Day 4 of 14 · AI in Trade Credit Control

Setting Credit Limits That Fit

⏱ 6 min 📊 Beginner

Yesterday's risk snapshot ended with "approve with a low limit." Today's question: how low? How do you actually pick the number?

In most businesses, honestly, the number is a guess. Sales asks for 20,000, the last customer got 15,000, this one "feels similar," done. The problem with gut-feel limits isn't that they're always wrong — it's that they're inconsistent and unreviewable. Nobody can say why Customer A got double Customer B's limit, so nobody can learn from the ones that go bad.

A credit limit is really a bet: you're wagering that this customer will pay for what you ship before you ship more. Good limit-setting means sizing the bet so your business survives being wrong.

The limit logic: three forces

Every sensible credit limit balances three things:

Their capacity. Can they realistically pay this much within terms? Their net assets, cash position, and size are the ceiling. A limit that's large relative to a customer's whole balance sheet is a fantasy, not a limit.

Your exposure. What does losing this amount do to you? A limit that would sting is a business decision; a limit that would sink you is a mistake, no matter how good the customer looks.

The evidence. How much verified history do you have? A prospect with a glossy pitch has zero payment history with you. Twelve months of clean payments is real evidence — and limits should grow with it.

Which leads to the single most useful rule in limit-setting: start low, grow with history. A modest opening limit costs you almost nothing — if the customer is genuine, they'll trade within it for a few months and you'll raise it on evidence. If they push hard against a cautious starting limit before shipping a single order, that itself is information.

Knowledge Check

A promising new customer pushes back hard on your modest starting limit, demanding triple the amount before their first order. Per the "start low, grow with history" rule, what's the right read?

A

Hold the low starting limit — unusual pressure for large credit before any trading history is itself a risk signal worth noting

B

Grant it — pushing back shows they're a serious, high-volume buyer

C

Split the difference and double the limit as a compromise

D

Decline the account entirely — negotiating a limit is a red flag by itself

Genuine customers usually accept a sensible starting limit because they know it will grow with clean payments. Heavy pressure for large credit from a stranger — before any history exists — is a known pattern in credit fraud and distressed buying. Hold the line, document the request, and let evidence raise the limit.

A simple rubric AI applies consistently

You don't need a data-science team. A five-factor rubric scored 1–5 covers most B2B situations:

Financial strength — net assets, profitability, trend. Payment behavior — days beyond terms with other suppliers, judgments. Trading history with you — none, some, or long and clean. Sector and situation — stable industry or stressed one, concentration risk. Information quality — recent full accounts and references, or stale abbreviated filings.

Add the scores: high scores earn a limit near what their capacity supports; mid scores get a fraction of it; low scores mean minimal credit, or payment up front until history builds. Your own thresholds will depend on your margins and risk appetite — the point is that the same customer facts always produce the same limit.

This is exactly where AI shines. Humans apply rubrics inconsistently — Friday-afternoon generosity is real. AI applies the same rubric the same way every time, shows its scoring line by line, and drafts the file note. Then a human — you, or whoever holds credit authority — approves, adjusts, or overrides the number. An override is fine; an undocumented override is how gut-feel sneaks back in.

The right limit sits where their capacity, your appetite, and the evidence overlap — and it's a dial you keep adjusting, not a number you set once.

🤖

ChatGPT

online

Score these two anonymized customers on my rubric (1–5 each: financials, payment behavior, history with us, sector, info quality) and propose limits with reasoning. Requested limit is 20,000 each. Customer P: net assets 450,000 and stable, pays 5 days beyond terms, 3 years with us with zero late payments, stable sector, full recent accounts. Customer Q: net assets 40,000 and falling, pays 25 days beyond terms, brand new to us, stressed sector, abbreviated accounts filed late.

**Customer P — score 23/25** Financials 5 · Payment 4 · History 5 · Sector 5 · Info 4 Strong on every factor, and three years of clean history with you is the best evidence there is. The 20,000 request is comfortably within their capacity. **Proposed limit: 20,000 approved.** Routine annual review. **Customer Q — score 9/25** Financials 2 · Payment 2 · History 1 · Sector 2 · Info 2 Thin, declining balance sheet; slow payer elsewhere; no history with you; stressed sector; weak information. A 20,000 limit here is half their net assets — that's not a limit, it's a hope. **Proposed limit: 4,000 starting limit**, 30-day terms, review after 3 months of clean payments; consider proforma for anything above it. **Review triggers for both:** any payment 15+ days late, a sudden order spike above normal pattern, or adverse filings/news. Same rubric, same math — the gap between 23 and 9 is why they shouldn't get the same number. Final approval is your call.

↻ Replay conversation

Knowledge Check

Both customers asked for 20,000. What justified proposing 4,000 for Customer Q instead?

A

New customers must always wait a year before receiving any credit

B

The AI is programmed to cut every requested limit by eighty percent

C

Q scored low on every rubric factor, and 20,000 would equal half their net assets — an exposure their capacity can't support

D

Customer Q's sector makes trade credit unavailable by regulation

The rubric surfaced weakness on all five factors — thin falling financials, slow payment elsewhere, no history, sector stress, poor information. Sizing the limit against their capacity (20,000 vs 40,000 net assets) made the mismatch obvious. Q still gets credit — just a bet-sized amount that grows with evidence.

Limits are living numbers

A limit set once and forgotten quietly rots. Build in review triggers — events that force a fresh look:

Late payments — an on-time payer drifting to 15, then 25 days beyond terms is telling you something. Order spikes — a sudden order far above their normal pattern can be great news or a customer loading up before trouble; look before you ship. Adverse news — new judgments, late filings, lost contracts, sector shocks. On Day 11 you'll turn these into a full early-warning system; for now, just make sure every limit has a review date and a trigger list attached.

Notice what you've built across two days: check the customer (Day 3), score and size the limit (today) — a repeatable onboarding decision that takes minutes and stands up to scrutiny. Tomorrow, the final piece before the first invoice: the paperwork — applications, terms, retention of title, and guarantees.

🎯

Day 4 Complete

"A credit limit is a bet sized to survive being wrong."

Tomorrow — Day 5

Credit Applications, Terms & Guarantees

Tomorrow you'll get the paperwork right before the first invoice — payment terms, retention of title, and guarantees in plain English.

Continue to Day 5 →