A bank's back office is a set of queues: alerts, returns, disputes, documents, reconciliations. Each queue has a written procedure, a deadline and a person who answers to an examiner. That structure is what makes the work a fit for an AI employee, and it is also why the list of candidate jobs is longer than most operations leaders expect.
This guide is for a COO or head of operations at a US bank or credit union. It lists 20 jobs by department. For each one we note what goes in, what the AI employee does and where a person signs off. For the wider picture, see our pillar guide to AI employees in banking and financial services.
What are the main AI employee use cases in banking operations?
How is this list organized?
The 20 jobs sit in six departments. Each entry names the inputs and systems the job touches, the work the AI employee does and the human sign-off. Where we have a deeper guide, it is linked. A job with no link has no dedicated Zamp guide yet.
Two conventions apply throughout. The AI employee works from a written procedure and cannot act outside it. Anything it cannot resolve goes to a person with the evidence already assembled.
Retail and deposit operations
1. Periodic KYC refresh
Inputs are the customer file, risk rating, last review date and any new documents. The AI employee finds accounts due for review, requests missing items, re-screens the customer and drafts the refreshed profile. A reviewer approves any rating change. See our guide to how KYC automation works and the glossary entry on perpetual KYC.
2. Account maintenance requests
Address changes, signer updates, title changes and card-limit requests arrive by branch form, secure message or mail. The AI employee reads each request, validates it against the core system record and identity checks, prepares the update and flags mismatches. A deposit operations analyst approves changes to signers and ownership.
3. Dormant accounts and escheatment preparation
Inputs are the core account file, last customer-initiated activity dates and each state's unclaimed property rules, which differ by state. The AI employee builds the dormancy list, drafts owner-contact letters on the required schedule and assembles the remittance file. The compliance owner approves the state filings and the remittance.
4. Reg E dispute casework
Inputs are the customer's notice, transaction history and card or ACH network data. The AI employee logs the claim, tracks the clock under 12 CFR 1005.11, gathers evidence and drafts the provisional credit and the customer letter. An investigator signs the decision. Details are in our guide to Reg E dispute investigation automation.
Payments operations
5. ACH returns and Notifications of Change
Inputs are the daily return and NOC files from the ACH operator and the original entry records. The AI employee matches each return to its entry by trace number, applies change codes where policy allows and tracks return rates per originator. A payments lead approves any re-presented entry and any originator action. See ACH returns and NOC handling.
6. Wire repair and exception handling
Inputs are wire messages that fail validation, the payment system queue and customer or beneficiary records. The AI employee identifies the defect, proposes the correction from the source documents and routes the repaired wire for release. A second person approves release under dual control. See wire payment repair and exception handling.
7. Check and deposited-item exceptions
Inputs are item images, the deposit record and the account file. The AI employee flags amount mismatches, duplicate presentment and missing endorsements, then prepares the adjustment or return. An analyst approves adjustments above the bank's own threshold.
Lending and credit operations
8. Commercial loan onboarding
Inputs are the signed application, entity documents, beneficial ownership information and the loan origination system. The AI employee checks the package for completeness, runs the required identity and screening steps and sets up the file for booking. Credit and compliance staff approve exceptions. See commercial loan onboarding automation.
9. Loan document review
Inputs are the loan file and the checklist for the product. The AI employee classifies each document, extracts key terms, checks that figures agree across documents and lists what is missing or stale. An underwriter makes the credit decision. See loan document review with AI.
10. Financial spreading and covenant tracking
Inputs are borrower financial statements and the covenant terms in the credit agreement. The AI employee spreads the statements into the bank's template, computes each covenant and drafts the compliance memo. A credit analyst reviews and a relationship manager decides any waiver.
11. Insurance and collateral document tracking
Inputs are the collateral file, insurance certificates and expiry dates. The AI employee finds lapsed or expiring coverage, requests renewals from borrowers and logs the response. A loan servicing lead approves force-placed coverage.
BSA/AML and fraud operations
12. AML alert triage, first-level preparation
Inputs are the monitoring alert, customer profile and transaction history. The AI employee pulls the context, notes the likely typology and drafts a disposition for each alert. An analyst closes the alert or escalates. Our guide to AML alert triage covers false-positive reduction, and the AML automation hub covers the wider program.
13. Sanctions hit review preparation
Inputs are the screening hit, the customer or payment record and the list entry. The AI employee compares names, dates of birth, addresses and identifiers and writes up why the hit looks like a match or a false positive. A sanctions officer decides every disposition. See OFAC sanctions screening automation.
14. Currency transaction report preparation
Inputs are cash transactions aggregated by customer for the business day. A CTR is required for currency transactions over $10,000 under 31 CFR 1010.311. The AI employee aggregates, populates the report from the customer record and flags missing data. A BSA analyst reviews before filing. The glossary has an entry on the currency transaction report.
15. Fraud alert review
Inputs are card, online banking and account-takeover alerts, device data and recent customer contact. The AI employee assembles the timeline, checks it against the customer's normal pattern and recommends a hold or release. A fraud analyst makes the call on anything that restricts the customer. See fraud alert review with AI employees.
Compliance and regulatory operations
16. CFPB complaint intake and response
Inputs are complaints from the CFPB portal, the branch and email, plus the account history. The AI employee logs and categorizes each complaint, pulls the relevant records and drafts the response for review. A compliance officer approves every response. See CFPB complaint handling automation.
17. Third-party due diligence
Inputs are the vendor questionnaire, SOC reports, financials, insurance certificates and the bank's risk tiering. The AI employee checks the package against the policy, notes gaps and drafts the review summary. The vendor owner and risk committee approve onboarding. See third-party risk management for AI vendors and the glossary entry on TPRM.
18. Exam and audit evidence collection
Inputs are the examiner or auditor request list, policies, logs and sample populations. The AI employee maps each request to its source, pulls the evidence into a folder and tracks what is outstanding. A compliance lead reviews the package before it leaves the bank.
Finance and accounting
19. General ledger reconciliation
Inputs are the GL, core system balances, payment network settlement files and subledgers. The AI employee matches items, ages the breaks, proposes journal entries and drafts the explanations. A controller approves entries and clears the aged items. See bank reconciliation automation.
20. Call report data preparation
Inputs are the trial balance, loan and deposit subledgers and the prior quarter's report. The AI employee maps balances to FFIEC call report schedules, runs the validation edits and produces a variance explanation file. The financial reporting manager reviews and certifies the filing.
How do you pick the first job?
Pick a job with a written procedure, a clear volume, one owner and a cheap error. Returns and NOCs, account maintenance and reconciliation breaks fit that description. Alert triage, sanctions and anything that restricts a customer come later, once the review process has proven itself.
Run the AI employee next to your analysts first. Review every disagreement and widen scope only when disagreements are rare. Examiners will ask who approved what, so decide the approval gates before the first case, not after.
For credit unions the same list applies with a smaller team, which makes the first job more important. See our guide to AI employees for credit unions and the roundup of AI agents for banks and credit unions.
How Zamp Runs Banking Back-Office Jobs
Zamp builds AI employees that run jobs end to end with human approval gates and a full audit trail. Each job follows an Agent Operating Procedure, the written document that sets the steps, the systems the agent may read, the fields it may change and the cases it must hand to a person.
When a case falls outside the procedure, the agent marks it Needs Attention and sends it to a reviewer with the evidence attached. Reviewer corrections change the procedure through a tested and reviewed process, and the model itself is not retrained. Banks with data residency requirements can run the platform inside their own cloud account or on premises. Read the deployment models guide for the options, and our note on four-eyes approval for AI agents for how dual control works with an agent in the loop.
