Can an AI agent put a customer account on stop?

Sometimes, under a written credit rule; usually it recommends and a credit controller decides. Putting a trade account on stop blocks new orders, which protects the business and can damage the relationship. An agent can check the rule, confirm the debt is undisputed and prepare the case. Disputed balances and customers in difficulty go to a person.

What does a stop actually do?

A stop flag on a customer account tells the order system to refuse or hold new orders. It does not chase the debt, charge interest or end the contract. It is pressure: a signal that supply depends on payment.

That makes it a commercial decision as much as a credit one. A stop on a major customer can end a relationship worth far more than the overdue amount. A stop on a customer whose payment is sitting unallocated, perhaps in a suspense account, is simply an error made in public. So the decision depends on facts the agent can check, and judgements it cannot.

When could the agent apply a stop itself?

Only when a written credit policy says a stop is automatic and every condition is met:

1. The rule is explicit. For example, balance above the credit limit and oldest debt more than a set number of days overdue.
2. The debt is undisputed. No open query, credit request or complaint on the overdue invoices.
3. No payment is in transit. The agent has checked the bank feed and any unallocated receipts, including the suspense account, for money from this customer.
4. No sign of difficulty. No insolvency notice, payment plan request or other sign that the customer is struggling.
5. The account manager has been told. With time to object before the stop takes effect.
6. The customer is in the segment the rule covers. Many firms exclude their largest accounts from automatic stops entirely.

If any condition fails, the agent prepares a recommendation for the credit controller instead.

What does the agent prepare, and what does a person decide?

Separation of duties sits in the override row. A salesperson who wants an order released for a stopped customer cannot approve it themselves. The credit manager does, and the reason is recorded.

An illustrative example

Take a building supplies merchant with about 2,000 trade accounts. Its policy stops accounts automatically when they are over limit and more than 60 days overdue, except for its top 50 accounts.

On Tuesday the agent finds 14 accounts meeting the rule. For nine, every condition is clear. It notifies each account manager in the morning; none object by the afternoon deadline, so it applies the stops and sends each customer a standard notice. For two, receipts matching the overdue amounts are sitting unallocated; it sends those to cash allocation and applies no stop. Two have disputed invoices, so it recommends no stop. One is a top-50 account, so it prepares the case for the credit manager.

The credit manager calls that customer, agrees a payment date, and decides against a stop.

What controls make it safe?

• The stop is enforced where orders are placed. The order system or an API gateway in front of the agent's tools checks the agent's authority to set the flag, not just its instructions.
• Pre-stop checks every time. Disputes, receipts and difficulty signals are read immediately before acting.
• Notice to the account manager first. With a window to object.
• Lifting is fast. Once payment is allocated, the stop comes off without waiting for a person.
• A record per stop. The rule, checks, notices and any override are filed.

Some actions are worth stopping for in a different sense: bank detail changes are one, as argued in supplier bank detail changes are where finance agents must stop. The wider controls are in AI agents for finance: books you can audit.