- Key insight: Americans don't trust their banks' virtual assistants.
- What's at stake: With more people using frontier models like Claude and ChatGPT for financial help, banks need to step up to compete.
- Forward look: Stronger data management, better fraud mitigation and experimentation with generative AI in virtual assistants could all help boost credibility, experts say.
Americans have little faith in their banks' virtual assistants, a just-released survey has found.
The survey, part of RFI Global's U.S. Innovation Monitor, asked 4,000 U.S. consumers how much they trust their bank's chatbot (among other things). About 33% said they trust their bank's virtual assistant "a lot" to answer product questions, 32% trust it for fraud alerts, 30% for learning about money, 28% for budget insights and 27% for reviewing finances. The vast majority of respondents trust these assistants a little or not at all.
What gives? About 78% of respondents said they hold banks to a higher AI standard than tech firms. Most said they would trust their banks' use of AI more if a human was clearly responsible for the outcome. And 63% said they would trust AI used by their bank more if a human was clearly responsible for the outcome.
There are some things banks can do to build trust. Dan Latimore, head of research, North America at RFI Global, said the path to trustworthy chatbots is through improving fraud mitigation, which will lead to consumers trusting their bank more, and that trust will extend to the bank's virtual assistant.
When financial institutions have strong anti-fraud measures and helpful 24-7 support, "you build on that solid foundation," Latimore told American Banker. "So you earn the trust, and then you ask for the next step in the trust ladder. It's tough to disentangle trusting the bank's AI versus just trusting the bank."
Piermont Bank, a seven-year-old, New York-based digital commercial bank built for entrepreneurs, innovators and fintechs, is trying to combine the trust and expertise of a regulated bank with a tech-enabled approach to banking, according to Chief Growth Officer Wenni Wu. It's navigating a balance between offering advanced AI tools and making sure they provide accurate answers.
The hard part, she said, is balancing data security and risk management across company systems while at the same time incorporating this new technology.
"I think the tricky part is that AI is only as good as the data that it can access," Wu said. "Access to data for small business owners is so different from those of [larger companies with] strict policies around data access. Most of us who are working in banking carry two phones. We use a work laptop, but the SMB owners do not necessarily do that. They don't have the same control around data access, and their strong demand to allow AI to access and run their business will accelerate the urgency to solve these fragmentation issues." By fragmentation issues, she meant the plethora of software programs small-business owners use to run their companies and finances.
Banks also have to work on improving access to their own data, she noted. "Bank data also lives in various different systems, and I think a lot of us are trying to figure out how we have clean data, so that AI agents can be effective," she said.
Read more:
- These are the
Top Performing Banks of 2026 in four asset tiers - New York bank
regulator says vendor flaw hit its firms - The
best credit unions to work for in 2026 Tokenized deposits are here . Banks need to manage the risks.
Some experts argue there's room for banks' virtual assistants to improve without using generative AI.
Most banks' virtual assistants draw upon libraries of pre-written answers rather than having generative AI create answers. Some use other forms of AI like natural language understanding to figure out what the customer is asking for and map the question to an appropriate answer.
"They're a step up from scrolling through all sorts of menus," Latimore said. "But they're not much better than that."
Latimore, who is a Bank of America customer, recently asked the bank's Erica virtual assistant if he should buy a CD.
"Bank of America knows all my transaction history and everything else, and Erica just sent me to the CD page of the mobile app," he said. Then he asked Anthropic's Claude, "I have $15,000 sitting in my checking account, should I open a CD? If so, where and what are some indicative rates and other things I should know about CDs?" Claude gave a long answer that included information about federal deposit insurance, rates, potential Fed rate cuts and the liquidity trade-off of buying a CD.
"It was a much more helpful answer, and that's what I think banks have got to be on the lookout for: Are people going to the LLMs first for questions like that," Latimore said.
Bank of America did not respond to a request for comment.
Banks with bigger budgets should be experimenting with generative AI assistants in sandboxed ways, Latimore said, "and iron out the kinks in a low-risk kind of way, and then as they figure out what works and what doesn't, roll it out."
