Predictive Analytics – a “Crash” Course for Improving Your Subrogation Results

By now, everyone in the Insurance industry has heard about “predictive analytics” (P.A.). It is rare to attend a meeting without hearing someone say “predictive analytics can help with that”. While most people have a general understanding of the term, not everyone knows how P.A. works and how vital a role it plays in today’s subrogation world.

To level-set, let’s start with the definitions of a few key terms:

• Predictive Analytics - The use of data, statistical algorithms and machine learning to identify the likelihood of future outcomes based on historical data
• Machine Learning - The use and development of computer systems that are able to learn and adapt without following explicit instructions by using algorithms and statistical models to analyze and draw inferences from patterns in data
• Data Mining – The practice of analyzing large databases in order to generate new information
• Predictive Modeling – A commonly used statistical technique to predict future behavior which works by analyzing historical and current data generating a model to help predict future outcomes
• Topic Modeling –Date Mining claim notes looking for topics (groups of terms such as those related to  subrogation), finding patterns in the data using built-in search terms and text analytics

Now we can outline how the processes are created, developed and utilized along with some of the benefits gained by incorporating them into a subrogation program.

Where to Begin?

Installing a P.A. engine can be accomplished in one of three ways: (1) build it from scratch; (2) license or purchase an existing model from a vendor/consultant; or (3) a combination of both. Regardless of which option you choose, some basic information is required to get started.
First, historical data on both successful and unsuccessful recovery outcomes must be gathered to create the baseline of files that are normally referred or not referred for subrogation pursuit. The full dataset can be broken into training data (used to build the model) and testing data (used to quantify performance). Other factors, such as state laws and line of business particulars, can be added to establish a Referral Score (the files most likely to be sent to the recovery unit for pursuit) and/or a Recovery Score (the files most likely to result in an actual recovery).

What actually happens?

Once the platform is built and testing produces the desired results, the analytics engine is ready to be put into action. On a daily basis the structured data (claim records, policy records) is analyzed. The text mining of unstructured data (adjuster claim notes) occurs simultaneously to identify the best recovery claims. Predictive and Topic Modeling helps group terms to find common topics and allows for machine learning. Business Rules (client-specific, loss-scenario specific) can also be added to further refine/amplify results.
Once the inputted data processing finishes its run, routing decisions are made and files are placed into one of three categories - Yes, Maybe and No. The files with the highest scores are routed directly to recovery specialists. The files tagged “Maybe” can be triaged to determine if further pursuit is warranted. The files scoring on the “No” scale can be placed into the archives without any human interaction.

What are the Benefits?

Some of the benefits of an effective P.A. engine include, but are not limited to:

• Files that likely will provide recoveries are identified quickly; speeding up the receipt of funds
• Time-savings: Time is not spent on files with no potential/low potential; freeing up resources
• Expense savings: avoid spending money on investigations/experts for files that are non-viable
• Analysts only receive files with the highest recovery potential, which will create efficiencies and increase the likelihood that assigned files will result in a recovery
• Triage will review and investigate the files classified as “Maybe” and either close those with no recovery potential or route the files with recovery potential to an analyst
• Enhanced internal (Underwriting, Claims) and external (insureds/clients) customer relations
• Peace of mind: Since every file is  scored there is peace of mind knowing that nothing is overlooked – there is no reliance on any sort of manual referral process
• If facts change, an archived file can be rescored and potentially opened to pursue recovery
• MORE $$ IS RECOVERED

Conclusion

Predictive analytics and its extensive capabilities are here to stay. Don’t be caught with antiquated identification processes. Failure to embrace the potential of P.A. will lead to missed opportunities and inefficient use of ever-shrinking resources.
To learn more about how P.A. can help you and your team, contact Dan D’Imperio.