BNP Paribas has integrated a generative AI assistant into its ESG assessment process as part of its internal AI AT BNP PARIBAS series.
The move aims to drastically cut the time relationship managers (RMs) spend on client assessments. Currently, completing the ESG questionnaire, which can contain up to 80 questions, can take RMs hours. The new AI tool is designed to provide suggested answers and justifications, significantly reducing the filling time and gaining operational efficiency.
ESG assessment embedded in credit risk:
The ESG Assessment tool is housed within the Group’s internal data platform, impACT. Its purpose is to incorporate ESG parameters into a client’s risk assessment. This is achieved through a sectorial questionnaire and a review of ESG controversies.
The assessment is fully embedded in the bank’s credit risk process for lending decisions. The ESG questionnaire is a key pillar, collecting data on areas like governance, climate pollution, biodiversity, and human rights. It has been deployed across strategic clients, large corporates, relevant commercial clients, and financial institutions. Its use is progressively expanding into commercial real estate, project finance, and aviation.
How the AI tool works:
The AI assistant, developed internally using OpenAI’s Chat GPT 4 o mini model via Azure API, analyzes a client’s ESG or CSR documents uploaded by the RM.
The process involves the AI scanning and indexing the documents. It then displays suggested answers for the questionnaire, complete with comments and links referencing the relevant sections of the document. This helps RMs identify client risks and simplifies the analysis of the client’s ESG strategy.
AI does not replace RMs:
BNP Paribas emphasized that the AI is strictly an “assistant” and does not replace the relationship manager. As Enam Ehe, Head of ESG Data & Systems, noted, “Besides, AI makes the job simpler and quicker for the relationship managers, who remain responsible for the inputs provided.”
The RM remains crucial, as they must actively select and verify the AI suggestions before the answer is finalized. The RM also retains the ability to edit the generated content, ensuring accuracy and reliability.
Initial results and future goals:
The AI Assistant has shown good early results, with 600 users completing almost 1,600 ESG assessments since its launch. The tool is now used to complete approximately 35% of all ESG assessments, and this adoption rate is growing.
Future development plans include continuous improvement of the user experience, such as translating AI suggestions into various languages used across the Group. Data scientists will also work to enhance the prompts for more precise suggestions from the AI model.
