Digitalization has made and will continue to make a huge difference in many industries, and insurance is not an exception. My colleague recently wrote about the possibilities that the Internet of Things will bring in insurance. Being able to track driving or living habits with the help of the data gathered by sensors may result in safer and healthier behavior and also lower insurance fees.
However, accidents continue to happen, and digitalization also affects the way they are dealt with. People do not visit their local insurance office anymore, but they want to make the claim online wherever they happen to be. And in order to defend and grow their market shares by providing as smooth customer experience as possible, the insurance companies will process the claims as fast as possible – perhaps completely automatically.
How does digitalization affect insurance fraud?
Unfortunately, not all claims are legitimate. It is a well-known fact that approximately 10% of the compensations paid are due to fraudulent claims. In practice, this can mean anything from opportunistic exaggeration to completely fake claims and even organized criminal activities.
In the past, the most important part of fraud detection in insurance companies have been the people who process the claims. Their review of the claim and often a gut feeling filters the claims that have something suspicious in them. There are also various types of rules to sort out potential fraud, but usually they focus on rather simple questions: When did the accident happen? What is the value of the claim? What is the near history of the claimant like?
This now raises a question: Who will check the claims for fraud, if the process is completely automated? And further: Will this open up a chance for fraudsters to take an advantage of the system?
There is indeed a danger that digitalization enables creating fraudulent claims with appropriate characteristics, which would pass the system and being paid automatically. This is why insurance companies need to include more advanced analyses as a part of the automated claims handling process, which will aim at filtering suspicious claims for more detailed review.
What do we mean with more advanced analyses?
First of all, we can look at the claim in more detail. Are there suspicious features such as personal injuries that are of high compensation value, but hard to prove by medical inspection? Does the description include words or phrases that are typically used in previous proven fraud cases?
Second, we need to analyze the claimant more thoroughly within his or her peer group. Is the value of claims exceptionally high considering the claimant’s age, profession and other characteristics? Is the frequency of claims unnaturally high, and do the claims often occur at suspicious points of time, such as launching of new devices?
Thirdly, and perhaps most interestingly, we need to pay attention to the claimant’s network. Is this person linked with previously identified fraudsters through address, telephone number or other connection? Is there a number of similar cases in a short period of time linked with each other through a service provider, for example?
All this can actually be done with the help of modern analytical techniques: anomaly detection, predictive modeling, text analytics and network analysis. These methods along with appropriate collection and transformation of data will produce insight about suspicious cases based on previous experience, statistically anomalous behavior and linkages between entities. Their main advantage compared to simple rules is their ability to narrow down the target group for investigation so that it includes as many fraudulent cases and as few false positives as possible. Furthermore, analytics can be applied even in connection with online processes by utilizing real-time technologies such as event streaming.
Analytics supports, but does not replace humans
Can these methods then replace human claims handlers and investigators completely? No, because an information system cannot prove that a case is fraud. It can only tell that a case is suspicious. We still need people who examine the suspicious cases, ask clarifying questions from the claimant and finally make the decision.
As a conclusion, insurance companies need to adopt advanced analytics along with their transformation towards digitalization. At the same time, they need to bear in mind that analytical technologies are only a part of the defense line, and people will remain equally important. In their intersection, we need to carefully tie both of them together in processes, in which fraud is detected. When all these aspects are in balance, digitalization in insurance claims processing may continue to be an asset that improves customer experience and efficiency – without sacrificing legitimacy.