BNP Paribas has identified hundreds of use cases for AI and is using the tech to enhance its virtual agent capabilities and automate data extraction, improving its ability to analyze unstructured data.  

“The purpose of AI is to take unstructured data, access it, structure it, analyze it and to create a model around it, and that’s what we’re trying to apply in each and every use case that we do have,” Bruno Campenon, head of financial intermediaries and corporates at BNP Paribas, said last week at the Sibos 2023 event in Toronto.

The Paris-based bank is implementing AI throughout its organization, including using it to improve client experience with virtual agents, he said.  

Utilizing AI, the bank’s “virtual agents are capable of accessing robots, and those robots, through API, will be retrieving information, taking that back and feeding it back to the client,” Campenon said. The API-to-API connection presents an opportunity for system-to-system automation.  

“That’s quite powerful,” he said. 

When a client calls the bank, instead of being routed to a relationship manager, the virtual agent sends an API that talks to its operators, uses another API to retrieve the results and still another API to send the results back to the server to answer the client’s question, Campenon said. 

 This API-driven tool has been in the works for five years, with improvements expected, he said.  

Automated document processing

The $2.8 trillion bank is also implementing AI into document processing, more specifically fund prospectus analysis, where there was a need for AI-driven model training, controls and compliance, Campenon said. 

“What we’ve done is train a number of models in order to read those prospectuses and detect those controls,” he said.  

With AI in place, the prospectuses are being read and analyzed with 91% to 92% accuracy, Campenon said, noting that human accuracy is between 92% and 95%.  

“We’re getting very close to reaching the level of the human,” he said, “and I guess we’ll go above that.”