A deposit franchise is, at its core, a behavioural asset. Customers leave money in current accounts not because the rate is competitive but because moving it is effortful. Inertia is the product. Switching costs are the moat. This is about to change.
McKinsey's analysis of agentic AI in retail banking identifies a structural shift: when customers delegate financial decisions to AI agents, the friction that protects deposit balances disappears. An agent optimising for yield will not leave money idle in a current account paying 0.1% when a savings product at another institution pays 4.5%. It will move the money. Automatically, repeatedly, and without the emotional attachment to a banking brand that has historically kept balances in place.
The Mechanics of Deposit Erosion
BCG's 2026 research on agentic AI in retail banking describes three emerging economies. The first, which they call "Adaptive Customer Experiences," enables real-time interface personalisation. The second, "Agentic Twins," represents customers with delegated authority across financial relationships. The third is the most consequential for deposit franchises: AI agents acting with secure, user-granted tokens to initiate transactions across multiple institutions.
The deposit erosion pathway runs as follows. A customer authorises an AI agent to manage their cash. The agent maps all available deposit products across the market, including notice accounts, money market funds, and overnight sweep facilities. It calculates the optimal allocation given the customer's liquidity needs, tax position, and FSCS protection limits. Then it moves the money.
For a bank that has built its treasury model on the assumption of sticky retail deposits funding its lending book, this is a structural risk. Not because every customer will adopt agentic cash management immediately, but because the customers most likely to adopt it early are also the ones with the largest balances and the lowest switching costs: mass-affluent and high-net-worth individuals whose deposits are disproportionately valuable.
What the Banks I Work With Are Asking
Three questions come up consistently in board conversations about agentic banking.
The first: how fast? The honest answer is that nobody knows. Meow Technologies launched what it calls the first agentic banking platform for AI agents in early 2026. Open banking rails make multi-institution money movement technically feasible today. But consumer adoption of delegated financial authority is a trust problem, not a technology problem. In the cases I have been involved in, the most useful planning assumption is that 5 to 15% of mass-affluent customers will be using some form of agentic cash management within three years.
The second question: what is the P&L impact? This depends on the bank's funding mix. A bank that funds 60% of its lending book with retail deposits and earns a net interest margin of 180 basis points on those deposits has material exposure. If 10% of those deposits become "agentic" and move to higher-yielding alternatives, the funding gap must be filled from wholesale markets at a higher cost. The treasury team can model this, but only if the scenario is on the agenda.
The third question: what do we do about it? This is where the strategic choice lies.
Three Strategic Responses
Response 1: Compete on rate. The simplest response is to match the yields that agents will find elsewhere. This is expensive and margin-destructive, but it keeps deposits in-house. It also changes the economics of the deposit franchise permanently: if balances only stay because the rate is competitive, the behavioural moat is gone and the bank is competing on price alone.
Response 2: Become the agent's preferred platform. Banks that offer delegated-consent flows, open banking APIs, and agent-friendly interfaces can position themselves as the hub through which agents operate rather than the institution agents optimise away from. Oracle's February 2026 launch of an agentic banking platform signals that core banking vendors see this as a viable architecture. The bank becomes infrastructure rather than a product provider. The margin profile changes, but the relationship survives.
Response 3: Build proprietary agentic services. Rather than waiting for third-party agents to move customer money, some banks are building their own agentic cash management tools. The advantage is control: the bank's agent optimises across the bank's own product set first, using external products only when the customer's interest clearly requires it. The risk is regulatory: a bank's own agent that consistently recommends the bank's own products over better alternatives will attract conduct scrutiny under Consumer Duty.
Most banks will need a blend of all three. The proportions depend on the bank's funding structure, customer mix, and regulatory appetite. But the strategic conversation needs to start now, not when the first material outflows appear in the treasury report.
The Conduct Dimension
The FCA's Consumer Duty creates an interesting tension in the agentic deposit world. If a bank builds an AI agent that manages customer cash, the agent must deliver "good outcomes" for the customer. This means recommending products that serve the customer's interest, even when those products are at a competitor institution. A bank-operated agent that systematically favours in-house products will fail the fair value and consumer understanding outcomes under Consumer Duty.
Conversely, if a third-party agent moves a customer's deposits to a higher-yielding product at another institution, and that product carries higher risk or lower FSCS protection, the question of who is responsible for the customer's outcome becomes complex. The FCA's Mills Review is examining exactly this question: when an AI agent performs a function that looks like regulated advice, who bears the conduct obligation?
For banks, the practical implication is that any agentic cash management strategy must be designed with Consumer Duty compliance from day one. Retrofitting conduct obligations onto an agentic system that was designed for yield optimisation alone is significantly more expensive than building it in.
Planning for the Transition
The deposit franchise will not disappear overnight. Behavioural inertia is real, trust in AI agents is still building, and regulatory guardrails will slow adoption. But the direction of travel is clear, and the banks that prepare now will have structural advantages over those that treat agentic banking as a distant hypothetical.
The starting point is a treasury scenario analysis: what happens to the funding model if 5%, 10%, or 20% of retail deposits become rate-sensitive overnight? The answer determines the urgency of the strategic response, and most treasurers I work with find the answer more concerning than they expected.
*To discuss how the 90-Day AI Acceleration programme can help your bank prepare for the agentic deposit transition, contact the Value Institute.*