Document Processing · Risk Management
Relationship Managers needed hours to complete ESG Assessment questionnaires of up to 80 questions from client ESG and CSR documents, creating operational inefficiency and data quality issues in a process embedded in BNP Paribas's credit risk workflow.
Reported impact
The AI Assistant helps Relationship Managers complete ESG Assessments faster, simplifies analysis of client ESG strategy and risks, reduces data quality issues, and has achieved good user adoption while keeping Azure LLM operating cost low.
BNP Paribas built an internal AI Assistant into its impACT platform to accelerate ESG Assessments used by Relationship Managers and risk teams. In the workflow, Relationship Managers first upload relevant client ESG and CSR documents that were either publicly disclosed or shared during client meetings. The AI Assistant then launches analysis in impACT, where the documents are scanned and indexed and prompts are sent through Azure API to OpenAI's ChatGPT 4o mini model. The model returns suggested answers and justifications for questionnaire items, along with links to the exact document passages supporting each suggestion. The generated content is served back inside impACT rather than being auto-filled, so Relationship Managers must review, select, edit, and validate the proposed answers before submission. The resulting assessment feeds BNP Paribas's ESG synthesis and credit decision process, while the bank monitors answer relevancy, user feedback, and Azure LLM operating cost.
Relationship Managers upload client ESG and CSR documents into impACT, then trigger the AI Assistant. The documents are scanned and indexed, prompts are sent through Azure API to ChatGPT 4o mini, and the model returns suggested answers plus comments and links to supporting passages. Those suggestions are displayed in impACT for Relationship Managers to review, edit, and use in the ESG Assessment and downstream credit-risk process.
Documents are being scanned and indexed by the AI, and the AI displays suggested answers for the questionnaire from the documents uploaded, and comments referencing the relevant sections of the document
The AI Assistant has been used by 600 users since the Go live, helping them to complete almost 1600 ESG Assessments
The AI Assistant is used to complete around 35% of all ESG Assessments
The ESG Assessment is implemented in the impACT platform, (the Group’s internal platform dedicated to ESG data).
The tool is the result of internal development, using Open AI’s Chat GPT 4 o mini model via Azure API.
it takes hours for a relationship manager to fill in the ESG Assessment, which can contain up to 80 questions
the main goal of implementation of Generative AI is to assist Relationship Managers in completing the ESG Assessment questionnaire faster and more efficiently, by providing suggestions answers and justifications based on quick comprehensive analysis of the documents disclosed by the clients and uploaded by the Relationship Manager.
we also monitor the cost of running the LLM model in Azure, which is very limited in this use case.
Review how this reference was scored and trace each normalized claim back to its documented source.
Scoring methodology
Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.
Traceability
Field coverage shows which facts each source supports. Dates are shown only when documented by the source.
Related
OpenAI Sample Oct 16, 2026 Benefit Street Partners Benefit Street Partners needed to answer investment-committee follow-up questions faster, compare lender marks across shared loans, and review 300-page credit agreements for clauses that could shift value away from… 10–12 hours total. BDC mark tracker · Codex · Credit agreement analyzer OpenAI Sample Dec 4, 2024 Endex Investment professionals needed AI that could reason through complex financial documents and data with the precision, traceability, and multistep thinking required for real analysis rather than generic search results. Financial experts preferred responses generated by OpenAI’s o1 model 70% of the time AI Analyst · GPT-4o · o1-mini OpenAI Sample Oct 15, 2026 Sullivan & Cromwell Sullivan & Cromwell needed to review draft agreements against relevant precedents, client priorities, and market practice in one workflow so lawyers could do deeper legal analysis without losing transparency into… Agreement Analyzer · ChatGPT Enterprise 