Customer Service Automation · Document Processing
BNP Paribas Global Markets sales teams were manually interpreting client requests for quotes from chat and email and filling response forms on a dedicated platform, which took several minutes per quote and made speed and reliability difficult at high volume.
Reported impact
The automation lets sales teams answer quote requests faster and with greater reliability, reduces manual work to less than a second for the translation step, and supports high request volumes across multiple asset classes.
BNP Paribas built Chat2Trade as an internal AI workflow for its Global Markets business to turn unstructured RFQ text into a structured format that downstream pricing systems can use. The system ingests client requests arriving through chat or email, extracts the plain text, and passes it to a pre-trained open-source BERT language model that BNP Paribas fine-tuned internally using historical client requests and synthetic data generated from templates. The model performs text classification and named entity recognition so the request can be understood and tagged with the relevant product characteristics. Those outputs are assembled into a final structured output format and handed off to internal pricing tools, which then generate the quote for sales teams to return to clients. BNP Paribas says the deployment is fully rolled out, closely monitored, and updated with new model versions almost every week.
Client RFQs arrive through chat or email, where Chat2Trade extracts the plain text and submits it to a fine-tuned BERT model. The model classifies the request and tags entities, and those outputs are converted into a structured output format that internal pricing tools use to produce quotes for BNP Paribas sales teams and their clients.
"Before the AI-based tool 'Chat2Trade' was developed by the Global Markets Data & AI lab in 2016, these requests were managed manually."
"Chat2Trade automatically translates client requests submitted by chat or email into a structured output format that can be used by internal pricing tools to provide a quote."
"Chat2Trade now handles around 50,000 requests across several asset classes every month. That volume of requests, without Chat2Trade, would correspond to hundreds of hours of manual pricings every month."
"This step is now reduced to a simple copy-and-paste in less than a second."
"Though Chat2Trade is now fully deployed... Each model is also monitored closely, and a new version is released almost every week."
"the models learn by performing 'text classification' ... and 'named entity recognition'"
"the request must first be understood. To do this, the Global Markets Data & AI lab uses a pre-trained open-source language model, BERT - developed by Google - that is then fine tunned internally on historical requests from BNP Paribas clients, or synthetic data built from a set of templates."
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
In-house Sample Apr 8, 2025 Bank of America Bank of America needed to improve employee productivity and client service at scale across a global workforce of 213,000 people, while reducing routine support demand, accelerating employee access to information,… 20 million clients now actively using the virtual assistant Erica · Erica for Employees · PRIVATE BANK In-house Sample May 8, 2024 Mapfre Mapfre wanted to improve customer experience and streamline insurance operations such as claims handling, risk assessment, underwriting support, document management, internal knowledge access, and digital content… 75 use cases under study for the application of generative AI ControlExpert · Cyberwrite · Holistic AI In-house Sample Apr 9, 2024 ClearBank ClearBank needed to deploy multiple generative AI use cases for banking API workflows in a regulated financial environment while safeguarding security, robustness, transparency, bias control, and regulatory alignment. Large Language Models (LLMs) · evaluation pipeline · model experiment tracking system 