Before offering 30-day terms, get a signed credit application, confirm the business exists and that you are really dealing with it, pull a business credit report, and take two trade references. Then set a starting limit no bigger than you could afford to lose, often one or two typical orders, and raise it only after the customer has paid on time.
Most small-firm bad debts come from skipping these steps for a promising new customer with a big first order, or from limits nobody revisits. AI is useful for summarising credit reports, drafting reference requests and spotting mismatches in an application; the decision stays with a person. If the customer is a sole trader, you are assessing an individual, which brings consent and data-protection duties.
A $15,200 first order from a customer you've never met
The illustrative business is a subscription box company with a corporate gifting side: it makes branded gift boxes for businesses to send to clients and staff. In October an events agency it has never dealt with emails asking for 400 boxes at $38 each, $15,200 in total, delivered in early December on 30-day terms. That is more than the gifting side's whole November turnover. The order is attractive, the deadline is tight, and the temptation is to say yes and sort out the paperwork later. The rest of this page is what "sort it out first" looks like, and it takes about two hours.
Write a one-page credit policy before the next application
Decisions made on the spot, under a deadline, are the risky ones. A short policy written in advance makes the answer mostly mechanical. The gifting side's policy, drafted with AI from the owner's notes and then edited:
- Who gets terms: registered companies and established organisations only. Sole traders and new companies (under two years old) pay up front or a 50% deposit.
- Standard terms: 30 days from invoice, under our terms of business.
- Checks: every new account needs a signed application, identity checks and a credit report. Orders over $5,000 also need two trade references.
- First limit: the lowest of the agency's suggested limit, two typical orders, and $6,000.
- Approval: up to $6,000, the sales lead; above that, the owner.
- Late payment: no new orders on account while any invoice is more than 14 days overdue.
- Review: each account after its first three invoices, then yearly.
The terms themselves belong in your terms of business, which the application must refer to; writing terms of business with AI covers drafting them and what a lawyer should check.
A credit application form that feeds every later check
A one-page form, completed and signed by someone with authority to commit the business, collects what every later check needs:
- Full legal name, trading name, registration number where one exists, registered and trading addresses
- How long trading; names of directors, partners or owners
- Accounts payable contact, invoice email address and any purchase order requirements
- Bank name and account holder name (for reference and for matching payments later)
- Two trade references with contact details
- Credit limit requested and expected monthly spend
- Consent to credit checks, including on the individuals named if the business is not a company
- Signed acceptance of your terms of business
Some suppliers also ask directors of small or new companies for a personal guarantee. That is a legal document with real consequences for the person signing, so have it drafted by a lawyer rather than an AI, and use it sparingly.
Check they are who they say they are
Before assessing whether a customer can pay, check the customer exists and that the person emailing you represents it. A common fraud pattern targets exactly this kind of order: someone poses as a buyer from a real, creditworthy company, uses an email domain one character different from the real one, asks for goods on credit, and has them delivered to an address that is not the company's. The goods vanish and the real company has never heard of the order.
The checks take 20 minutes:
- Look the company up on the official company register where it is incorporated. Do the name, number, registered address and directors match the application?
- Check the email domain against the company's real website, found by your own search rather than a link in the email.
- Ring the company on a number from its website or the register, not the email signature, and ask for the person by name.
- Check the delivery address is the company's own premises or a venue that makes sense for the order.
AI speeds up the comparison. Paste the application details and the public register entry into a chat and ask it to list every difference. Illustrative output for the events agency:
Differences between application and public register entry:
1. Registered address matches. Trading address on application is different
(a business centre in the same town) - common for small agencies; confirm.
2. Application lists 2 directors; register lists the same 2 plus a third appointed
last month.
3. Company incorporated 3 years ago; application says "trading since 2019" -
possibly a previous business; ask.
4. Email domain on application matches the company website.None of these are red flags on their own, and the call to the agency's main number confirmed the buyer. Point 3 turned out to be a sole trader business that incorporated three years ago, which is normal, but worth asking about because the company's own credit history only starts at incorporation. Staff who handle orders should know the fake-buyer pattern; training staff to spot AI-written phishing emails covers the lookalike-domain tricks these frauds use.
Reading a business credit report with AI
A business credit report from a credit reference agency typically includes a score or rating, a suggested credit limit, filed accounts where available, payment behaviour (how many days beyond terms the business typically pays other suppliers), court judgments or insolvency notices, director appointments and other companies the directors run. What is available varies by agency and by country. Reports are dense, and AI is good at turning one into a decision sheet. Remove nothing but also add nothing personal beyond what the report holds, and use a business plan or switch off the model-training setting before uploading:
Summarise this business credit report for a credit decision on 30-day terms.
Return:
- Score and what the agency says it means (quote)
- Agency's suggested limit (quote it; this is NOT our approval)
- Accounts: latest filed, filing on time or late, profit/loss trend, net assets
- Payment behaviour: average days beyond terms, trend over 12 months
- Any judgments, insolvency events, or recent changes of directors
- Red flags and amber flags, each with the evidence line from the report
Do not recommend approve or decline. Say "not in report" for anything missing.Illustrative output for the events agency:
Score: 68/100, described by the agency as "low to moderate risk".
Suggested limit: $9,000 (agency figure, not an approval).
Accounts: latest filed 7 months ago; the previous year's accounts were filed late.
Small profit in both years; net assets positive but modest.
Payment behaviour: pays on average 11 days beyond terms; was 6 days a year ago.
Judgments/insolvency: none recorded.
Changes: a third director appointed last month.
Amber flags: late filing last year; payment days worsening.
Red flags: none.The first version of this prompt did not include "this is NOT our approval", and the AI's summary ended "Approved limit: $9,000", which a busy salesperson could easily take at face value. It also listed the late filing without comment; late accounts are often an early sign of a stretched finance function. Both points are why the prompt now forbids a recommendation. The worsening payment days matter too: a customer who pays 11 days late on average will probably pay your 30-day invoice in about 40 days, which affects your cash flow even if it never becomes a bad debt.
Trade references that tell you something
References are only useful if you ask specific questions. The gifting side's reference request, drafted with AI and sent by email:
"Hello, [customer] has given your business as a trade reference for a credit account with us. Could you tell us: how long they have had an account with you; their usual credit limit or typical monthly spend; your payment terms and whether they usually pay within them; and whether you have ever had to put their account on hold? A two-line reply is plenty. Thank you."
Watch for weak references: suppliers so small the account is trivial, businesses that share a director with the customer, and replies that arrive within minutes from a free email address. One of the agency's references was a caterer with a $1,500-a-month account, paid on time, which is genuine but not much evidence for a $15,200 order. The other was a venue that had given them 30-day terms for two years with one late payment. Useful, and consistent with the credit report.
Setting the first limit
The policy's rule is the lowest of three figures. For the events agency:
- Agency's suggested limit: $9,000
- Two typical orders: unknown for a new customer, so the owner used the expected spend after this order of about $3,000 a quarter, giving $6,000
- Policy cap for new accounts: $6,000
So the limit is $6,000, and the order is $15,200. That does not mean refusing the order. It means structuring it so the exposure fits the limit. The owner offered two options: a 50% deposit ($7,600) with the balance on 30 days, leaving $7,600 on credit, which is slightly above the limit and so needed her sign-off as an exception; or two deliveries of 200 boxes, the second released once the first invoice was paid. The agency took the deposit option. The quick sum behind the decision: at a 40% gross margin, a total loss on the $7,600 balance would cost about $4,560 in goods and labour, which the business could survive. A total loss on $15,200 without a deposit would have cost about $9,120, which it could not comfortably survive in December. Run the same sum for your largest accounts from time to time; the total you could lose if your two biggest customers failed together is the number that matters most.
A scoring sheet that keeps decisions consistent
A simple score stops the decision depending on who is in the office. Each factor scores 0, 1 or 2:
| Factor | 0 | 1 | 2 | Events agency |
|---|---|---|---|---|
| Identity checks | Mismatch unresolved | Minor differences explained | All match | 1 |
| Time trading as this entity | Under 1 year | 1 to 3 years | Over 3 years | 1 |
| Credit report | Red flags | Amber flags only | Clean | 1 |
| Payment behaviour | Over 30 days late | 1 to 30 days late | On time | 1 |
| References | None or poor | One useful | Two useful | 1 |
| Order size vs limit | Over 3x limit | 1x to 3x limit | Within limit | 1 |
The policy's bands: 10 to 12, standard limit; 6 to 9, standard limit with a deposit or staged deliveries on large orders; 5 or below, payment up front. The agency scored 6, which matches the decision the owner reached. Keep the score sheet with the application; when an account goes bad, it shows whether the policy failed or someone bypassed it. Be careful that factors do not quietly discriminate, for example by treating customers from certain areas as higher risk; AI bias in small business decisions explains how that creeps in.
Write the decision down in five lines
Whatever you decide, record it on the application in a short, standard note. It takes two minutes, and it is what lets someone else, or you in six months, understand why the account looks the way it does. The note for the events agency:
"Decision, 14 Oct: account opened, limit $6,000, 30 days. Score 6 of 12 (amber: late filing last year, payment days worsening). First order $15,200 accepted as an exception with a 50% deposit ($7,600) received before production; balance of $7,600 on credit, approved by the owner. Review after third invoice or by 31 March."
An AI assistant can draft the note from the score sheet and the report summary, but a person should write the reason for any exception in their own words. Exceptions are where the losses hide.
After approval: watch the first three invoices
- Enter the credit limit in your accounting software and turn on its warning or hold for orders that exceed the limit or arrive while an invoice is overdue.
- Send the invoice the day the goods ship, to the accounts contact on the application, quoting any purchase order number.
- Diary a friendly reminder a few days before the due date for the first invoice. AI payment reminders that sound human covers the wording.
- After three invoices, review the account: average days to pay, any disputes, and whether the limit should rise. Payment patterns are the best predictor you will get; predicting late payers with AI shows how to spot the slide early.
- If your credit agency offers monitoring, add the customer so you hear about new judgments, late filings or director changes.
Sole traders, partnerships and individuals
When the customer is not a company, checking the business often means checking a person. That usually needs the person's clear consent, recorded on the application, and brings duties under data-protection law about what you collect, how long you keep it and who sees it. Personal credit files are more sensitive than company reports, so restrict access to the one or two people who make credit decisions.
Automation needs extra care here. If you sell to customers in the EU, AI systems used to evaluate the creditworthiness of individuals are listed as high-risk in Annex III of the EU AI Act, with the main obligations for such stand-alone systems deferred to 2 December 2027. Using AI to summarise a report for a person who then decides is very different from letting a system score and decline individuals automatically. Keep the decision human and documented, and get advice before automating any refusal.
Football clubs and nursery groups need different checks
- A sports equipment shop supplying kit to a junior football club faces a governance risk rather than a credit one. Clubs are often run by volunteer committees, and the treasurer who ordered the kit may have left by the time the invoice is due. The shop asks for the club's formal name and bank account, and an order form signed by two committee officers, and invoices the club, not the individual.
- A toy shop supplying a group of nurseries finds that they pay reliably but slowly and reject any invoice without a purchase order number. Its check is lighter on credit and heavier on process: it will not ship without a purchase order, and it invoices exactly to the purchase order's details. If your quoting and invoicing are joined up, connecting quotes, invoices and payments makes that automatic.
For the events agency, the whole check took about two hours: 15 minutes on the application, 20 on identity, 20 on the credit report with the AI summary, a day's wait for references, and 30 minutes on the decision and the offer. The agency paid its balance 38 days after the invoice, in line with its record, and the account now has a $6,000 limit and a history.
Credit check questions from small suppliers
Will asking for references put a new customer off?
Rarely, if you present it as routine. Established businesses expect a credit application and are used to supplying references. Say it is your standard process for all new trade accounts, and offer to start with a deposit or a smaller first order while the checks run. A customer who is offended by a normal credit check is telling you something useful.
What if the customer won't pay a deposit or accept a lower limit?
Then the choice is yours: decline, ask for payment up front, split the order into smaller deliveries that each fit the limit, or accept the risk knowingly with a senior person's sign-off. Write down why you made an exception. Most bad debts in small firms trace back to an exception that nobody recorded or revisited.
How much does a business credit report cost?
It varies by agency and country, and between pay-per-report and subscription plans. Small suppliers checking a few customers a month often pay per report, while those checking many choose a subscription with ongoing monitoring alerts. Compare the agencies that cover your customers' markets and check whether monitoring of existing customers is included in the price.
Further reads
- How to Automate Invoicing With AI: From Finished Job to Paid — Invoice promptly so the terms you agreed start on time.
- How to Catch Duplicate Invoices and Payment Fraud With AI — Spot payment fraud on the supplier side too.
- Automate Client Onboarding With AI: Forms, Contracts, and Emails — Build the credit application into customer onboarding.
- How to Keep Customer Data Private When Your Team Uses AI — Handle customer financial details safely when using AI.
- How to Build a 13-Week Cash Flow Forecast With AI Help — See what slow-paying customers do to your cash in the weeks ahead.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: general trade credit practice; the EU AI Act, Annex III and its deferred high-risk timetable. Credit report prices and contents vary by agency and market, so none are quoted.