From fast tracking claims to spotlighting fraud, predictive analytics continues to provide increasing value to claim operations. Predictive analytics enables companies to develop operational efficiencies, enhance customer service, and increase cash flow by maximizing recovery opportunities.
Simply stated, predictive analytics is a way to project outcomes (with a certain level of probability) based on patterns derived from historical events. A key benefit of this strategy is that it can be a one-size-fits-all application. If there is a business question to be resolved with underlying data, then predictive analytics can provide insights not previously recognized.
While fraud detection and fast tracking claims are currently the most popular applications, predictive modeling can be very useful for subrogation. For example, predictive analytics can help managers provide solutions to minimize subrogation leakage and speed up the recovery process. Before starting, however, managers should determine what functionality predictive modeling will serve. Will it be used to provide real-time guidance from the first notice of loss as a means to improve upfront subrogation reorganization? Or will it be used to uncover errors in underlying processes by identifying missed opportunities? The answer to these questions will shape the way processes should be implemented and data prepared.
Predictive analytics is not a black box that produces perfect predictions all of the time. Ultimately, industry experts must interpret the output to give it value. When used correctly, predictive analytics can help organizations create efficiencies by ordering workflows and targeting particular events or situations.
The core of any analytics solution is the statistical model. Many hours can be invested in developing models. But as any modeler will admit, model development is 80 percent data preparation and 20 percent design. Even more relevant is the ability to implement the output from the models into existing processes and systems as part of a larger strategy.
Data Preparation
When developing predictive models, data preparation is a time-consuming but critical step. Anyone can process a set of data through a model and generate results. It is essential to understand what the models say about the data.
Modelers must be able to deal with incorrect and null data, rein in outliers, and transfer datasets to identify business issues for evaluation. Another challenge is to ensure the output makes good business sense. Subrogation can apply to multiple lines of business: automobile, property, and workers' compensation, to name a few. These losses have many different attributes, such as the types of accidents involved and the data captured. But they also can share attributes, such as legal jurisdiction. For that reason, ensuring that differentiating factors are incorporated in the modeling output is important for model optimization and accuracy.
Examples of differentiating factors within subrogation include state laws regarding negligence and the ability to subrogate for PIP and Med Pay. For instance, if left undifferentiated, a standard rear-end collision loss in a no-fault state like Michigan could receive the same probability score as a similar claim in a pure state like Florida. Ultimately, models produce the best results based on the available data and resulting probabilities of the desired outcome. They do not know the laws or other subrogation-specific information. They may, however, infer the laws and dissect the data accordingly based on the underlying historical results. Unfortunately, these inferences are sometimes difficult to recognize in model output. One way to solve this problem is through preparation and segregation of the data into like populations. Therefore, in addition to ensuring that variables are addressed from a quality standpoint, it is equally important to ensure that the dataset as a whole is separated into like segments.
Strategies and Pitfalls
Before initiating data preparation and the modeling process, one must resolve the question of how the results will be implemented. Models by themselves provide guidance and action items. But it is the operation of the predictive analytic outcomes that provides the real value. In the case of subrogation identification, predictive analytics can be applied as a front-end claim scoring tool or a back-end auditing tool for prioritizing the search for missed subrogation recovery opportunities.
In the front-end application, scoring can be utilized to increase the timeliness of subrogation reorganization and pursuit. Typically, models are developed on retrospective data. In developing a real-time solution, it is important to ensure that data/variables will be available in the real-time environment. The real-time application also allows for claims to be continually scored as information changes on a file. While this may allow a claim to work its way into the process as new information is gathered, it may also continually flag false positives that have already been deemed non-recoverable. In this situation, processes should be established in which claims are re-scored only when certain pieces of information — such as additional accident details — are added to the file.
When predictive analytics is used to review claims post-close, the benefit of timely subrogation recovery is diminished. Used in this fashion, predictive analytics serves more of an audit function for identifying missed opportunities. The effectiveness of technology (from the auditing perspective) can be measured and, more importantly, conclusions can be drawn as to why subrogation is being missed within existing processes.
For instance, perhaps individual adjusters have problems identifying subrogation potential, or issues exist within certain branches or within particular losses. Answers can be derived by using predictive analytics to streamline auditing processes. In this case, for example, data points are not predictors of subrogation recovery potential, but instead provide insight as to why a subrogation opportunity was missed. This information could be added to the data analysis to further improve the accuracy of the auditing process.
Identifying trends and even effectiveness is difficult in a real-time environment. Was it a score that led an individual to identify liability, or would the user have made the same determination without the help of the score? Be it real-time scoring or back-end auditing, the human factor is still a necessary component in the subrogation evaluation process. To that point, the key is to implement processes that require users to take appropriate action and to ensure that the action can be measured and monitored.
Model Revision and Evaluation
Even after the data has been prepped, models developed, and a successful process implemented to ensure action is taken, the job is still not complete. The final aspect of a predictive analytics strategy is model revision. The genius of predictive modeling is that it continually learns from new data. Should patterns change or the underlying data morph, the models are able to capture and adjust to those changes. The challenge is being proactive rather than reactive in this step. Furthermore, the underlying results from the models should be continually validated by the users in the field. If the models are off-base and a number of false positives are generated, information should be fed back and the models reevaluated. After all, in order for the models to be useful, people must believe in them.
That leads to the final point. A predictive analytics strategy should be approached in a team environment. Skill sets and subject matter expertise between the modeling community and subrogation department typically do not overlap. Effective process implementation and model development with these two groups working side by side will eliminate several of the challenges posed in the model revision and the data prep stages.
Predictive analytics for subrogation works. Utilizing this type of solution can maximize subrogation recovery potential, increase the timeliness of subrogation identification, and be used as an auditing tool to fix root causes. The challenge is looking beyond just the score to ensure that the true value of the models can be extracted via an executed predictive modeling strategy that includes the right blend of people and a firm understanding of the operational objective.
Norman McKnight, CSRP, manages the subrogation identification and large deductible reconciliation teams for Paragon's third-party deductible and subrogation practice. He also is an active member of the National Association of Subrogation Professionals. He may be reached at 412-375-6451, norman.mcknight@paragonbenfield.com, www.ParagonBenfield.com.
© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.