Financial institutions must consider the efficiencies and limitations of artificial intelligence when deploying chatbots as client adoption grows and regulators address pain points of the technology.

According to the Consumer Financial Protection Bureau’s (CFPB) “Chatbots in Consumer Finance” report, released yesterday, adoption of bank chatbots is expected to hit 110.9 million users by 2026. In 2022, there were 98 million users of bank chatbots.

As adoption increases, the CFPB warns financial institutions that compliance must be top of mind, according to the report. For example, while cost savings and internal and external application of the technology presents opportunities for banks, chatbots do not come without flaws.

However, even with potential AI-related risks, banks are accustomed to maintaining compliance when implementing new processes, Chris Willis, co-leader of Troutman Pepper’s consumer financial services regulatory practice, told Bank Automation News. “I think what the announcement reminds us, is that like any other business process, if they’re not well designed, if they don’t have the appropriate controls, or if they’re not monitored appropriately — things can go wrong.”

AI risks

The report identified the following risks:

• Security risk: Chatbots present an opportunity for fraudsters to gain personal or payment information from a bank client. FIs are obligated to protect client information, including personally identifiable information, and that protection needs to be applied to chatbots, too.
• Failure to handle disputes: FIs are required to accurately identify a dispute. Depending on how the technology is programmed, a chatbot may not be equipped to recognize a dispute or complaint.
• Inaccuracies: Chatbots may provide an inaccurate response to a bank client. This could lead to disinformation, which would violate a bank’s responsibility to provide accurate and reliable information to clients, including the request for financial advice.

Today’s chatbots

Chatbots on the market today differ based on the technology underpinning their solutions, according to the report. Types of chatbots include: those with a database of keywords that trigger preset responses, large language model (LLM)-based bots that analyze patterns and predict responses to questions, and domain-specific chatbots that help accomplish specific tasks.

Financial institutions including Bank of America, Truist, U.S. Bank and Wells Fargo all employ chatbots on their platforms.

Bank of America, for one, deployed its chatbot, Erica, for both internal and external use, Chief Executive Brian Moynihan said during the bank’s first-quarter earnings call in April. The $3.1 trillion bank can leverage the technology for data input automation, research and analytics, and coding.

“Since incorporating Erica into our internal processes, we’ve heard from teams across the country that utilizing Erica’s capabilities to compile client information saves our bankers hours of research and manual work,” Jorge Camargo, senior vice president of digital product management at Bank of America, previously told BAN.

As banks continue to invest in AI, FIs should “not take this [report] as a sign that they can’t use chatbots… because they can be used in a totally safe and compliant way,” Willis told BAN.