Claims and Predictive Modeling: Where Art Meets Science
By Adam Wesson
The use of predictive analytics in claims is becoming increasingly common due to a number of compelling factors. Pressures on insurers, particularly in the area of workers' compensation, have spurred closer attention to claims costs. In another trend, a recent Towers Watson survey found one of the greatest challenges facing the claims operations in the property/casualty industry is recruitment and retention of top talent, followed closely by effective implementation of data capture, metrics, and analytics.1
In an environment where twenty-plus years of experience for claims adjusters may be becoming rare—and the effective use of data remains a complex skill—can leveraging technology help fill the potential knowledge gap that may arise out of waning claims experience? And how can a company effectively accomplish this?
In considering the cream of the crop—experienced and high-performing claims adjusters—the expertise and finesse they typically bring to achieving great claims outcomes likely comes to mind. From their keen insight grounded in years of experience to their ability to accurately predict claim costs and consistently achieve good outcomes, there’s a special art to mastering the role of claims handling. Experienced claims handlers may rightfully be seen as the most important asset a claims department possesses. As claims adjuster experience is possibly dwindling, predictive analytics usually becomes more critical as a tool to educate less experienced adjusters and assist them in learning the claims-handling process.
Analytics (also commonly known as big data or data science) refers generally to a variety of mathematical and statistical techniques for extracting information and value from historical data with the goal of improved decision making. Analytic techniques cover a broad spectrum, from queries and reports that examine the past claims experience to scoring and forecasting methods that can predict the future. This can provide significant value to insurers facing workers' compensation claims through improvements to the efficiency and effectiveness of their claims-handling processes.
What can happen if the deepest claims experience (art) is combined with predictive analytics (science)? Can an insurer benefit from having the best of both worlds? More insurers are finding that such a combination is possible. Predictive analytics can provide a decision support tool and a more timely way to alert adjusters to potential issues, enabling earlier and more informed intervention. The result can be mitigation of unnecessary costs and potentially better outcomes. Such a combination also offers a safety net that can focus the attention of adjusters and managers where it’s most needed—on claims of higher severity.
Successfully applying a predictive model requires development, implementation, and maintenance of the model. This calls for a team effort, involving data modelers, product experts, information technology departments, and experienced claims staff. Not only must a good model yield accurate results, it needs to provide actionable insights. It also must fit an insurer’s operational needs, possess scalability for growing data sophistication, and be reasonable to implement and regularly refresh with current data. That’s a pretty tall order, which can be daunting even for the most sophisticated insurers. The outcome, however, can be a solution that pays dividends year after year—and one that only grows stronger as it’s nurtured by the organization.
So, where do you start? Initially, any insurer should determine its needs. What problem cries out to be solved? Next, ask why a solution is being sought, which helps to establish that the appropriate issue is being addressed. Once those steps have been taken, the data’s availability and reliability can be assessed. Other criteria that need to be considered include: What type of data is needed for the model? Does an organization have the proper scale and data integrity to create a reliable model? Is the necessary expertise available? If an insurer lacks sufficient data quantity or quality, are viable solutions available within the market to fulfill its needs? As with any large project, defining the problem and breaking down the necessary components to solve that problem should lead to the best solution.
After passing the hurdles of defining needs and assessing resources—that may result in either building or procuring a model—it becomes time to focus on implementation. Even the most detailed and accurate model will not produce stellar results if it’s not effectively built into an insurer’s business process. For claims operations, this can be particularly challenging, and it’s where a number of insurers can stumble. Overly complex models that require user input to work effectively can meet considerable resistance from claims adjusters who typically already have a number of systems to log in to daily. Manual data entry is also prone to errors and can affect model accuracy. The simplest solution is a model that uses either data captured within codified fields or text mining to obtain data on the back end, which can significantly lower the need for user interaction. That allows an insurer simply to focus on incorporating the output into its process for claims decisions.
When devising the claims process, it’s important to think back to the original problem statement: What problem needs to be solved? The process should include a review of modeling results with this specific issue in mind. For example, if an insurer is attempting to mitigate loss costs, adjusters should consider predictive output when developing action plans and reviewing components such as reserves and claims disposition. It’s important for claims management to ensure this is taking place consistently and effectively and to provide training as necessary. Adherence to the process won’t change overnight, especially without consistent monitoring and reinforcement.
Every day, more insurers are putting their data to good use in the claims department by using predictive modeling to help adjusters and managers identify potentially high-cost claims and focus their attention on improving outcomes. By taking a holistic approach to model development and implementation, chances of successfully achieving such goals are greatly improved. When insurers continually update and improve their models—and possibly add new models as new opportunities appear—they can transform their organization from reactive to strategic, adding a critical advantage over the competition.
- Ramsay, F., & Stoll, B. (2015, August 1). Hopes and Hurdles of Tech-Driven Claim Operations: 2015 Property & Casualty Insurance Claim Officer Survey. Retrieved August 25, 2015.