Credit controllers calibrate on the overdue business accounts the agent held: disputes, broken promises, stop decisions and small write-offs. Each controller decides the same ten accounts alone, the team compares, and the credit manager agrees a gold standard. The standard goes into the agent's brief, and sampling over the next month shows whether the team now treats customers alike.
Why do credit controllers need calibration?
In business-to-business credit control, an agent typically sends reminders on overdue invoices at depth 3 within the company's own rules, drafts statements and chasers, matches remittance advice to open items, and holds anything unusual for a controller. Controllers decide disputes, payment plans, account stops and write-offs.
Those decisions affect sales relationships as much as cash. One controller stops supply to a customer two days after a broken promise; another waits for a call with the account manager. One writes off small balances under 50 without question; another chases every one. Customers notice, sales teams complain, and the agent learns contradictory rules. The general method is in how to run calibration sessions for AI agent reviewers.
The credit manager owns the sessions. A senior controller facilitates. A sales or account management lead attends quarterly.
Which accounts belong in the set?
Eight to ten accounts held in the past fortnight:
- An invoice in dispute over a short delivery, where reminders should pause on the disputed amount only.
- A broken payment promise, the second in three months.
- A remittance advice by email claiming payment was made, with no matching receipt in the bank. The email is external.
- A small residual balance after a customer short-paid by a few units of currency.
- A large customer near its credit limit with an order waiting.
- A customer the agent refused to chase because a dispute flag was set, to test whether controllers read the refusal correctly.
The session, step by step
- Person (credit manager): selects the accounts and removes controller names.
- Person (each controller): decides each account alone: next action, whether to stop the account, whether to write off, with a reason.
- Agent, depth 1: shows its proposed action and the bound that held the account, for example the company's own write-off limit, or a dispute flag.
- Person (facilitator): reveals the spread, account by account.
- Person (controllers): explain their reasoning, especially on stops and disputes.
- Person (credit manager): agrees the gold standard, consulting finance on write-off limits.
- Person (credit manager): sends the standard to the brief owner, including changes to reminder timing and dispute handling.
- Person (credit manager): samples decisions for a month.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Select accounts | Nothing | 3 on reminders | Choosing the set |
| Decide alone | Nothing | 3 | Action, stop, write-off |
| Agent's proposal | Shows action and deciding bound | 1 | Comparing |
| Discussion | Nothing | 3 | Reasons |
| Gold standard | Nothing | 3 | Credit manager |
| Update brief | Chases per new standard | 3 | Brief owner |
| Sampling | Works live accounts | 3 | Credit manager |
What does a gold standard look like for credit control?
It sets the trigger and the action. For example: "A dispute pauses reminders on the disputed invoice only; undisputed invoices on the same account continue. The controller logs the dispute within one working day and the agent's bounds hold any reminder on the disputed amount."
On remittances: "An emailed remittance with no matching receipt is held. The controller does not mark the invoice as paid until the money appears in the bank feed, which is sealed."
On stops: "A second broken promise within three months triggers a stop on new orders, after a call between the controller and the account manager on the same day."
On write-offs: "Residual balances below a set amount are written off by the agent at depth 3 within the company's own limit, and listed monthly for the credit manager."
What does a session look like in practice?
Take a hypothetical team of six controllers at a wholesaler. On the emailed remittance, three marked the invoice as paid because the customer was long-standing, two held it, and one called the customer. The discussion showed that the three who marked it paid had been told by sales that this customer was reliable. The credit manager set the standard that no invoice is marked paid without money in the bank feed, whoever the customer is, and asked sales to route any concerns through the controller rather than around them. The brief owner changed the agent so emailed remittances no longer update the account's status, only add a note for the controller.
Which controls does calibration support, and what evidence does it leave?
Consistent treatment of customers. Evidence: spreads over successive sessions.
Dispute handling. Evidence: the dispute standard and sampled accounts.
Payment evidence. Evidence: the remittance standard, tied to the bank feed.
Agent alignment. Evidence: brief changes and approvers.
What should the credit manager measure?
- Agreement across controllers on shared accounts.
- Reminders sent on disputed invoices. The target is none.
- Invoices marked paid on a remittance with no receipt. The target is none.
- Complaints from sales about stops, before and after calibration.
Where does it go wrong in credit control?
Sales not involved. Stop decisions affect sales directly. If account managers never hear the standard, they fight every stop.
Calibrating only at quarter end. Pressure to collect distorts decisions. Use a quieter week.
Treating the agent's reminders as fixed. If controllers agree a new timing but the brief does not change, every controller overrides the agent manually.
Financial controllers and credit analysts calibrate on neighbouring decisions; see calibration for financial controllers and for credit analysts. For planning, see the AI agent training plan template for credit controllers and how long it takes to train credit controllers to supervise AI agents. The topic guide is training teams for AI agents.
Questions people ask
How often should credit controllers calibrate?
Monthly, and after any change to credit terms, write-off limits or the agent's reminder rules.
Can the agent stop a customer's account?
It can propose a stop. Stopping supply affects a commercial relationship, so most companies keep the decision with a controller.
Should the agent ever accept a payment promise?
It can record a promise made through an authenticated channel and set a follow-up date. Whether to accept a promise after a broken one stays with a controller.