US capital markets operations run on a one-day settlement cycle, so affirmation must finish on trade date. Teams also fix audit trail errors, close out fails and support the customer reserve computation. AI agents can chase affirmations, repair reporting errors within limits and track close-outs. People decide buy-ins and own regulatory findings. The SEC and FINRA set the rules.
What do capital markets operations leaders in the United States spend their time on?
Trade date is now the working window. With a one-day settlement cycle, allocations, confirmations and affirmations must be completed on the day of the trade, and exceptions caught by evening. Operations chases investment managers and custodians, fixes instruction mismatches and escalates anything likely to fail.
Reporting is the second load. Broker-dealers report order and trade events to a market-wide audit trail, and errors must be corrected within set times. Error rates are monitored by regulators, and repeated errors attract scrutiny. Books and records must be complete and retrievable.
Third, fails and customer protection. Fails to deliver on certain sales must be closed out by deadlines, or the firm faces restrictions. Broker-dealers compute, on a regular schedule, the reserve they must hold to protect customers' cash and securities, and operations supplies the data behind it.
Where can AI agents help?
- Affirmation chasing (depth 3). The agent tracks unaffirmed trades through the day, sends standard chasers to the right counterparty contacts and records responses. Requests to change economic terms are held.
- Audit trail error repair (depth 3). For rejected or flagged reports where the firm's own order record is clear, the agent corrects the field and resubmits within limits. Errors that suggest a systemic problem are held for a person.
- Close-out tracking (depth 1). The agent tracks each fail to deliver against its close-out deadline, flags those at risk and drafts the close-out instruction for a person to approve.
- Reserve computation support (depth 2). The agent assembles and checks the inputs to the customer reserve computation, identifies anomalies against prior periods and drafts the explanation for the financial operations principal.
Each of these uses runs against the clock. A chaser sent after a counterparty's cut-off is wasted, and a close-out flagged on the deadline day leaves no room to act. Give the agent each counterparty's hours and each deadline as sealed sources, and measure it on how early it raises an item, not only on whether it raises it.
As agents take on these queues, the team's structure changes. Someone must run the agents: watch their held items, retune briefs, manage certificates and report on their performance. The agent operations lead is a real job argues why that role needs naming, and our guide on how to organise an agent operations team sets out the structure. The method for introducing agents is in deploying AI agents: train, shadow, certify, widen.
What stays with people?
Buy-ins and close-out executions commit the firm's money and stay with people. So do decisions about who bears the cost of a fail.
The reserve computation is signed off by a qualified principal who must be satisfied it is correct. The agent's checks support that judgement; they do not replace it.
Responses to SEC and FINRA inquiries, decisions to self-report, and any finding that customer assets were at risk stay with named people and counsel.
Watch for one trap. An agent that corrects audit trail errors quickly can hide a broken upstream feed, because the errors keep being fixed and nobody asks why they keep appearing. Set a threshold: when the same error type recurs beyond it, the agent stops correcting, holds the queue and reports the pattern to a person, who fixes the cause.
Which rules apply in the United States?
| Area | Who supervises | What it means for the agent |
|---|---|---|
| Broker-dealer operations | SEC and FINRA | Settlement, close-out, customer protection, audit trail reporting and books and records rules apply to work the agent does |
| Bank-owned dealers and custodians | OCC, Federal Reserve, FDIC | SR 11-7 model risk guidance applies to models used in material processes |
| Financial privacy | Federal regulators | GLBA governs customer financial information |
| State matters | State regulators such as NYDFS | State privacy laws such as California's CCPA as amended by the CPRA may apply to personal data |
Records the agent creates, such as chasers sent and corrections made, may be books and records the firm must retain. There is no single federal AI law. The NIST AI Risk Management Framework offers a voluntary structure that maps well to supervisory expectations.
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 affirmation chasing. It is routine, high volume and central to meeting the settlement cycle. Name the agent operations lead and decide record retention for its messages.
Weeks 5 to 8: shadow the agent. It logs the chasers it would send; your team works as normal. Compare timing and accuracy, and check it never proposed changing economic terms.
Weeks 9 to 13: certify at depth 3, then shadow audit trail error repair. UK operations face a different rulebook and their own settlement change; see our brief on operations agents in UK capital markets.
Questions people ask
Can the agent execute a close-out buy-in?
No. It tracks deadlines and drafts the instruction. Execution commits the firm's money and stays with a person.
Are the agent's chasers subject to communications rules?
They are business communications and should be retained and supervised like any other. Use approved templates.
How many people does an agent operations team need?
It depends on the number of agents and their held-item volumes. Start with a named lead and grow as agents are added.