US business banking operations handle check positive pay exceptions, ACH returns, wire verification and treasury management onboarding, all under heavy fraud pressure. AI agents can work routine ACH returns within limits, draft positive pay decisions and assemble treasury onboarding files. People approve wires, exposure limits and fraud decisions. OCC, Federal Reserve and FDIC expectations, including SR 11-7, govern the agents.
What do business banking operations leaders in the United States spend their time on?
Check fraud keeps operations busy. Businesses still issue checks, and criminals steal, alter and counterfeit them. Positive pay services match presented checks with the issue files businesses upload, and every mismatch becomes an exception the customer or the bank must decide on before a deadline that day.
ACH is the second load. Businesses originate payroll and supplier payments and collect from customers, and returns arrive for closed accounts, insufficient funds and unauthorised debits. Each return code has its own handling and timing rules, and a business with high unauthorised returns can become a risk to the bank.
Wires and treasury services are the third. High-value wires need verification, especially when instructions arrive by email, because business email compromise remains common. Onboarding a company to treasury services means setting ACH origination limits, assessing exposure and configuring users and approvals.
Where can AI agents help?
- ACH return handling (depth 3). For administrative returns such as closed accounts or invalid account numbers, the agent posts the return, notifies the originating customer through approved templates and updates their records within limits. Unauthorised-debit returns are routed to a person, because they may signal a problem with the originator.
- Positive pay exception drafting (depth 2). The agent compares the presented check image with the issue file, the payee and amount history and the check stock, and drafts a pay or return recommendation with its reasons. The customer or a bank officer decides.
- Treasury onboarding assembly (depth 2). The agent gathers the company's beneficial ownership certification, financials and expected volumes, calculates proposed ACH exposure using the bank's methodology, and drafts the setup. A treasury officer approves limits.
- Wire verification preparation (depth 1). For each wire above the threshold, the agent compares the instruction with the customer's history and flags new beneficiaries, changed accounts and unusual destinations before the call-back.
ACH return handling suits depth 3 because each return code maps to a defined action and the effects are reversible. Unauthorised returns are a different matter: they may point to an originator debiting people without consent, which is a risk decision. Our guide on how to deploy AI agents in a regulated business explains how to separate the two inside one agent.
What stays with people?
Wire release stays with people, as does the call-back verification itself. A wire sent to a fraudster is rarely recovered, and the bank's controls exist precisely to put a person in the loop.
Positive pay decisions on altered or counterfeit checks, ACH exposure limits, decisions to suspend an originator, account exits and suspicious activity reports all stay human. Each involves risk judgement and, often, a customer conversation.
The operations leader should own the agent's performance, with model risk management providing independent validation.
Which rules apply in the United States?
| Area | Who supervises | What it means for the agent |
|---|---|---|
| Model risk | OCC, Federal Reserve, FDIC | SR 11-7 guidance expects inventory, validation, monitoring and documented limits for models that inform decisions |
| Third parties | Prudential regulators | Interagency expectations on managing third-party relationships reach the agent's supplier |
| AML | Prudential regulators | Customer due diligence and beneficial ownership duties, following the FATF Recommendations |
| Financial privacy | Federal regulators | GLBA governs customer financial information the agent reads |
| State law | State regulators such as NYDFS | State privacy laws such as California's CCPA as amended by the CPRA may reach data on individuals |
An agent working ACH or check exceptions supports critical operations and belongs in continuity plans; our playbook on running business continuity tests with an AI agent shows how. There is no single federal AI law; our guide to how AI agents are regulated in the United States covers the rest.
This is an explanation of the rules, not legal advice; confirm how they apply to you with counsel.
How do you start in the first 90 days?
Weeks 1 to 4: choose administrative ACH returns. Register the agent in the model inventory, agree validation with model risk, and define the return codes it may handle. Our guide on how to deploy an AI agent in a business covers the groundwork.
Weeks 5 to 8: shadow on live returns, comparing every action with your team's. Validation reviews the results independently.
Weeks 9 to 13: certify at depth 3 for administrative codes only, then shadow positive pay drafting with a focus on altered checks. Retail operations in the US face different queues; see our brief on operations agents in US retail banking. The full method is in deploying AI agents: train, shadow, certify, widen.
Where Pryme fits
In Pryme, a wire instruction that arrives by email is stamped external and cannot authorise a payment at any depth. The wire preparation agent compares it with sealed records of the customer's past wires and hands a flagged case to the person making the call-back, with the verdict filed.
Questions people ask
Can the agent decide positive pay exceptions for the customer?
No. The customer decides through the bank's service, or a bank officer where the agreement provides. The agent drafts a recommendation with reasons.
Is an ACH return agent a model under SR 11-7?
If it makes decisions on the bank's behalf, treating it as one is prudent. Validation and monitoring should be proportionate to its impact.
Can the agent do the wire call-back itself?
No. The call-back exists to put a person in contact with the customer through a known channel. The agent prepares the facts for that call.