One line, the wrong guess

A payment to a person looks like wages. It was a contractor with an invoice on file. A rule that only reads the text gets this wrong every week. A person answers it once.

Borrowed money is not a sale

A draw on the credit line arrives in the bank like a customer payment. It is kept in its own lines, below operations, so a week that was rescued by borrowing never reads as a good week.

Own account to own account

The transfer that funds payroll leaves one account and lands in another. Both sides are excluded, or the same cash is counted twice.

It spreads totals. Cash moves one invoice at a time

The usual model takes receivables in total and spreads them by payment terms. Here 162 open invoices, bills, purchase orders and fixed payments are each placed on the week their cash moves, with each customer's own habit of paying late. That is 200 lines behind the 13 weeks, rebuilt every week.

It loses what does not fit a column

Formulas that match a due date to a week quietly drop anything already past due or beyond week 13. Past due invoices are the most collectable cash in the business. Here every open item is in a week, beyond week 13, or excluded with a reason, and the page proves it adds back to the aging: $1,525,830 of receivables, $20,900 excluded, nothing unplaced.

It forgets last week

The file is saved over, so last week's forecast is gone and nobody learns whether it was right. Here each week is locked when it closes and the next one is measured against it: last week was 90.1% accurate, with a reason for every difference.

It does not know what is on the shelf

Purchases were planned at 75% of sales, whatever the stock. The month-end count says 9.6 weeks on hand against a target of 8, with another $600,800 already on order. Until the forecast reads that, the largest payment the company chooses to make is the one line nobody is managing.

What AI does here

Reads the words the bank prints and suggests a category for a line no rule has seen, with the reason.

Learns each customer's habit from their own paid invoices, so a collection date is their history, not their terms.

Notices what a formula would not. New sales were forecast as half after 6 weeks and half after 7. The recent invoices showed 30% after 5 weeks, 30% after 6 weeks, 15% after 7 weeks, with slower customers after that. Left alone, the forecast showed a shortfall in week 7 that was never going to happen.

Asks the 6 questions that need a person, one at a time, and writes each answer down where next week will find it.

What AI does not do

The arithmetic. Fixed code places every line and adds every week, the same way each run. A number cannot change because it was asked for differently.

The checks. 15 of them run before anyone sees the page. This week 11 passed and 4 were flagged, each shown with the decision a person made.

The decisions. A new category, a difference, a changed assumption, which action is real, and anything that is sent. Those stay with the person who signs.

Left alone

Week 13 closes $29,470 under the $250,000 minimum. The usual answer is to draw on the credit line and pay interest on it.

The 2 obvious actions

Splitting an import payment and a payment plan for an overdue customer help the middle weeks. Week 13 is still $8,570 short.

Buying to the stock target

Week 13, $264,900 higher than with the actions alone, and no week under the minimum. $292,839 was sitting in stock above the 8-week target. Purchases still to be decided go from about $189,700 a week to $159,700, and stock ends week 13 on the target. No new borrowing.