2022 integrated report

The digitalisation of lifestyles leads to many requests for immediacy while significantly increasing the risk of fraud. To meet the requirements of our customers and better protect them, artificial intelligence offers a new avenue. With a defined framework: make this booming technology a lever to better understand and serve our customers.

While split payments are increasingly frequent, BNP Paribas Personal Finance is strengthening its means of combatting fraud on e-commerce sites without deteriorating the customer experience, by combining the contributions of an artificial intelligence system and a graph database⁽¹⁾.
This technology, proposed by the publisher Neo4j, makes it possible to use, in real time, the links between financing requests to improve decision-making. Already used in the recommendation engines of retail sites to suggest purchases to internet users, it consolidates the arsenal of resources mobilised by BNP Paribas Personal Finance to detect fraudsters. Thus, files sharing information, such as the same telephone or bank card number can be linked to form a network whose characteristics are used to distinguish suspicious situations from legitimate cases. By providing a more sophisticated detection capability, graph analysis provides valuable information to reduce fraud while limiting the rejection of legitimate requests.

As insurance requires the collection and prior validation of supporting documents, BNP Paribas Cardif designed the CardX solution, based on artificial intelligence techniques. This innovation automates processes to facilitate interactions with the insurer’s customers, reducing response times from several weeks to a few minutes, whether to subscribe to a guarantee or to make a claim. Connectable to all of the insurer’s information systems via an API⁽²⁾, this secure solution has three functionalities: document recognition; data matching and data extraction. It adapts to all types of documents and is based on state-of-the-art machine learning models. Over time, the algorithm learns to identify the data of interest and to extract them correctly from the documents. Operational in Poland, Spain, Brazil and Colombia, this solution has already analysed nearly 150,000 pages and continues to be deployed in other countries where the insurer is present.