like eyeglasses help many of us read the small print on a dinner menu, claims predictive models represent "eyeglasses" for the claims adjuster's mind, allowing him or her to read the claims file leveraging a powerful but consistent mathematical equation. Crafting communications that underscore this important point, prior to the deployment of the claims predictive models, can help end users view this as a powerful tool in their tool belt instead of a threat to their jobs or professional judgment.

The introduction of a claims predictive model may necessitate some changes in the claims-handling process, such as the introduction of new claims intake protocols to enhance the collection of data the model needs to run effectively as soon as possible (such as at FNOL). Claims system changes will also be required so that the output of the claims predictive model can be displayed to the claims handler in a user-friendly manner. The integration with technology is of vital importance in presenting the model output to the claims professional in a manner that will enhance their claims-handling abilities.

Technology Drives Action

Last but not least, a successful model implementation requires a well-conceived and implemented technology solution. The predictive power of the model—for example, the ability of the model to segment claims severity by injury group—and return on investment can only be achieved through the use of the right enabling technologies. The scoring solution is a combination of processes that collects both the internal and external data, transforms the data into the appropriate format, calculates the score, generates the reason codes explaining 85 to 90 percent of the score in "English," and feeds the business rules that ultimately drive the appropriate business actions. The scoring operation is essentially the technical manifestation of the claims predictive models, often referred to in the industry as the "scoring engine" or "scorecard."

Leveraging this added analytics intelligence requires the modification of business processes, the re-routing of lost time and medical claims, claims staff screen updates/enhancements, the development of new analytics based KPIs, and other measures. The modifications may translate to enhancing existing infrastructure, all the way to integrating new business rules engines and/or workflow management tools. In the end, the most successful companies seamlessly integrate the analytics results in a non-intrusive way to the end users, and they do so in real time. If you find you are compounding the workload of your professionals, then your technology integration may need to be revised.

The Journey Continues

In the end, a claims predictive model deployment is more than just a bunch of actuaries, statisticians and PhD's having fun with numbers. The end-to-end analytics journey includes:

1. Model deployment strategy.
2. Claims predictive model development (that is, cool mathematical equations).
3. Scoring engine development and technical integration.
4. Business and operational implementation.
5. Organizational change management.
6. Performance management and loopback improvement. 

If the organization can clearly articulate its strategy, develop a well-defined implementation plan, and foster a strong sense of ownership from the end users as discussed in this article, we believe the organization can be well on its way to realizing the financial benefits from using claims predictive models.

Deloitte refers to one or more of Deloitte Touche Tohmatsu Limited, a UK private company limited by guarantee, and its network of member firms, each of which is a legally separate and independent entity. Please see www.deloitte.com/about for a detailed description of the legal structure of Deloitte Touche Tohmatsu Limited and its member firms. Please see www.deloitte.com/us/about for a detailed description of the legal structure of Deloitte LLP and its subsidiaries. Certain services may not be available to attest clients under the rules and regulations of public accounting.

This publication contains general information only and is based on the experiences and research of Deloitte practitioners. Deloitte is not, by means of this publication, rendering business, financial, investment, or other professional advice or services. This publication is not a substitute for such professional advice or services, nor should it be used as a basis for any decision or action that may affect your business. Before making any decision or taking any action that may affect your business, you should consult a qualified professional advisor. Deloitte, its affiliates, and related entities shall not be responsible for any loss sustained by any person who relies on this publication. For related information, visit www.deloitte.com.

Footnotes

[1] See Enhance Workers' Comp Predictive Modeling with Injury Groupings.