The end of inertia: Agentic AI’s disruption of retail and SME banking

| Article

The agent is here: “Hey TravelBot, book a seven-day itinerary for two people direct from San Francisco to Paris with refundable economy flights for under $2,000, and pay with the best credit card based on rewards.”

Agentic AI—technology that can perform tasks and solve issues largely on its own—will be able to handle complex requests like the one above, shifting AI from reactive helper to proactive financial agent for shopping, payments, and investing.

These developments are expected to shake up the economic foundations of finance, affecting billions in revenue and posing a threat to business models and revenue at banks, small and medium-size enterprises (SMEs), credit-card companies, and others. Pioneers willing to lead the way could see a game-changing upside, while those who don’t innovate will lose out. The changes will be felt first in markets where open banking has taken root.

Two revenue “engines”—credit cards and deposits—are especially exposed. Both rely heavily on customer inertia and brand familiarity. But AI agents won’t care about brand loyalty. They will optimize for outcomes. When logic, not habit, drives product selection, the rules will change.

OpenAI’s Operator Mode1 can already browse the web, make a hotel booking, and carry out multistep tasks. Start-ups like Manus AI2 and tools like Perplexity3 are pushing even further—working across business platforms, interpreting user intentions, and taking action. Google4 has introduced an AI Mode shopping experience that delivers personalized discovery and price tracking via Gemini AI. To facilitate these moves, payments players such as Visa,5 Mastercard,6 and PayPal7 have recently released new capabilities to facilitate agentic commerce, allowing agents to trigger payments autonomously on behalf of users.

A simple voice request—“Hey SaveBot, please maintain my checking-account balance at $1,000 and move excess cash to my high-yield savings account, and top up from my savings if I dip below $500”—could become routine. Before long, we may not even have to ask. An AI agent will notice payment or investment opportunities via email, text, or app alert, act automatically, and tell us what it did afterward. The technology to make this happen is already here. Griffin8 has begun building an “agentic-first” banking core for its fintech partners.

Financial institutions must prepare now to understand where their value is at risk, which control points will matter, and how they can remain relevant in an agent-mediated world. In this article, we review the financial stakes; the disruption agentic AI will cause for deposit institutions, credit-card issuers, and others; the opportunities that lie ahead for those willing to innovate; and how institutions can best prepare for the coming agentic banking economy.

The value at stake

The global payments industry generates more than $2.7 trillion in annual revenue from sources such as deposit spreads, card economics, cross-border flows, and merchant acquiring.

Roughly half of that revenue pool is concentrated in two broad areas:

  • Net interest income on deposits. The spread between the interest paid out on deposits to account holders and the interest earned on banks’ investments
  • Consumer-card economics. The revenue from interchange, interest on revolving balances, and rewards that aren’t redeemed

Both revenue engines rely on a simple truth: Most consumers don’t optimize every dollar, every day. Margin accrues in the gap between what consumers could do and what they actually do.

Agentic AI breaks that equation. Even partial adoption of the technology can meaningfully compress those margins or shift revenue to other businesses.

Deposits and liquidity: The inertia dividend

Deposits—from consumer checking to SME operating accounts—have long powered bank profitability. Globally, net interest income accounts for roughly 30 percent of retail-bank profit.

Most consumers don’t notice the interest rate they’re receiving, or they lack the time, tools, and incentive to optimize interest returns on their deposits. They instead are more focused on the convenience of a single low-maintenance banking hub that waives fees and integrates ATMs, bill payments, and wealth portals.

The national rate that banks pay on interest-bearing checking and savings accounts in the United States was just 0.07 and 0.38 percent, respectively,9 in June 2025, even though top online saving rates exceed 4 percent. Further, over the past five years, non-interest-bearing demand-deposit account balances grew at a CAGR of 28 percent, compared with just 3 percent for interest-bearing equivalents.

Agentic AI systems flip that logic. They can monitor balances in real time, compare returns across institutions, sweep idle cash into higher-yield accounts, and then sweep cash back to a checking account in time for bills. That allows more of the spreads once captured by banks to go to account holders.

Recent interest-rate surges (the 2022 flight to high-yield saving10) proved consumers will move cash on their own when a clear rate gap emerges. China’s Yu’e Bao,11 a money-market fund launched in 2013, illustrates how fast the trend can grow. By 2017, users had shifted $268 billion in idle cash into the money-market fund (see sidebar “Yu’e Bao: When optimization scales”). Platforms like Raisin12—a mainly EU- and UK-focused marketplace for deposits—have captured more than $80 billion in deposits, placing the cash in higher-yield savings accounts at over 250 banks. These examples offer proof that marketplace optimization can happen, even before agents take the wheel.

SMEs are already harnessing API-driven treasury automation to optimize cash and foreign exchange (FX) in real time. For example, multiple SMEs have adopted cash-management platforms that automate daily reporting, forecasting, sweep operations, and even foreign currency hedging. The platforms put real-time bank data at the customer’s fingertips. Agentic AI would go further, integrating these capabilities into continuous, preference-driven treasury operations.

The author William Gibson once said: “The future is already here—it's just not evenly distributed.” Thus geography will shape the speed of change, as some global regions embrace open banking before others. Agentic AI’s impact on deposits will be felt first in Europe and the United Kingdom, where Payment Initiation Service Provider (PISP) and Account Information Service Provider (AISP)13 licensing, coupled with low-cost instant-payment rails, make switching accounts relatively easy.

The European revenue pool from deposits is worth more than $100 billion. If just 10 to 20 percent of consumers adopt agent-driven cash sweeps, bank net-interest margins could tighten by 30 to 50 basis points (bps).

Agents will be able to go further than Yu’e Bao by continuously reallocating funds across multiple banks and rate tiers, guided by yield, settlement timing, and user-set risk limits. For banks, the threat is clear: loss of low-cost funding and pressure on liquidity assumptions as deposit stickiness erodes. For account holders, the upside is equally clear: finally capturing interest income long left on the table.

Credit cards: Optimized out

Credit cards generate extraordinary revenue—$234 billion in 2024—through a blend of interest income from customers who carry a balance,14 interchange fees, annual and penalty fees,15 and unredeemed rewards (exhibit).

Consumer general purpose credit card revenue breakdown.

Much of this lucrative model is held together by consumer inertia. Recent survey data16 suggests that more than 20 percent of cardholders didn’t redeem any rewards in the prior 12 months. “Forfeiture incidence”—the share of accrued rewards points lost through account closure or expiration—ranges from 3 to 5 percent annually, according to the Consumer Financial Protection Bureau.17

AI agents, however, can help passive consumers by automatically directing spending to the best card in real time, triggering new applications to snag a better card deal, and rolling balances to another card before promotional rates expire.

Some of this is already happening. Klarna’s Money Story feature nudges spending limits based on transaction data;18 Curve’s Smart Rules feature lets users set category- or amount-based routing on its Curve card;19 and the Apple Wallet app introduced a circular-dial payment slider that lets cardholders instantly adjust their monthly repayment amount on their Goldman Sachs–issued Apple Card.20 These micro automations foreshadow full delegation of card tasks.

Account-to-account checkout threat

As open banking gains traction, AI agents can launch account-to-account (A2A) payments at checkout—skipping card interchange systems and undermining rewards economics. In the European Union, interchange fees are capped at 0.2 percent for debit and 0.3 percent for credit cards—limiting arbitrage opportunities.21 But in North America, interchange fees typically range from 1.30 to 3.25 percent, providing a saving incentive for real-time card switching.22

Such dynamic optimization hinges on transactions going through a digital wallet, since physical-swipe transactions must be routed through interchange systems. That gives platforms like Apple Pay a head start as an AI agent conduit. Even modest agent-led A2A adoption could divert a meaningful share of interchange fee and interest revenue.

Momentum meets reality

In sum, what was once passive in banking is becoming programmable—and more dynamic. Sticky deposits become liquid; reward economics become transparent. Even modest adoption can meaningfully dent deposit spreads and card margins. And deposits and cards are merely the opening act; personal-loan pricing, installment fees, and merchant acquiring could come under similar pressure as agentic optimization spreads.

Yet adoption won’t be friction-free. Agentic finance collides with several real-world constraints that will determine how fast it grows and who benefits first:

  • Credit-score boundaries. Today’s models dock points against credit scores for new inquiries and short account age, limiting how often agents can switch cards or open balance-transfer offers without hurting consumer scores. Agents will need embedded credit-awareness logic until alternative cash-flow or behavioral scoring models emerge.
  • Agent errors. Early versions of agentic technology are likely to make mistakes. An agent might pay the wrong bill or improperly shift funds, for example, reflecting a misinterpretation of a user’s preferences or a software bug.
  • Trust and transparency. Delegating meaningful financial actions demands more than a “black-box” AI. Agents must clearly communicate with users, provide real-time alerts that can be overridden, and keep audit trails that let users—and regulators—verify exactly what happened and why.
  • Liability and regulation. Clear regulations to enforce accountability for AI actions are still developing. The EU AI Act23 (approved in 2024 and enforceable in 2026) classifies agentic finance tools as “high risk,” requiring AI explainability, human controls, and third-party audits. In the United States, the CFPB’s Section 103324 rulemaking could mandate standardized agent access to customer-authorized data, setting a compliance framework for regulated API usage. Individuals and institutions will be reluctant to delegate high-stakes functions to agents until a clear accountability framework is in place.
  • User security concerns. Consumers are understandably wary of handing over account credentials to a bot. Agents need tokenized, zero-trust architectures that grant the minimum permissions required and automatically revoke access if anomalous behavior is detected. Multifactor reauthentication—especially for large or out-of-pattern transactions—will at a minimum be needed to gain consumer trust.
  • Fraud and anti–money laundering (AML). Rapid sweeps and multiaccount orchestration can trigger suspicious-activity alerts or the suspicion of mule-account schemes. To satisfy regulators, agents must have embedded velocity caps, periodic Know Your Customer (KYC) revalidation, and real-time AML monitoring.
  • Operational rails. Modern instant-payment systems like FedNow25 in the United States and Europe’s TIPS/SEPA26 Instant have all but eliminated settlement lag, enabling agents to sweep funds and make payments in real time. Meanwhile, early pilots of stablecoin-based settlement (for example, Visa–USDC27 and Circle28 treasury integration) point to subsecond, programmable transfers—though such rails also introduce on-chain KYC and token-governance considerations that issuers will need to address. Many institutions, however, still rely on legacy rails such as Automated Clearing House, which may slow agentic adoption.

None of these hurdles will stop agentic finance, but they will set the tempo and shape the early winners.

Control points in the agentic economy

The locus of value in retail and SME banking and payments will shift as agents handle more financial decisions. Competitive advantage is likely to concentrate around five key control points:

  • Credentialing and identity. Agents need secure, user-granted tokens before they can fetch balances or initiate transactions across multiple institutions. Firms that already manage high-trust credentials—whether through OAuth2-based open banking APIs in Europe (PSD2) or network tokenization services like Visa Token Service in the United States—start with a clear advantage. Likewise, Big Tech identity providers and banks offering delegated-consent flows can provide a control gate for every downstream action. Success, therefore, hinges on:
    • Zero-trust architectures that never assume persistent access.
    • Dynamic consent via standardized protocols (for example, OAuth2/OpenID Connect).
    • Continuous audit trails to satisfy both customers and regulators—with companies mastering these elements able to control the first, and most critical, choke point in agentic finance.
  • Trust and liability wrappers. Businesses that build agent-compatible products, offer embedded guardrails, and share accountability may become preferred partners in the ecosystem. Trusted brands could agree to assume liability for a fee, and in doing so accelerate user adoption of agentic technology.
  • Merchant and platform integration. Real-time optimization—for example, routing spending, invoking offers—will depend on deep integration at the checkout point, both online and offline. Today most transaction acquirers have limited insights into consumer purchasing behavior, meaning software-centric providers may have the most to gain.
  • Decisioning logic. Agents will need to compare rates, features, rewards, and credit implications in milliseconds. Platforms that build agent-readable, optimization-friendly products for their aggregator platforms and their own websites will have a better chance of getting selected by agents. Today’s existing comparison websites may be the initial winners, having already assembled data and information with clear affiliate marketing monetization logic.
  • Behavioral data and intent signals. The first generation of agents will need to be instructed, while subsequent generations will be able to infer user needs before they are expressed. This learning dynamic will reward institutions that have developed better insights into user needs. Thinking like a consumer app means orchestrating every touchpoint—push notifications, in-app suggestions, contextual prompts—so agents can prompt dynamic nudges that keep users in a continuous optimization loop.

What should businesses do next?

The shift to agent-mediated finance is an inflection point. For banks, thriving in this new era means reengineering products around performance, not loyalty. Here are three universal pillars for developing a successful strategy:

  • Product strategy. Audit every offering for inertia dependence. If an agent was making decisions on a customer’s behalf today, would your product still win?
  • Technical infrastructure. Make products machine-readable and optimization-friendly, using standardized APIs, transparent pricing logic, and rich metadata that agents can parse.
  • Distribution and engagement. Decide whether—and how—to integrate into third-party agent ecosystems or build proprietary ones. In either case, products must be easy for consumers to understand and control.

In the short term, banks must brace for a new competitive tempo. Relevance must be re-earned moment by moment—every time an agent refreshes its rankings. As they develop their strategy (see sidebar “Five strategic questions shaping the agentic future of finance”), banks can get agent-ready by publishing real-time rate and liquidity APIs, launching savings subaccounts designed for automatic sweeps, and sharing a portion of incremental yield with customers to keep balances on-platform.

Other institutions can likewise position themselves ahead of the disruption:

  • Card issuers face erosion of interchange and interest income when agents rotate cards or default to A2A rails. Issuers could respond by translating rewards structures and balance-transfer rules into machine-readable formats; exposing always-on preapproved credit offers via API; and piloting experiential benefits—think VIP event access, surprise travel upgrades, or curated member‐only experiences—that an algorithm can detect but only a customer can truly appreciate.
  • Wallet providers and superapps must guard against disintermediation if agents integrate directly with bank APIs. Instead, they can position themselves as the go-to operating system for agents: Publish developer tool kits, bake in transparent consent and override controls, and secure the user’s intent—whether through voice, chat, or device-level widgets.
  • Card networks face a threat as A2A rails threaten to bypass them entirely. Networks are already responding by running instant payment rails, tokenizing credentials across multiple rails, and packaging premium friction-reducing services such as fraud coverage, dispute resolution, and data insights that agents can invoke by default.
  • Merchant acquirers and payment service providers (PSPs) operate in a world where agents run mini-auctions at checkout, picking the least-cost rail in milliseconds, much like today’s online advertising auctions. To stay relevant, they should expose real-time fee quotes via APIs, surface agent-visible promos or rebates, and integrate with wallets to offer built-in, least-cost routing logic.
  • Cross-border specialists will see FX spreads squeezed as agents compare prices across providers in real time. Winning companies will publish all-in landed-cost feeds, guarantee settlement time frames, and automate refunds when service-level agreements aren’t met—turning reliability into a competitive differentiator.
  • Big Tech platforms can cement leadership by embedding agentic commerce into voice assistants, which already have strong consumer use—then monetize referral flows, data insights, or subscription services.
  • Merchants themselves now face a micro-auction at every checkout. Large retailers can drive agent preference by offering dynamic, API-driven coupons and loyalty hooks; SMEs can plug into marketplace-agnostic offers through APIs or partner with acquirers that expose agent-friendly promotional tools.

Across the business landscape, there is a common thread: Make it effortless for an optimization engine to discover, trust, and select your product—while giving the human a clear reason to feel good about that choice. Agentic AI marks a shift from user-driven to system-mediated decision-making. It also marks a change in retail and SME finance from brand-led loyalty to performance-led selection. Institutions that adapt first can embed trust and visibility in the logic layer that could soon make decisions on every customer’s behalf. Those that wait risk becoming invisible balance-sheet utilities. This is not just about protecting margin; it is about reimagining retail and SME finance. The future will favor those who perform, not those who are familiar. Agents are likely to prefer top-performing products. Will they find yours?

Explore a career with us