The AI application that most bank customers in the United States have already met is the customer service chatbot. In a June 6, 2023 report, the Consumer Financial Protection Bureau (CFPB) estimated that about 37 percent of the U.S. population, over 98 million people, interacted with a bank's chatbot in 2022, and that every one of the 10 largest commercial banks used one. The same report warned that chatbots handle simple questions well, handle complex disputes poorly, and must still follow all applicable federal consumer financial laws.
Key takeaways
- No source supports naming one "top" AI app for finance; the best-documented category is the bank chatbot, which the CFPB examined in 2023.
- The CFPB distinguished rule-based chatbots, which follow decision trees, from complex ones built on machine learning, including large language models.
- Its named problems were limited ability to solve complex problems, inaccurate information, failure to provide meaningful help, and hindered access to a human.
- The report says chatbots must comply with federal consumer financial laws, while stating it is not an interpretation of any regulation.
- In Canada, the Financial Consumer Agency of Canada (FCAC) tells consumers that AI answers may be inaccurate and that banks must give clear, non-misleading information.
What the CFPB measured and what it described
The CFPB's issue spotlight on chatbots in consumer finance is an overview, not a controlled study. It combines public information, vendor and bank announcements, and consumer complaints. Its headline figures are an estimate of 98 million U.S. users in 2022 and a projection of 110.9 million by 2026, plus the finding that all of the top 10 commercial banks deploy chatbots of varying complexity. It also cites a 2017 industry projection of USD 8 billion in annual savings, about USD 0.70 per interaction, which is a forecast from a third party and not a measured result.
The report counts no complaints, and its examples are quotations from individual consumers. That makes it good evidence for how widespread chatbots are and for the kinds of problems people report, and weak evidence for how often those problems occur.
Two kinds of chatbot, two kinds of failure
The report separates simple bots from complex ones, and the failures differ. The table summarizes its descriptions.
| Type | How it works, per the CFPB | Typical failure it names |
|---|---|---|
| Rule-based | Decision trees or keyword databases trigger preset, limited responses | Disputes are recognized only when a customer uses specific words, and the bot may repeat the customer's own information back |
| Complex | Machine learning or "artificial intelligence" simulates natural dialogue; large language models predict the text that should follow a question | Wrong or invented answers, because training data can contain misinformation |
Banks reach these systems in different ways. The report notes vendor-built bots, in-house systems, and bots built on large technology platforms. It describes the arrangement without ranking any provider, and this article does the same.
The problems the report names
Four problems recur. First, limited ability to solve complex problems: customers in a dispute need an accurate record, and a bot that only parrots information does not meaningfully handle the dispute. Second, inaccurate or unreliable information: where the law requires accurate information, a wrong chatbot answer may violate those obligations. Third, failure to provide meaningful assistance, including what the report calls "doom loops," in which a customer receives repetitive responses and never reaches a person. Fourth, hindered human intervention, supported by a survey cited in the report in which 80 percent of consumers said they were more frustrated after a chatbot interaction and 78 percent needed a human afterward.
Harms in the report include wasted time, late fees when no one could take a payment, and a complainant who was told a dispute had been opened when the record showed none. A fifth category is security: the report describes a 2018 incident at a UK ticket seller involving a third-party chatbot that exposed data on 9.4 million people, including 60,000 payment card details, and warns about impersonation bots used for phishing.
What the legal language does and does not say
The report says chatbots must comply with all applicable federal consumer financial laws and that institutions may be liable when they do not. It adds that customers expect timely, straightforward answers regardless of the processes or technologies used, and it cites a CFPB circular on protecting sensitive consumer information. Its footnote also states the document is not intended to impose obligations, define rights, or interpret any regulation or statute. The CFPB said it was monitoring the market and would hold entities to account if they failed to meet their obligations. Whatever the agency's priorities may be today, the report remains a dated description of the 2022 to 2023 market and not a statement of current enforcement.
The Canadian view
The FCAC's page on artificial intelligence in banking, with page details dated October 16, 2025, lists virtual assistants among banks' uses of AI and warns that answers may be inaccurate. It also tells consumers to protect the personal information they share with AI tools and to watch for scammers who use AI to imitate people they know. The page says federally regulated banks must offer products suited to a customer's needs and give clear, simple, non-misleading information, and it points to the agency's complaint process. Its reference to a proposed federal AI law is out of date, because that bill lapsed in January 2025.
Questions a customer can ask about a bank's chatbot
- Does the chat window offer a clear route to a human, and how long does it take?
- Is a dispute or complaint logged with a reference number, and can the status be checked outside the chat?
- Does the bot state that it is automated, and what does it say about how chat logs are stored?
- Is the answer one the bank would confirm in writing or on its official website?
This article gives general information about reported findings and public guidance. It is not legal or financial advice and does not recommend a bank or product. A person with a dispute should use the bank's complaint process and, where needed, a regulator's complaint channel or a qualified professional.
FAQ
Is there a single best AI app for banking?
No source ranks one. The documented category is the bank chatbot, and the CFPB found that every top 10 U.S. commercial bank used one in 2022.
Do bank chatbots use generative AI?
Some do. The CFPB described complex chatbots built on machine learning and large language models, and it also described simpler rule-based bots that follow decision trees.
Are banks responsible for what a chatbot says?
The CFPB report says chatbots must comply with federal consumer financial laws and that institutions may be liable for failures. It states that it is not an interpretation of any statute, so specific questions need a professional.
Where can a Canadian customer complain about an AI service?
The FCAC page advises first complaining to the federally regulated entity and links to the agency's complaint process.
Related reading
- Measured benefits of AI for industry
- A consumer chat app profile
- Which industries keep human work
- Another chatbot profile: Baidu's Ernie Bot
More in AI in finance.
Last verified: October 2026
Sources
- Chatbots in consumer finance, Consumer Financial Protection Bureau, June 6, 2023
- Artificial intelligence in banking, Financial Consumer Agency of Canada, page details October 16, 2025
- Prioritizing privacy in a data-driven world (2024-25 annual report, on the lapse of Bill C-27), Office of the Privacy Commissioner of Canada, 2025