BNP Paribas

BNP Paribas Global Markets sales teams were manually processing thousands of client requests for quotes (RFQs) submitted via chat and email. Each request required salespeople to interpret free-text instructions and manually fill in structured response forms, taking several minutes per quote depending on complexity. This created bottlenecks and limited the team's ability to focus on client support.

The Global Markets Data & AI lab developed Chat2Trade in 2016, an NLP-based tool that automatically translates client RFQs from plain text (chat or email) into structured output formats compatible with internal pricing tools. The system uses a fine-tuned BERT language model trained on historical BNP Paribas client requests and synthetic data, performing text classification and named entity recognition to parse and structure each request.

Chat2Trade reduced the manual quoting step to a simple copy-and-paste taking under one second. The system now handles approximately 50,000 requests across several asset classes every month — volume that would otherwise correspond to hundreds of hours of manual work. The tool has become a benchmark within the bank, and NLP automation is now considered a standard component of new sales and trading workflow systems.

• Fine-tuning open-source models (BERT) on domain-specific historical data produces reliable, production-grade NLP automation without requiring proprietary model development.
• Automating high-volume, repetitive intake workflows frees sales teams to focus on client relationship and market opportunity — a quality and speed gain that is critical in trading.
• Once embedded, AI components can become architectural standards: Chat2Trade made NLP automation a default expectation for all new workflow tooling at BNP Paribas CIB.