Executives at financial institutions are looking into ChatGPT to unleash a new era of innovation, education and customer experience.  

While many FIs, like Fifth Third Bank, are using ChatGPT to improve the efficiency of their chatbots, others, like DNB, are exploring generative AI to increase operational efficiency.

Financial institution executives discussed the potential and pitfalls of using ChatGPT in their operations at the Artificial Intelligence in Financial Services conference in London last week.  

ChatGPT for CX 

Generative AI can be used to improve customer experience by improving chatbots and providing the tech for customer support teams to improve chatbot efficiency, said Nathan Huntley, data science manager at Atom Bank, a British neo-bank. 

“We’ve seen examples like, customer support — not directly chatbots with the customer ChatGPT — but rather with customer support teams,” Huntley said. Customer support teams will “have all have their expertise to be able to step in and say, ‘Well actually, I think it’s gotten quite a bit wrong here’” and provide the correct responses while expediting problem-solving. 

Similarly, Norwegian bank DNB is exploring the use of generative AI to improve its internal customer support chatbot, Juno, and make their customer support teams more efficient, Maia Sognefest, project manager at DNB, told Bank Automation News. 

Instilling guardrails for ChatGPT

ChatGPT offers no reasoning for its outputs, which can be risky when implemented in financial services. 

“What these technologies [generative AI] appear to do is they don’t reason,” said Simon Thomson, head of AI, ML, and data science at GFT, an IT solutions company. “They look like they reason but they are not reasoning, they are retrieving from past training, and that’s very different. And the [financial services industry] is really zeroing in on the differences there.” 

While inaccurate responses or hallucinations can be hurdles in AI implementation, financial institutions can train large language models on large diverse data sets to address those issues, said Ines Riguad, innovation analyst at HSBC, noting, that ChatGPT and other generative AI models “cannot do anything more than we can do, and it can only learn what we can do.  

“And it can only have the bias, the false empathy, the errors, the inaccuracies that we have, so it’s not going to be any worse than us,” Riguad said. 

Educating users 

While ChatGPT can boost innovation, educating users about potential pitfalls and limitations is essential, Hasintha Gunawickrema, former chief control officer of wealth and personal banking at HSBC, told BAN.  For the best and most efficient use of ChatGPT, financial institutions must provide employees with education and training of the tech.

“One of the most important steps is to educate all our colleagues, making sure that they understand what this tool can do, what they’re capable of, what their limitations and risks are as well,” Atom Bank’s Huntley said. “I will say education is probably the main step in” proper implementation of the tech in financial services.

Join Bank Automation News for the upcoming webinar, Global Ideas for Better Banking AI, on Thursday, Sept. 14, at 11 a.m. ET.  For more information on this free webinar and to register, click here.