Netflix for Insurers: Preventing Fraud with Predictive Modeling
Feb 2, 2017
Gian Luigi Chiesa is data scientist at FRISS. He builds analytical tools that provide insight into the data that insurers use in their fight against fraud. Gian sees how Netflix profits from big data and concludes that the insurance industry can learn from this. In this blog he explains how predictive modeling can help to reduce insurance fraud. The explosion of data, in recent years, has opened an unprecedented opportunity to find new and deeper insights into complex problems. Traditional methods such as business intelligence, simple statistical models, slice and dice were developed around 30 or 40 years ago. Many firms still only use these traditional methods, which do not allow to answer to much deeper questions on your data sets.
Machine learning, deep learning and network analysis
Much modern tools are available nowadays, such as: machine learning, deep learning, network analysis. Those can be employed in your business with high performances. One of the most powerful and influential companies in the entertainment industry already heavily capitalizes on such methods. And I’m talking about Netflix.
A couple of years ago, Netflix realized that it needed to solve the problem of having so much content that the user would just get lost. In order to deal with this, Netflix built very large-scale networks which connect users and shows. The power of this methodology speaks for itself. The company estimates that about 75% of the viewer activity is driven by these recommendations. How does this concern the insurance world?
Connecting claims and people
Well, people are connected via their claims in a natural way. As Netflix connects users and shows, we can connect people and claims. In particular we can define subjects; which can be persons, companies, aliases, or objects; which can be vehicles, addresses, locations, and events; which can be entering a claim, having an accident, entering a policy. It makes a lot of sense to build a network based on these relationships.
And in such networks we can look for interesting patterns. Like the network of Netflix can profile a user as one who may like Fight Club, our network can profile a person as potential fraudster. Those predicted profiles can greatly enhance the value of your portfolio. For example, by apply these methods at the gate, when people are requesting a policy, you can differentiate how you are going to deal with people based on their predicted profile. In the future the evolution of these tools can lead to a more honest insurance, with less room for fraud and potentially a 100% straight-through processing of claims and policy applications. If you embrace those tools, you’ll be rewarded.