No, not by hand. Interest and contractual fees should be applied by the core banking system's rules, which are tested, versioned and the same for every customer. An AI agent can explain charges, check them and refund or waive them within set limits. Applying a discretionary charge stays with a named person at the bank.
Why shouldn't an agent apply charges itself?
A charge takes the customer's money. When the core system applies interest or a monthly fee, it does so under rules that were approved, tested and applied evenly to every account. When anyone applies a charge by hand, that discipline is gone. The charge depends on someone's judgement at that moment.
An agent applying charges by hand would make that problem worse, not better. It would act at volume, and a wrong reading of the terms would reach hundreds of accounts before anyone noticed. Refunding them later is possible, but the customers have already been overdrawn, missed payments or lost trust. That is a compensable action, not a reversible one.
There is also a fairness question. A charge applied to one customer and not another in the same position is hard to defend. The FCA Consumer Duty in the UK expects firms to deliver good outcomes for retail customers, and inconsistent charging works against that.
What may the agent do, and what stays with a person?
| Task | The agent | Stays with a person |
|---|---|---|
| Explaining a fee or interest charge | Explains it from the terms and the account history | Disputes the explanation does not resolve |
| Checking a charge was correct | Recalculates it from the terms and flags differences | Deciding that the system was wrong |
| Refunding a fee | Refunds within a set limit and policy, at depth 3 | Refunds above the limit or outside policy |
| Waiving a fee for hardship | Drafts the waiver with the reason | Approving it |
| Applying a discretionary fee | Never | A named person, if the terms allow it |
| Changing interest rates or fee tariffs | Never | Product governance |
| Correcting interest after a system error | Calculates the correction for all affected accounts | Approving and running the correction |
The refund row is where agents add most value. Customers often ask for a fee to be refunded, and many requests fall within a clear policy. An agent can handle those quickly and pass the rest on. See how to answer questions about fees and charges with an AI agent.
A worked example
Take a bank with an unarranged overdraft fee. A customer asks why they were charged.
The agent reads the account. A card payment took the account overdrawn by 15 for two days, triggering the fee under the terms. The agent explains this plainly, with the dates. The customer says their salary was delayed by their employer. The bank's policy allows one goodwill refund of this fee in twelve months for customers with no previous refunds. This customer qualifies, so the agent refunds the fee at depth 3 and records the policy it applied.
The same week, the agent's daily check finds that interest on a savings product was calculated on the wrong day count for 300 accounts after a system change. It calculates the shortfall for each, drafts the correction, and holds it for the product owner. The correction runs only after a person approves.
What controls make it safe?
- No write access to charges. The agent's role scope excludes posting fees or interest. It can post refunds within limits, nothing else.
- Policy-bound refunds. Each refund cites the policy rule that allowed it. Requests outside the rule are drafted for a person.
- Recalculation, not judgement. When the agent checks a charge, it shows its calculation from the terms, so a person can verify it.
- Batch corrections held. Any correction across multiple accounts is reviewed and approved before it runs.
- A full record. Each explanation, refund and correction is filed with its source and the bound that allowed it.
Final interest and fees on a closing account follow the same rule: the system calculates them and the agent explains them, as covered in whether an AI agent can close a bank account. For related questions, see whether an AI agent can answer questions about interest rates, and for the wider limits, AI agents in banking: what they may and may not do.
Where Pryme fits
In Pryme, refunding a fee and applying a fee are separate capabilities. A bank can certify the refund capability on its own past cases and set a limit; applying fees by hand would sit outside the agent's role scope. Every refund is filed with the rule that allowed it and the bound that decided.
Questions people ask
Can the agent apply a fee the customer agreed to, such as for a replacement card?
If the fee is set in the tariff and triggered by a customer request, the core system should apply it as part of the request. The agent triggers the request; the system applies the fee.
Can the agent stop fees on an account in financial difficulty?
It can flag the account and draft a hold on charges for a person to approve. Stopping fees is usually part of a wider support arrangement that a person should set up.
What if a customer was overcharged for a long period?
The agent can find the affected period and calculate the refund. Long-running overcharges may point to a wider problem across many accounts, so a person should decide the remediation.