You know roughly what is coming in and roughly what is going out, and on most Fridays that is enough. Then a key account pays eleven days late, the fuel card direct debit and payroll land in the same week, and you are moving money around at nine at night to make sure eight drivers get paid.
A 13-week cash flow forecast fixes that. Not the version your accountant builds once a year for the bank, but a working sheet you own, that shows you the tight weeks while you still have time to do something about them. Below is how to build one with ChatGPT using paperwork you already have, and how to keep it current in about fifteen minutes a week.
Why 13 weeks, not 12 months
Thirteen weeks is one quarter. It is long enough to see the problems coming and short enough that the numbers are still real.
- Your debtors are inside the window. Net-30 invoices raised today land in week four or five. Net-45 accounts land in week seven. Everything on your aged debtor report resolves inside 13 weeks, so you are forecasting actual money, not guesses about future sales.
- Your fixed costs are known. Truck finance, insurance, payroll, rent and the fuel card cycle are all contractual. You know the dates and the amounts.
- You can still act. Thirteen weeks gives you time to chase two accounts, delay a trailer purchase, or talk to a factoring company. A twelve-month forecast tells you nothing you can change on Monday.
- Weekly beats monthly. A monthly view shows a comfortable month. A weekly view shows the Thursday in week six when the balance goes to $4,000 because payroll and the insurance renewal collided. The monthly view hides exactly the thing you need to see.
The inputs you already have
You do not need new software. You need six things, and five of them are already on your computer.
- Your aged debtor report. Export it from Sage or QuickBooks. Customer, invoice number, amount, invoice date, terms.
- The realistic pay date for each invoice, not the due date. This is the part most forecasts get wrong. If a customer has paid you at 44 days on average for the last year, their net-30 invoice is not due money, it is week-seven money. Go through the ledger by account and write down what each one actually does.
- The fuel card cycle. Weekly or fortnightly direct debit, with the amount driven by miles run. Pull the last six statements and note the dates and the range.
- Payroll dates and amounts. Drivers, yard, office, and the employer taxes that follow. Include the weeks where a fifth payroll run falls if you pay weekly.
- Truck finance and insurance. Every HP or lease agreement with its debit date, and the fleet insurance renewal or monthly installment. Add the trailer, the forklift, and anything else on a finance agreement.
- The big supplier runs. Tires, parts, the maintenance contract, the subcontract carriers you use on overflow work. These are your controllable payments and they are the ones you will move when a week gets tight.
Put all of that in one place as plain text. Then paste it into ChatGPT and ask it to build the grid. It will do in two minutes what takes an hour in Excel, and the cash flow forecaster skill keeps it in a consistent shape so week two looks like week one.
The prompt
I run an eight-truck haulage firm and I need a 13-week cash flow forecast, weekly columns starting Monday 21 September. Opening bank balance is $41,200. Below is my aged debtor list, my fixed outgoings with their dates, and the average days-to-pay for each customer account. Use the average days to pay to work out when each invoice actually lands, not the invoice terms. Show money in, money out, net movement and closing balance for each week. Flag any week where the closing balance drops below $15,000 and tell me which two customer accounts would fix it if they paid on terms. Mark anything you have had to assume. Here is the data:
Then paste the lists. Do not tidy them first. Columns out of an accounting export are fine.
A worked example
The numbers below are an illustration, not a real company.
Northline Freight is an eight-truck regional haulier. Monthly payroll is $34,000 including employer taxes, paid on the 28th. Fuel card debits every second Wednesday at $9,000 to $11,500. Truck finance is $7,400 a month across five agreements, split across the 5th and the 20th. Fleet insurance is $3,100 monthly. The maintenance contract is $1,850 a month. Opening balance is $41,200.
On the debtor side there is $186,000 out, and it is not evenly spread:
- Castlereagh Retail Group, $78,000 across nine invoices, net 30 on paper. Average days to pay over the last year: 52. This is the key account and it is the slow one.
- Merrow Plastics, $31,000, net 30, pays at 29 days like clockwork.
- Hallam Building Supplies, $24,000, net 45, pays at 47.
- Eleven smaller accounts, $53,000 in total, mostly net 30 and mostly on time.
When ChatGPT lays that out week by week, the picture that comes back is not the one the monthly view showed. Weeks one to four are comfortable. Week six is the problem: a fuel debit on the Wednesday, the $7,400 finance run, and the 28th payroll all inside five days, against a week where the only meaningful receipt is a Castlereagh invoice that is technically due but, at 52 days, will not arrive until week eight. Closing balance week six: $6,300. Week seven recovers to $19,000. Week eight is fine.
That is a one-week hole, and there are four ways to fill it.
Chase the right two accounts. Not everyone. Two. Castlereagh has $78,000 out, and $22,000 of that is over 60 days even by their own slow standard. Getting one invoice of $14,000 moved forward by ten days closes the week-six gap on its own. Ask ChatGPT which specific invoices, if paid on terms, would lift the tight week above your floor, then write the chase. The overdue invoice chaser skill drafts the sequence: the polite one now, the firmer one in seven days, and the point where you stop emailing and pick up the phone to their finance manager.
Move a controllable payment. The maintenance contract invoice and the tire order are yours to time. Paying the tire supplier in week seven instead of week six is a phone call, not a crisis. Decide this in week two, not on the Thursday.
Time the purchase. If you were going to put a deposit down on a replacement trailer in week six, put it in week nine instead. Same purchase, different week, no hole.
Know your factoring option before you need it. If Castlereagh at 52 days is structural rather than a bad quarter, invoice finance on that one account may be worth pricing. Get the real cost as a percentage of the invoice value and check with your accountant how it sits against your other facilities before you sign anything.
There is a fifth thing worth doing over a quarter rather than a week. If a key account is slow and also sends you a quarterly scorecard on your on-time delivery, the conversation about their payment behavior belongs in the same relationship as the conversation about your service. The supplier scorecard response skill helps you answer their numbers with evidence, which is the right footing for asking them to pay at 30 rather than 52.
What to check before you rely on it
ChatGPT will produce a clean grid from messy input. That is useful and it is also the risk, because a wrong number looks exactly as tidy as a right one. Before you act on the forecast:
- Check the opening balance against your bank, today. Not last week's statement.
- Check every assumption it marked. If it guessed a fuel debit date or an average days-to-pay, correct it. Ask it to list assumptions separately so you can see them.
- Check the payroll weeks. If you pay weekly, confirm the forecast has the right number of runs. A missed fifth run is a $8,500 error in the wrong direction.
- Re-add the debtor column yourself. Total the aged debtor figures and compare to the forecast's money-in total. They should match within rounding.
- Check nothing is counted twice. Invoices raised this week that are also in the export are the usual culprit.
- Keep VAT, sales tax and any tax payment dates where your accountant put them. Confirm the amounts and dates with them rather than estimating.
Updating it in fifteen minutes
The build takes an hour. The update does not, and the update is the whole point.
Every Monday, do three things. Pull the new bank balance. Pull the new aged debtor export. Paste both back in and ask for the forecast to be rolled forward one week, with a note on what changed and whether any week has moved above or below your floor.
That is it. What you get is a running record of which accounts are drifting, which weeks are tightening, and whether the thing you decided last Monday actually worked. After six or seven weeks the average days-to-pay figures stop being estimates and start being your own data, and the forecast gets sharper without you doing anything extra.
The value is not the spreadsheet. It is knowing on the 21st that week six is tight, instead of finding out on the Thursday.