Can Big Data help in detecting insurance fraud?

The technology that most insurers have currently in place to help to fight frauds is a blend of business rules and database searches, where the results rely heavily on the sensitivity of the claims auditor. While these techniques have proved being successful in detecting known fraud patterns, insurers today need to invest in new analytical capabilities to help them to spot unknown and complex fraud activities. These analytical capabilities include incongruity detection, predictive modelling, unstructured data mining and social network analysis.

Anomaly detection aims at discovering fraud by identifying those elements that vary from the norm. 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 fraud pattern.

Predictive models use past fraud events to produce fraudpropensity scores. Adjusters simply enter data and claims are automatically scored against the likelihood of them being fraudulent. These scores are then made available for review. Use of predictive modelling makes it possible to understand new fraud trends.

Since around 80 percent of claims data is unstructured, the use of tools able to mine unstructured data enables insurers to analyse information arising from medical chronicles, police records, external and internal database sources or even e-mails.

Social network visualisation tools allow investigators to actually see network connections so they can uncover previously unknown relationships and conduct more effective and efficient investigations.

By using Big Data technologies companies are able to manage all of these issues and to ‘learn’ from experience to improve their fraud detection and pattern identification capability. Reply has established a proven methodology to apply a Bayesian model in fraud recognition combined with Big Data analysis techniques. This is a comprehensive approach, which includes data discovery through all the available internal and external structured and unstructured data sources, combined with the powerful computational capabilities of a Big Data infrastructure to support the claims manager in every phase of the investigation.

First of all, a network analysis will identify any historical relationship between the actors in a specific claim, revealing any connection in the past that could suggest a propensity to commit a fraud. Then a clusterization of the actors and related behaviors based on a self-learning statistical model let emerge similarities in the data model, to better represent relations and attitudes to plausible fraud existence.

While this technology is still in its early stages, the bottom line is that new Big Data analytics can be used to explore large volumes of networked data, using high-speed processing with configurable data entry from multiple internal and external sources, to reveal fraudulent behaviour. Can you imagine how far you could go using a so strong paradigm change in tracking frauds?