Credit limits: the number nobody enforces until it is too late

A credit limit exists only at the moment an order is accepted, and only if something refuses the order. Written in a spreadsheet, reviewed at month end, or held in the head of whoever answers the phone, it is a note about a number. It becomes a control when it is computed from what the customer owes, checked before the goods are released, and overridable only by a named person whose reason is recorded.

Where does the limit live now?

Usually in three places, none of them where orders are taken: a column in a spreadsheet one person updates, a free-text field on the customer card that nobody fills, and the owner's memory of who is safe to sell to.

The test costs nothing: accept an order past the limit and watch what happens. If the answer is "the salesperson would ask me first", the control is somebody's attention — and attention is not available late on the day before a holiday, which is when the largest orders arrive.

Who sets the limit, and against what?

Not against turnover. Turnover says how much a customer buys, not how they pay, and the slowest payer is often the biggest buyer, because nobody wants to stop a large account.

Set it against the ageing: average days late, the oldest item open, whether they crossed a promised date and by how much. A customer who pays in full and always late is a different risk from one who pays quickly and leaves a residue that never clears.

Then against exposure — invoiced and unpaid, ordered and uninvoiced, plus this order — and against margin, since a thin-margin customer who pays slowly is a loan at your own cost.

The limit is a decision about appetite, not an accounting fact, so it belongs to one role, with a date, a reason and a review date. A limit never revisited is either too generous forever or blocks a good customer forever.

What has to block, and when?

At order entry, on total exposure. Not at invoicing, not at delivery, not at month end.

The order of events decides whether the control is worth anything. Check at invoicing and the goods are already picked, loaded and often delivered: the choices left are to hold an invoice for goods the customer has, or to send it and hope. At month end it is a report on a decision paid for in stock.

Order entry is cheap because nothing has moved, and the only stage where the order-taker can still offer an alternative — a smaller quantity, a part payment, a cheque.

One condition: exposure must include orders confirmed but not yet invoiced, or a customer places four orders under the limit and the fifth is the one that hurts.

Per customer, or per group?

Both, in most wholesalers, because one person often trades under more than one name.

Scope	What it measures	When it applies
Customer	One account's open balance	One legal entity pays
Group	Every linked account together	One payer, several names
Guarantee	Exposure against a cheque or note	Payment is secured

A limit per customer is really a limit per name, and adding a name is trivial. Three shops registered separately, one person signing the cheques: three comfortable limits are one uncomfortable exposure. The grouping has to exist as data — an identifier linking the accounts — or it lives in the head of whoever knows the family.

Guarantees pull the other way: a cheque can justify a higher limit, and the system should record which limits are secured.

What happens when a payment arrives late?

It must free capacity the day it is recorded, and it must be recorded the day it arrives. Most businesses fail at both.

The first is mechanical: a customer pays, the receipt is entered three days later, and in between the counter refuses an order from a customer who is square. The response is predictable — raise the limit, and never bring it back down.

The second is informational: days late must be stored, not remembered. A review that begins with "they usually pay, I think" is a review of the reviewer's mood. Promised against paid date, per invoice, is the only evidence that survives a change of staff.

Why must a blocked order stay overridable?

Because a block with no exit gets disabled, and a disabled block protects nothing. Its purpose is to force a decision at the right moment, not to prevent the sale.

So the override is designed: the people who may use it are named in advance, the reason comes from a short list rather than free text, and the override shows on the order and the customer's account. A weekly look at that list is the cheapest credit review available: it shows which limits are wrong.

An invisible override is a limit with a hole, and the counter finds it within a month.

One case needs a written rule rather than a screen: the hours when the system is down. A fallback that takes the order on paper unchecked is a decision about risk, and the contingency planning NIST publishes for information systems treats it as a documented, tested procedure, not something improvised.

How do you put a limit into force?

1. Write the exposure definition: open invoices, confirmed orders, this order.
2. Compute payment behaviour from the ageing — days late and oldest open item.
3. Set the limits with a named role, a date, a reason and a review date.
4. Decide the scope per account: customer, group or guarantee.
5. Put the check on the order screen, and make it block.
6. Record payments the day they arrive, and release capacity at once.
7. Name the people who may override, with a short list of reasons.
8. Read the override list weekly; re-set limits monthly against the ageing.

Where this goes wrong

The quarterly spreadsheet. Correct the day the file is sent, stale a week later, unopened before any order is accepted.

The ratchet. A limit is raised to unblock one order and never lowered, so every limit drifts upward until the first default reveals it.

Three names, one payer. Group exposure ignored, three separate limits, and a total nobody is measuring.

The payment nobody entered. A customer blocked while paid, an angry call, and a limit raised to end the argument.

What this does not cover

This is not legal advice on collection, cheque law or insolvency, and it does not say what a limit should be — that is an appetite decision, and any figure would be wrong for the next business. It assumes orders are taken in a system that can refuse them; where much of the trade arrives by phone on paper, the control starts when the paper is entered, which is later and weaker.

We build the limit into the order screen, computed from the ageing, with an override that leaves a name and a reason behind it.