Auto insurance fraud costs companies billions of dollars every year. Those losses trickle down to policyholders who absorb some of that risk in policy rate increases.

Thélem assurances, a French property and casualty insurer whose motto is “Thélem innovates for you”, has launched an artificial intelligence program, prioritizing a fraud detection use case as its initial project.

Fraud detection is a model that lends itself well to online machine modeling and is a project that would allow us to enter into artificial intelligence starting with the analytical field that we have prioritized. A successful fraud detection project would deliver immediate, significant financial gains for the company.

Tapping into IBM Services

We carried out a few preliminary tests and experiments internally with our data scientists and data engineers but encountered problems with tools and with the environment. Therefore, in order to go a step further, we decided two things. First, we needed to find a solution that would make it possible for us to free ourselves from storage and performance constraints. Second, to increase our expertise we realized we needed to engage experts in the field.

During the course of our research, we met with various representatives from IBM who showed the advanced analytics capabilities of IBM Watson Studio and IBM Cloud. We discovered that the value proposition that they proposed corresponded exactly to our needs.

At the beginning of the collaboration, an IBM Global Business Services (GBS) team met with different Thélem assurances teams including marketing and claims management to identify use cases for artificial intelligence. Car insurance is the area in which we experienced the majority of our cases of fraud, so we chose to begin there.

In addition to IBM Watson Studio, which is used as the development environment for analytical models and cases, additional solutions we employed include IBM Cloud with Secure Gateway Service to transfer data from Thélem to the IBM core; IBM Cloud Object Storage, which hosts data stored in the cloud; and IBM Watson Machine Learning, used for deploying IT scripts.

We also had to take into account GDPR legislation and regulations in Europe. Because of these regulations, we paid special focus to minimizing the amount of personal data uploaded and of securing the personal data we received throughout the initiatives implemented.

IBM GBS worked with us on the architecture definition and the addition of data in a secure manner to the cloud. Then, we worked together to define the method to implement use cases and followed up with training using data science models.

Discovering five times more potential cases of fraud

We realized an advantage right from the start: flexibility. The flexibility in launching the fraud detection solution, without having to concern ourselves with storage capacity, machine performance or services used, was phenomenal.

The IBM solution also makes it possible to facilitate joint work among the different data scientists working on initiatives. More than one of them can work on a single initiative, share their work and the progress of their algorithms.

More concrete, tangible advantages include the fact that we’ve increased five fold the relevance of cases identified as potentially fraudulent. Additionally, IBM GBS helped us develop a methodology that we can use again on our own. We now have a tool, the methodology and data modeling know-how. This makes it possible for us to enrich our models over time, make better progress and ultimately increase the relevance rate, which should allow us to save an additional several hundred thousand euros every year over the next few years.

Going forward, we plan to begin exploring additional Watson tools such as testing aspects of image recognition or of chatbot services.

Read the case study for more details.