Family trusts make onboarding the hardest queue in Australian business banking operations, ahead of signatory changes, payment exceptions and small business loan servicing. AI agents can assemble trust onboarding files, draft signatory changes and clear routine payment exceptions within limits. People approve anything that changes who controls money. APRA's CPS 230 and the Privacy Act 1988 shape how agents run.
What do business banking operations leaders in Australia spend their time on?
Onboarding is the biggest queue. Australian small businesses often operate through a company acting as trustee of a family or unit trust, so the bank must identify the trustee company, its directors, the trust deed, the appointor and the beneficiaries before an account opens. Files stall when the trust deed is missing pages or the corporate trustee changed.
Servicing comes next. Signatory changes, new authorities for online banking, address updates and account closures arrive by form, email and branch visit. Each change to who can move money needs careful verification, because criminals impersonate business owners to add themselves as signatories.
Payments operations handle real-time payment exceptions, returned payments, payee name mismatches and urgent same-day requests. Lending operations settle and service small business loans. Behind it all, the operations leader must keep critical operations within the tolerance levels the bank set under CPS 230.
Where can AI agents help?
- Trust onboarding files (depth 2). The agent reads the trust deed, company extract, director identity results and the KYB checks, builds the control map from trustee to appointor to beneficiaries, and drafts the onboarding file with missing items listed. A person approves the account.
- Signatory change drafting (depth 2). The agent compares the change request with the authority on file, checks that the people signing the request are authorised to make it, and drafts the change. A person verifies by call-back to a known number and approves.
- Payment exception handling (depth 3). For routine exceptions, such as a returned payment with a clear reason code, the agent reconciles, notifies the customer through approved templates and closes the item within set limits. Exceptions involving amounts above the bank's limit or suspected fraud are held for a person.
- Loan servicing tasks (depth 3). The agent updates repayment schedules after agreed variations and issues standard statements, within bounds that exclude any change to the credit decision itself.
Each capability earns its depth on the bank's own cases. The guide to deploying AI agents: train, shadow, certify, widen describes how an agent moves from drafting to routine execution only after shadowing proves it.
What stays with people?
Anything that changes who controls a business's money stays with a person: approving a new signatory, adding an online banking administrator, changing a mandate. The fraud risk is too high and the harm hard to undo.
Account approval for complex structures, decisions to decline or exit a customer, credit decisions and hardship arrangements for small businesses stay human. Industry code protections for small businesses, where the bank subscribes to them, set expectations a person must apply.
Ownership of the agent itself must be clear. Our answer on who should own an AI agent in a business recommends that the operations leader who owns the process owns the agent's performance, with risk and technology in support.
Which rules apply in Australia?
APRA supervises banks prudentially. CPS 230 on operational risk management requires banks to identify critical operations, set tolerance levels for disruption, manage material service providers and test business continuity. An agent working payment exceptions supports a critical operation; its supplier is likely a material service provider; and its failure must be covered in continuity plans. Our playbook on running business continuity tests with an AI agent shows how to include it.
ASIC regulates conduct, including for small business customers in some contexts, and AML/CFT duties follow the FATF Recommendations, including identifying beneficial owners of companies and trusts. The Privacy Act 1988, administered by the OAIC, governs personal information of directors, signatories and beneficiaries the agent reads.
Australia has no AI act. The government's voluntary AI safety standard gives useful guidance on accountability, testing and human oversight.
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?
Month one: write the business case. Pick the trust onboarding queue, measure its current turnaround and rework, and state what the agent will do and what it will not. Our guide on writing a business case for AI agents helps.
Month two: run the agent in shadow on live onboarding files. It drafts; your team works the file as normal; you compare. Track missed beneficiaries and misread deeds.
Month three: certify at depth 2 for trust onboarding, then start shadowing payment exceptions. Map the agent into your CPS 230 critical operations register before it touches live exceptions. France's business banking operations differ in rules and payments; see our brief on operations agents in French business banking.
Questions people ask
Can an agent approve a new business account?
No. It can assemble and check the file. Approval, particularly for trusts and multi-layered companies, stays with a person.
What happens if the agent's supplier goes down?
Your CPS 230 continuity plan must cover it. Work falls back to people, and you should test that fallback before relying on the agent.
How does the agent handle family trust deeds that have been varied?
It should read every variation in order and show the current position. If any variation is missing, it lists it as a gap for a person to chase.