The first step to combating insurance claims fraud is detecting it
Social network analysis is a new weapon in the SAS arsenal
It’s Friday evening rush hour in a downtown parking garage. Dave Jones, tired after a long week, is on his cell phone with his wife when the car in front of him brakes suddenly, causing Dave to crash into it. There’s very little damage, so the police aren’t called. Insurance information is exchanged, and everyone goes their separate ways.
Unfortunately, the story doesn’t end there. The driver and his four passengers all report neck injuries and file claims with Dave’s insurance company. Dave was the victim of an auto accident scam known as the “panic stop,” and if the fraud goes undetected, his insurance company will be on the hook for all damages and injuries reported.
Scams like this are all too common. Fraud costs the property and casualty insurance industry an estimated $30 billion a year in the US alone, and the impact is enormous. Fraud losses weaken an insurer’s financial position and lead to higher premiums for policyholders. Everyone pays the price.
A four-prong approach to fighting fraud
Thornton May in his book The New Know describes what he calls the three Knows:
1) Things we know we know.
2) Things we know we don’t know.
3) Things we don’t know we don’t know.
You could add a fourth: Who we know.
Based on years of experience working with insurers to fight fraud, SAS has developed technologies such as social network analysis into the SAS Fraud Framework for Insurance. This solution tackles each of these four categories of “knows.”
The scams we know
In insurance fraud, the first step is stopping the scams you know about by using business rules to automatically flag suspicious claims. For example, a business rule might target a claim for closer inspection if it exceeds a certain dollar amount, involves a rental vehicle, has no witnesses or police report, or shows excessive personal injury or property damage for the nature of the incident. These rule systems are a good first line of defense, but they can’t detect the scams you don’t know about.
The scams we don’t know
Fraudsters easily learn the rules and work around them, so to detect new schemes, the SAS Fraud Framework includes predictive models. These models use past fraud events to produce fraud-propensity scores. Using algorithms such as neural networks, regressions or decision trees, the system defines those elements of a claim, policy, claimant or supplier that are correlated to fraudulent claims. Adjusters simply enter data, and claims are automatically scored for their likelihood to be fraudulent and made available for review. Using predictive modeling makes it possible to reduce false positives and capture more fraud.
The scams that we don’t yet know that we don’t know
This is a more difficult category, but with the SAS Fraud Framework for Insurance, anomaly detection uncovers new fraud scams by finding those elements that vary from normal. Key performance indicators associated with tasks or events are baselined and thresholds set. When a threshold for a particular measure is exceeded, then the event is reported. Outliers or anomalies could indicate a new or previously unknown pattern of fraud.
The trick in anomaly detection is the difficult task of defining “normal.” Many poorly built anomaly routines return a high false-positive rate. However, with the ability to cluster accurately and take into account systematic factors such as trends and seasonality, SAS has reduced false positives while alerting insurance companies to previously unknown fraudulent schemes. In addition, the SAS solution includes robust multivariate outlier detection that assesses multiple fields jointly in the data, further reducing the false alerts that can be created by a simple threshold approach.
Uncovering who the scammers know
The most interesting “know,” however, may be who you know. Social network analysis opens a whole new window on fraud and abuse by mapping relationships between entities in claims to uncover organized fraud activities.
Using this linking technology creates a richer view of the alerts generated by the business rules, predictive models and anomaly models, but, more importantly, social network analysis can generate alerts on fraudulent networks that may not have been detected any other way.
Overcoming social network analysis challenges
While social network analysis can be highly effective, building the social networks is no easy task. The first challenge is the data, which is messy and incomplete at many organizations.
Using SAS data integration tools, insurers can use fuzzy matching on names, addresses and other text fields to eliminate data input errors and ensure accuracy.
Another challenge for social network analysis is the problem of super clusters. When linking data together, it is critical to find all the linkages, but the more links you have, the larger the cluster becomes. Making sense of clusters with tens of thousands of entities is impossible. To break these super clusters into more usable groups, SAS has developed an approach that allows insurers to see which links are interesting and which can be removed without loss of fidelity. Once meaningful clusters are created, insurers can score these networks for risk so that investigators spend their time following meaningful leads.
With a comprehensive fraud framework, including technologies such as social network analysis, you can reduce false positives while capturing more fraudulent activities and helping investigators focus on the biggest opportunities. Remember, fraudsters are always looking for the weak link – don’t let them find one in your organization.
Mark Moorman is a Senior Business Director in SAS Research and Development.