Insurance fraud is a worldwide problem. The numbers are staggering. According to the FBI, in the U.S. alone, insurance fraud (ex healthcare) costs Americans $40 billion a year. That works out to $400-to-$700 per family. That means higher premiums. So how do companies crack down on fraud? How can they ferret it out? One of the tools in their tool box is, or should be, text analytics.
Insurance companies collect massive volumes of text data every day from customers, agents, adjusters, health care professionals, hospitals, government agencies, police, witnesses, etc. For every claim or potential claim there are reports to fill out, valuations and estimates to be calculated. The vast majority of this data is unstructured text. It is where fraud starts and where it can be discovered. Text analytics can analyze patterns to catch fraudulent claims.
Content analysis software like WordStat can be given data sets of known fraudulent behavior and using machine learning algorithms can “learn” to recognize and flag suspicious claims. Using similar techniques, text analytics software can analyze victim statements, accident reports, workers’ compensation claims and others. The suspicious claims can be sent back for additional investigation or directed to the appropriate department. Insurers can improve their analysis by customizing categorization dictionaries to specific areas of insurance claims such as accident, home owners, commercial property, maritime, life insurance and so on. These different sectors probably have different fraud profiles and domain specific language associated with them. The domain specific dictionaries will help recognize fraud specific to that insurance sector. These dictionaries will also help reduce the number of “false” positives which take additional time and money to investigate.
As we stated at the outset of this blog, fraud is not limited to certain countries or certain types of insurance. It is pervasive across all domains and international. Tackling fraud is time consuming and expensive but done properly, there is a significant payback. Text analytics software isn’t a magic bullet but it should be part of your arsenal.