AI in credit checks on new customers before you extend terms

5 days ago
3 min read

Introduction

Giving a new customer thirty days to pay is lending them money. It is an unsecured loan, granted without an application, usually on the basis that they seemed reasonable on the telephone and the order was welcome. The same business that would think carefully before lending a thousand pounds will extend five thousand in credit to a company it knows nothing about.

Bad debt is also asymmetric in a way that is easy to underestimate. At a ten per cent net margin, one unpaid invoice consumes the profit from ten similar jobs. The purpose of checking is not to refuse business; it is to set the exposure at a level where a loss would not undo a quarter's work.

1. AI in credit checks on new customers is about sizing exposure

Not about approval and refusal.

The output should be a credit limit and terms, not a yes or no. Almost any customer can be traded with safely on the right terms, and the question is what those terms are.

2. Set a threshold below which you do not check

Proportionate effort.

Small first orders do not justify the cost of a check. Above a stated value, a check is mandatory. This rule alone puts the discipline in place without adding friction to every sale.

3. Use payment behaviour data, not just accounts

The more predictive signal.

Filed accounts are historical and often eighteen months old. Data on how a business actually pays its suppliers, where it is available, is a far better predictor of whether you will be paid.

4. Look at the pattern of enquiries and changes

Warning signs are visible.

Recent changes of directors or registered address, a sudden increase in credit searches, charges registered, or a newly incorporated entity replacing an older one. Each of these is worth a question.

5. Take a deposit rather than declining

The commercially sensible middle.

Payment in advance for the first order, a deposit, or a lower limit that grows with payment history. This wins business a refusal would lose and caps the exposure while you learn.

6. Confirm who you are actually contracting with

A frequent and expensive error.

The trading name, the legal entity, the correct registration, and whether the person ordering can commit that entity. Invoicing the wrong entity makes recovery difficult regardless of the credit position.

7. Review limits as the relationship grows

Both directions.

A customer whose orders have quadrupled is now a much larger exposure than the limit contemplated. A customer who has paid promptly for two years deserves better terms. Neither happens without a review.

8. Monitor existing customers, not just new ones

Where most losses occur.

Most bad debts are long-standing customers whose position deteriorated. Alerts on adverse changes across your existing customer base are more valuable than checks on new ones, and almost nobody has them.

9. Act on the first missed payment

The discipline that limits the loss.

A customer in difficulty pays whoever pursues them. Continuing to supply while an invoice is overdue increases the exposure at exactly the wrong moment, and a supply stop is a commercial decision to be taken early.

Credit reference data, its use, retention and the way decisions about individuals are made are all subject to rules that vary by jurisdiction, particularly where the customer is a sole trader or an individual rather than a company. Confirm the position that applies to you.

Conclusion

Treat this as sizing an exposure rather than approving a customer.

Set a value threshold above which a check is mandatory, use actual payment behaviour data rather than filed accounts alone, look for recent director, address and charge changes as warning signs, offer a deposit or a small growing limit instead of refusing, confirm the exact legal entity you are contracting with, review limits as order volumes change, monitor your existing customers for adverse changes because that is where most losses arise, and stop supply promptly when an invoice goes overdue.