Large language models (LLMs) will reinvent how many consumers and businesses interact with their bank and use their financial data. LLMs are inherent to generative artificial intelligence (AI) platforms and serve as data query, synthesis and analysis engines more powerful than any previous generation of AI tools. Customers will perform banking activities from within the LLMs themselves. They are also taking shape as platforms, not simply modes of interaction. From platforms come marketplaces, like the smartphone since 2008, and now LLMs such as those embedded in ChatGPT, Claude and Gemini. Financial institution-branded banking apps will appear and proliferate on these marketplaces.
The last revolution in banking channels — smartphones and their app stores — created an endpoint of the scale that nobody could have imagined. Data from Pew Research Center shows smartphone adoption in 2025 was 91% for U.S. adults and mobile banking has reshaped consumers’ relationships with their financial institutions. Today, mobile dominates consumers’ banking channel use. In 2024, 71% of U.S. consumers said that they used their smartphone to access their bank accounts online and 59% said it was their preferred channel.
LLM adoption is scaling even faster, since they can be accessed for free and do not require a hardware purchase like a smartphone. Pew in March 2025 also found 34% of U.S. adults said they had used ChatGPT, fewer than three years after its public release. That’s nearly double the number in 2023 and does not include usage of other popular generative AI platforms with LLMs. Today the number is surely higher. While LLMs will not fully replace mobile banking apps, they will become a new place for users to conduct banking.
Staff Get Smarter, Not Replaced
Tellers, bankers and customer service representatives will benefit immensely from LLMs on the administrative side of digital banking. The trend in workplace uptake of AI tools is clear: Since 2024, 27% of white-collar workers used AI a few times a week or more, according to a Gallup study. The most popular use cases can be derived from general-purpose AI tools: U.S. workers who used AI in their jobs at least once a year said that they used it primarily for consumer use cases, like to consolidate information or data (42%) or generate ideas (41%).
Workers’ adoption of AI in the workplace will grow as internal, specialized AI tools are developed. One bank reported that employees had replaced about half of their search queries with an internal AI tool. Any internal AI should in the end augment the staff’s abilities, make operations more efficient and facilitate a more personalized customer experience. That includes
enabling staff to be better-informed about their customers. LLMs, for example, will be integrated into administrative portals to give financial institution staff quick summaries of a customer’s recent banking activity, help them generate personalized offers and to create product and service collateral.
Guardrails Are Developing for LLM Banking
LLM banking must always be bound by safety mechanisms and standards when data is being shared. One important development has been the model context protocol (MCP). MCP is an open protocol that standardizes how AI applications connect to data sources. Institutions implement MCP servers that define what banking data is accessible, what operations are permitted and how data is formatted for LLM consumption.
Today’s guardrails prevent the LLM from making changes to banking data. Narmi’s MCP, for example, is configured to be read only. Agentic features, like autonomously initiated transactions, will have their day, starting with the lowest-risk use cases. But in the context of banking there should always be a human in the loop — the customer — to review and approve actions before the LLM takes them.
Will LLMs Disintermediate the Bank?
LLM banking should not cause an existential crisis for financial institutions. Questions over who should control a channel and own the customer provoked soul-searching during the headiest days of banking as a service (BaaS) and the rise of neobanks that raised billions in venture capital. Open banking and customer data portability likewise stoked fears over customer ownership. Both fears, while well founded, were ultimately blown out of proportion. Additionally, the purpose of LLM banking is different.
LLMs are an interface and a platform — a channel — that will facilitate consumers’ and business’ access to financial services. In some cases, they will even be customers’ primary access point for their bank, like mobile banking is today for a wide array of tasks. Like smartphones, LLMs will not be an alternative to the bank. They will be an extension of the bank. Financial institutions that enable LLM banking will own the customer — because consumers and businesses will switch to the banks that offer the channels they most demand.
Previous