Skip to content
allanninal.dev/build
Collaborate

Part 1 of 7 · Credit limit reviewer series ~6 min read

A credit limit reviewer on AWS for a few dollars a month

A customer you have traded with for three years, always paid, never any trouble, goes quiet in October and stops answering the phone in November. The limit was eight thousand. The balance when they went under was nineteen, because the limit was a number in a field that nothing ever compared anything to.

gray and blue Open signage
Photo by Mike Petrucci on Unsplash

Key takeaways

  • A limit is a loss you have agreed to be able to survive, not a compliment.
  • Exposure includes what you have committed, not just what you have invoiced.
  • Your own ledger is a better predictor than any bureau score.
  • Review on behaviour, not on the anniversary of the application.
  • Designed on AWS for about $3 a month.

The whole system on one page

Before any code, here is the shape of what we are designing.

System: credit applications assessed, exposure tracked, limits reviewedThree boxes across the top sit outside the AWS account. On the left, A new account applying for terms. In the middle, Your own ledger of invoices, payments and orders. On the right, Whoever takes the order, on the phone now. Each connects by an arrow to the AWS account container below. The application flows down into the account. Every order and payment feeds in. A number and a reason goes back out. Inside the AWS account are three components in a row. On the left, The opening limit, from the application and filed accounts. In the middle, Live exposure, counting what is committed rather than only what is invoiced. On the right, Review and answer, driven by behaviour and returned in seconds. A note says the middle box is the one most businesses do not have, and it is the one that decides whether the other two matter.AWS ACCOUNTA new accountapplying for termsYour own ledgerinvoices, payments,ordersWhoever takes the orderon the phone, nowThe opening limitapplication andfiled accountsLive exposurecommitted, not justinvoicedReview and answeron behaviour,in secondsapplicationevery order andpaymenta number and areasonThe middle box is the one most businesses do not have, and it is the one that decides whether the other two matter.
Fig 1. Three things outside the account, three pieces inside it. The opening limit is the easy part; keeping it true is the system.
  • Machine learning
  • Analytics
  • Front-end & mobile
  • People

What a limit actually is

A credit limit is not a rating and it is not a statement about how much you like the customer. It is the largest amount you are willing to lose if they fail, expressed as a number you can check an order against before you ship it.

That framing does most of the work. A customer who orders twelve hundred a month and always pays does not need a fifty thousand limit, and giving them one costs nothing until the month they order forty thousand of stock they cannot pay for.

What runs (the inside)

  • The opening limit. Set once, from the application and whatever public record exists, and deliberately small. Part 2.
  • Live exposure. What you are on the hook for right now, including the pallet on the van. Part 3.
  • Review and answer. Revisiting the limit when something changes, and getting the result to the person who needs it. Parts 4 and 5.

One customer, three years

One trade customer from application through to a held limitA horizontal row of five boxes joined by arrows. Applied, with the limit set at three thousand. Paid on time for eleven months. Limit raised to eight thousand, on evidence. Started paying late, at forty-seven days and then sixty-one. Limit held, before the order that would have hurt. A note says the fourth box is the one nobody notices and it is also the only one that was ever a warning.ONE CUSTOMER, THREE YEARSAppliedlimit set at 3,000Paid on time11 monthsLimit raisedto 8,000, on evidenceStarted paying late47 days, then 61Limit heldbefore the orderthat would have hurtThe fourth box is the one nobody notices. It is also the only one that was ever a warning.
Fig 2. The same system as one line. The increase in the middle is earned; the hold at the end is the whole return on building this.
  • Machine learning
  • Security & identity
  • Management
  • Analytics
  • Front-end & mobile

In plain words

A company applies for a trade account. The application and the filed accounts are read once, a small opening limit is set, and the reasons for it are written down where somebody can find them in two years.

From then on, every order and every payment updates one number: what you are exposed to right now. Not what has been invoiced, which lags reality by a week, but what you have committed — accepted orders, picked stock, goods on the van, invoices raised and invoices not raised yet.

When the customer’s payment behaviour moves, the limit is re-examined. Not on the anniversary of the application, which is a date with no meaning, but when the average days-beyond-terms drifts, when an order is unusually large, or when something changes at Companies House.

And when somebody is on the phone taking an order, the answer arrives in the order screen in under a second: how much room is left, and if there is not enough, what the options are. A refusal with an alternative is a sale on different terms; a refusal on its own is a lost customer.

Design rules that shaped every decision

  • The limit is what you can afford to lose, not what they can afford to owe.
  • Exposure counts commitments, not just postings. The van is exposure.
  • Your ledger beats any bureau score after the third invoice.
  • Review on a behaviour change, never on a calendar.
  • The system never refuses anything. It produces a number, and a person decides.
  • Every override is recorded with a name and a reason, and both are visible later.

What it does not do

It does not chase the money. Invoice chasing is a different system with a different rhythm, and merging them produces something that does neither well.

It does not set prices, it does not approve orders, and it does not replace the conversation with a customer who is struggling. It measures, and it makes sure the measurement is in front of somebody at the moment they can act on it.

The next four posts walk through each piece: how the opening limit gets set, why exposure is more than the unpaid invoices, what makes a limit worth revisiting, and how the answer reaches the sales floor. One diagram per post, a cost breakdown, and an engineering reference at the end.

All posts