Financial technology provider Finastra has launched Assist.AI, a chatbot designed to answer questions on trade finance for users of its Trade Innovation platform.
The company says that it has built the product to help bridge the widening knowledge gap in the trade finance industry between “seasoned professionals and new entrants”. It will allow users to get answers to questions of varying levels of complexity, with a particular focus on using Trade Innovation.
Examples of answers it can provide include outlining the difference between an advising bank and an advise through bank, explaining how to handle a discrepant letter of credit, and setting out how the Trade Innovation platform can support mixed payments, the company tells GTR.
Assist.AI is built on architecture from Microsoft Azure’s OpenAI Service, meaning it functions similarly to ChatGPT, analysing vast sets of training data to ‘learn’ how to respond.
To specialise this AI model for trade finance, Finastra used bespoke data sets including the Trade Innovation functional user guides, technical user guides and release notes, the company says.
“This timely and much needed solution represents a significant leap forward in our commitment to advancing open finance and leveraging AI technology to solve real-world challenges in financial services,” says Andrew Bateman, executive vice-president of lending at Finastra.
“By providing instant assistance to bank employees, we are empowering our clients to navigate the complexities of trade finance with greater ease and efficiency.”
Generative AI is increasingly popular in the banking sector, being used for customer service, fraud detection and credit checks. In December 2023, Standard Chartered said the technology is “set to supercharge the transformation of banking with unprecedented innovation and efficiency.”
However, chatbots based on generative AI have some unique challenges. The most prevalent is their tendency to ‘hallucinate’, providing inaccurate answers with confidence that they are correct. This can be particularly problematic when being used to gain knowledge, as a user is unlikely to know that the AI is wrong.
Asked about this potential issue, Finastra says its AI models “underwent rigorous testing and validation using separate datasets not seen during training”.
Metrics like accuracy, precision, recall, and F1 score were used to evaluate performance,“ a company spokesperson says. “Continuous monitoring and user feedback helped refine the models, ensuring reliable and accurate responses.”
Finastra and Microsoft have partnered several times in recent years following the 2023 signing of an agreement between the two focused on trade platform modernisation.
