Yes, for the actions they approve on customer accounts. When an AI agent drafts a chaser, proposes an account stop or suggests a write-off, the credit controller who approves it owns the result. Routine chasers sent within agreed bounds trace back to whoever set those bounds. The record shows which applies to each action.
What does a credit controller approve when an agent does the chasing?
Credit control is a steady flow of small decisions with commercial weight. An agent can reconcile receipts against invoices, send polite reminders on a schedule, log promises to pay and draft escalation letters. It can also propose that an account be put on stop, or that a small residual balance be written off.
The controller's approvals are the ones that affect the customer relationship: stopping supply, sending a final demand, writing off or disputing a balance. Those are the decisions that cause the most harm when they are wrong, and a key customer put on stop by mistake is not easy to undo.
What may the agent do, and what stays with the controller?
| Action | Usual depth | Who decides |
|---|---|---|
| Match receipts to invoices | Depth 2, drafts the match | Controller confirms exceptions |
| Send scheduled reminders | Depth 3, within bounds | The bounds the controller set |
| Draft a final demand | Depth 2 | Controller approves |
| Put an account on stop | Held | Controller, often with sales |
| Write off a small residual | Held, or depth 3 under a set limit | Controller or limit owner |
| Agree a payment plan | Held | Controller |
The agent should refuse to send anything threatening legal action unless the template and the decision have both been approved.
What does the record show about who decided?
Each action carries its verdict, the bound that decided it and its sources. The sales ledger and bank feed are sealed. A customer's email saying "payment sent yesterday" is external: useful, but not proof. A remittance advice from the customer's bank is closer to signed.
When a controller approves a stop, the record shows the account balance, the unallocated cash, the last contact and the sources behind each. If unallocated cash was sitting on the account and on screen, the controller who approved the stop owns that mistake.
An example: the stop that should not have happened
Take a building supplies wholesaler. A trade customer is 60 days overdue on 18,000. The agent proposes an account stop and holds it. On the same morning, 18,000 has arrived in the bank feed, but with a reference the agent could not match, so it sits as unallocated cash. The agent's draft mentions the unallocated amount in a note at the bottom.
The controller approves the stop. The customer's site team is turned away at the trade counter. The record shows the unallocated receipt, the note and the approval. The controller is accountable for the stop. The fix is twofold: unallocated cash matching an overdue balance should block a stop proposal, and controllers need practice reading the whole draft.
What controls make this fair?
- Block stop proposals when unallocated cash could clear the debt.
- Hold write-offs above a small limit, with the limit owner named.
- Practise before approving alone, using a two-week practicum for credit controllers recorded in a practicum log template for credit controllers.
- Review changes at each renewal. Agent certificates lapse after 90 days; see how to prepare credit controllers for 90-day recertification.
Upstream, the people who sign off the ledger face a related question; see financial controllers and AI agent mistakes. The wider topic is in AI literacy and certification for people who work with agents.
Where the Academy fits
Pryme Intelligence Academy's Operator track trains the people who review a live workspace daily, including human review and when something goes wrong. Its module tests are set in the learner's own industry and role, so a credit controller is tested on ledgers, stops and write-offs rather than generic cases.
Questions people ask
Is the controller accountable for reminders the agent sends alone?
Only through the bounds they set. If a reminder schedule was approved by the controller, a reminder sent under it traces back to that approval.
What if sales overrides a stop the controller approved?
Then sales owns the override. The record should show who lifted the stop and why.
Can the agent agree a payment plan on its own?
It can if the business grants that depth on evidence and sets tight limits. Most teams keep plans held, because they change what the customer owes and when.