Machine learning - A progressive solution for claims fraud detection

Driving lower costs and better outcomes with advanced predictive fraud analytics

Fraud is both an ever-present threat and growing opportunity for insurers who haven’t yet shifted to a data-driven detection model. EXL estimates the average fraud, waste, and abuse (FWA) rate in the United States hovers around 1-1.5% across most claims volumes. Despite the potential indemnity cost savings in the millions, and the seeming ubiquity of the problem, fraud and abuse has been notoriously difficult for insurers to mitigate.

Carriers face two main difficulties: a relatively small representation of known fraudulent customers, and a diversity of fraudulent behavior patterns. Criminals, it turns out, are quite creative when it comes to evading identification. One important angle to finding solutions in this space may lie in taking a new angle on addressing the problem. Rather than sifting through individual claims to identify potential instances of fraud, find the persona of a fraudster and work backwards towards the relevant activity patterns.

Using a combination of external data, structured internal data, and derived unstructured data using AI-enabled extract tools, it is possible to rapidly identify, develop, and deploy predictive models that can boost potential FWA identification by 3-5x. Moreover, these scalable modeling approaches are applicable to an array of products, and can be built to support multiple different data & technology ecosystems. Importantly, a critical success factor to realizing the benefit of fraud, waste, and abuse analytics is to understand the process end-to-end, fully taking into account the support mechanisms needed to assess, confirm, escalate, investigate, and remediate instances of potential fraud.

The value of human interpretation and experience is never discounted in this approach. Predictive modeling is a tool that greatly enhances identification capabilities of an end-user specialist enabling them with data-driven techniques that in combination prove to be more effective in the long run. Machine learning models are complex but their results are invaluable to accelerate the results organizations are looking for and to thwart seemingly ever present attempts to defraud.

Written by:

Karl Canty
Vice President, Analytics

Subha Datta
Senior Program Manager