Cathay Financial Tests Open-Source SLM to Improve Financial Customer Intent Recognition

Closed-source large language models can deliver strong performance when financial institutions adopt generative AI, but they may also increase costs and create data-governance and maintenance burdens. Cathay Financial Holdings has therefore studied fine-tuning open-source small language models, or SLMs, to identify customers’ financial-service intentions. The aim is to support use cases such as mortgages and credit-card payments with leaner models and system architecture.

Cathay Financial’s latest forward-looking research found that a fine-tuned open-source SLM delivered customer-intent recognition performance approaching that of mainstream closed-source large language models. The company also completed proof-of-concept tests in scenarios including mortgages and credit-card payments. Cathay Financial did not disclose the publication date, model parameter count, accuracy rate, investment amount or timetable for formal deployment. Its current focus is reducing the complexity of subsequent operations and maintenance.