The Bank May Lose the Customer Interface Before It Loses the Customer

Introduction

A customer can connect accounts from thousands of financial institutions to one conversational interface without moving any of those accounts.

OpenAI’s personal-finance experience in ChatGPT supports connections to more than 12,000 financial institutions in the United States. It can combine balances, transactions, investments and liabilities with financial goals and other context supplied by the user.

As of October 2026, OpenAI says Finances in ChatGPT is available in the U.S. to Free, Go, Plus and Pro users. Connected accounts remain read-only: the experience can analyse financial information but does not move money or alter the underlying accounts.

The bank can still hold the deposit, issue the card, service the loan and remain responsible for the regulated product while another company becomes the place where the customer asks what happened, what matters and what to do next.

The UK Financial Conduct Authority is already examining this possibility. Its 2026 Mills Review says consumers could increasingly interact with financial services through AI-mediated interfaces rather than directly with firms, with personal agents coordinating decisions across products and providers.

A bank does not have to lose the customer account to lose part of the customer relationship. The earlier shift is control of the interface where financial intent is expressed, options are compared and the next action is chosen.

At a glance

• General-purpose AI can aggregate financial information from several institutions before any account is moved.
• The first competitive pressure is likely to appear in discovery, interpretation and recommendation rather than custody.
• An agent with a cross-provider view may understand the customer’s financial position differently from a bank that sees only its own products.
• Banks can respond by exposing selected capabilities safely to external agents, building their own agentic interfaces, or doing both.
• Product information and servicing capabilities increasingly need to be understandable to software as well as people.
• The regulated relationship can remain with the bank even while the routine conversational relationship becomes thinner.

The account can stay where it is while the conversation moves elsewhere

For years, banks treated the digital interface as an extension of the customer relationship.

The branch became online banking. Online banking became the mobile app. The app accumulated payments, cards, savings tools, notifications, product offers, support and increasingly sophisticated financial insights.

Agentic interfaces introduce another possible layer above the bank application.

A user who connects several accounts to one financial assistant does not need to open five apps to answer a question about total spending, available cash or portfolio exposure. The assistant can provide the first interpretation from a consolidated view while each underlying institution continues to control its own accounts and products.

The FCA’s comparison with mobile wallets is useful here. Its Mills Review asks whether AI intermediaries could capture part of the customer experience in a similar way: controlling an important interaction layer without becoming the traditional regulated provider underneath it.

The change begins before conventional customer attrition. A customer can keep every existing account while moving the financial conversation elsewhere.

Product discovery can move before the account does

Banking relationships generate moments in which customers become candidates for another financial product.

A current-account customer accumulates excess cash and might need savings. A cardholder begins travelling more often and may benefit from a different card. Repeated overdrafts can expose another credit need. A maturing deposit creates a fresh decision point.

Historically, the bank that already owns the customer interaction has an obvious opportunity to identify and present the next product.

An external financial agent changes where that process begins.

If a customer asks a general-purpose assistant where to place an extra £10,000, the bank’s savings account may become one candidate among products from several providers. If the customer wants to lower borrowing costs, the assistant can compare alternatives before any particular banking app has been opened.

The FCA explicitly anticipates agents comparing, recommending and switching banking, credit and insurance products. It also raises the prospect that financial firms may increasingly have to compete for the attention of AI systems optimized for price, value, risk or customer fit rather than simply relying on brand familiarity.

That changes the commercial value of owning the interface.

A bank can retain a customer while losing the privileged position from which the next financial need becomes visible.

A cross-provider agent can see a different customer

Banks possess deep information about the relationships they hold.

They know transaction history, repayment behaviour, balances, product use, servicing interactions and other information generated through the customer relationship.

Yet a customer may bank with one institution, invest through another, hold a mortgage elsewhere and use a separate credit-card issuer. No single bank necessarily sees the complete financial position.

An agent connected across those providers can assemble another kind of information advantage: breadth.

OpenAI’s financial experience illustrates the architecture. Connected accounts can be combined with goals and contextual information supplied by the user, so the interface can reason across several financial relationships rather than treating each account in isolation.

In a September 2026 speech on digital innovation and banking business models, ECB Supervisory Board Chair Claudia Buch noted that banks have historically benefited from information accumulated through repeated customer interaction and from offering several services under one roof. Digital innovation can alter that competitive advantage as new providers enter different parts of the banking value chain.

An external agent will not automatically know the customer better than the bank. The institution may retain much richer information about its own products, credit relationship and servicing history.

The difference is that the interface layer can become the only participant looking across several financial relationships at once.

Banks are testing two different interface strategies

One response is to accept that some customer conversations will happen outside the bank and make selected bank data usable there.

In July 2026, First Internet Bank introduced a permission-based connection that allows customers to use supported assistants including ChatGPT and Claude to analyse their First Internet Bank account activity through Model Context Protocol.

Customers can ask about spending, cash flow, subscriptions, balances and account activity from the assistant they have chosen. The connection is read-only: the external assistant cannot move money, initiate transactions or modify account settings.

The bank is therefore allowing part of the informational customer experience to occur somewhere else while retaining the account and execution layer.

Other institutions are trying to make the bank itself the preferred agentic interface.

In March 2026, Starling launched an agentic financial assistant inside its own app, covering conversational spending questions, saving goals, budgeting, support and selected banking tasks.

Japan’s SBI Sumishin Net Bank—now DOCOMO SMTB Net Bank—began testing its NEOBANK ai Agent in February 2026. The beta interprets natural-language intent, dynamically creates relevant banking screens and can support actions including transfers inside the bank’s mobile environment.

Visa is enabling the same bank-owned strategy through its AI Financial Assistant, announced in July 2026. The product is designed for banks to offer conversational financial guidance inside their own applications and under their own brands.

Trust strengthens the case for retaining a bank-owned surface. Customers already entrust financial institutions with deposits, identity data, payments and regulated obligations. A general-purpose agent may offer wider context; the institution holding the money retains a different kind of credibility when the action becomes consequential.

The likely contest is therefore not simply banks versus AI companies. Banks can expose controlled interfaces to external agents while also building or buying agentic experiences of their own.

Intent-first banking changes what the bank has to expose to software

Traditional mobile banking begins with navigation.

The customer signs in and selects a function: transfer money, view a card, open savings, search transactions or contact support.

An AI interface can begin with a problem instead.

A customer says cash feels tight before payday, asks whether a large purchase is affordable, wants to reduce fees or needs somewhere to hold surplus funds. The agent determines which financial information, product or capability is relevant to the request.

When the journey starts from intent, ownership of the menu matters less.

The bank’s product still matters, but increasingly the product must be discoverable and understandable to software.

Rates, fees, eligibility criteria, limits and product conditions need representations that an agent can interpret reliably. Servicing capabilities also need clear boundaries: which actions are available, what information they require and when the customer must return to the bank’s controlled environment.

A polished savings page remains useful to a human browsing the bank’s app. An external agent needs structured information it can reason over without inferring commercial terms from presentation copy.

The same problem appears in servicing.

An assistant helping a customer understand a card charge, replace a card or review a standing instruction needs to know which capabilities exist and what their limits are. That does not require banks to expose every internal system publicly, but it does push part of the customer proposition toward machine-readable representation.

Interface competition therefore reaches further into product architecture than screen design alone.

Losing the interface also means losing intent data

Customer interfaces generate information before a transaction occurs.

Searches reveal what a customer is considering. Navigation shows which products attract attention. Abandoned applications expose friction. Support conversations reveal confusion or emerging needs. Planning tools indicate future priorities.

When an external agent becomes the place where customers explore those questions, some of that intent no longer originates inside the bank.

The customer may spend twenty minutes with a financial assistant comparing savings options, assessing affordability and adjusting a goal before opening the chosen bank’s application for the final action.

From the bank’s perspective, the action appears.

Much of the decision process happened elsewhere.

This is a deeper consequence than losing screen time.

Personalization has traditionally benefited from observing the customer inside the institution’s own channels. If the richest expression of financial intent moves to an external agent, the bank may retain transaction data while losing part of the deliberation that explains why the transaction happened.

The interface can therefore become an information asset even when it does not hold the account.

The regulated relationship can remain underneath a thinner customer relationship

An external financial agent does not become the bank simply because it mediates the customer journey.

The deposit remains a bank liability. The loan remains a regulated agreement. The card issuer still operates the account. Complaints, fraud handling, financial-crime controls, disclosures and other obligations do not disappear because another interface helped the customer reach a decision.

The FCA’s Mills Review already raises the possibility that AI systems could perform functions resembling advice or intermediation while sitting outside traditional regulatory categories.

That tension reinforces the business-model shift.

Customer-interface control and regulated-product ownership can separate. Banks may continue carrying much of the capital, compliance and operational responsibility while another platform captures more of the customer’s attention and decision context.

This form of disintermediation does not require customers to close accounts.

It requires the regulated institution to become less central to the moments in which customers interpret their financial situation and decide what should happen next.

Banks need to compete in both owned and external interfaces

Treating agentic banking solely as an app-design problem would miss the strategic choice.

Banks increasingly need to decide how their proposition works in two environments.

Inside an owned interface, the institution can combine proprietary data, servicing capabilities, risk controls and product knowledge into an agent designed around its own relationship with the customer.

Outside it, the bank has to decide what an authorized external agent can discover, read or request without forcing every low-risk interaction back through a human-oriented application.

First Internet Bank and Starling illustrate the two directions clearly.

One lets outside assistants become a read-only financial-analysis surface. The other is trying to make the bank itself the conversational layer.

Those approaches can coexist.

A bank could keep consequential execution and its richest functionality inside an owned environment while making selected product information, account data and servicing capabilities portable to external agents.

The strategic question becomes less about preserving one universal front door and more about deciding where the bank must own the experience, where it can participate through another interface and where control needs to return to the institution.

FAQs

What does it mean for an AI agent to become a banking customer interface?

It means a customer begins financial tasks through an AI assistant rather than directly through a bank website or app. The agent may aggregate information, interpret needs, compare options or connect the customer to supported workflows while the underlying financial product remains with the bank.

Can a bank lose its customer interface without losing the account?

Yes. Current services already allow external AI assistants to analyse bank-account information while the account itself remains unchanged. The customer can use another interface for financial questions even though custody, execution and regulated servicing remain with the institution.

Why could AI agents change competition between banks?

Agents can potentially compare products across providers using price, value, risk or fit. That may reduce the advantage banks receive simply from already having the customer inside their application and increase the importance of product quality, structured information and agent discoverability.

Are banks building their own AI agents?

Yes. Starling launched an in-app agentic financial assistant in March 2026, while SBI Sumishin Net Bank began beta testing its NEOBANK ai Agent in Japan. Visa has also introduced infrastructure that lets banks offer conversational financial assistance under their own brand.

Will external AI agents replace banking apps?

Not necessarily. Authentication, high-risk transactions, disputes, regulated disclosures and complex servicing may still require bank-controlled environments. A mixed model is more plausible, with customers moving between general-purpose agents and bank-owned interfaces depending on the task.

The relationship can thin before it breaks

The customer does not have to leave for the bank to lose something valuable.

The account can remain open. Salary can still arrive. The card can still be used. The mortgage can remain on the balance sheet.

Meanwhile, the questions that precede financial decisions can migrate elsewhere.

An AI agent can become the place where the customer understands their position, discovers alternatives, compares providers and decides what should happen next. The bank then receives the transaction or servicing request after part of the relationship has already taken place outside its interface.

Banks still have formidable advantages: regulated status, trusted financial infrastructure, proprietary account data and the ability to execute consequential actions.

The emerging competition is over which institution or agent gets to interpret customer intent before those advantages are called upon.

A customer can continue banking with the same institution while no longer starting their financial decisions there.