Money
The Thirteen-Week Cashflow
The one most operators say they have and most do not actually update.
Fix
1
·
Money
·
An hour to build, then twenty minutes a week
·
Retest quarterly
THE SYMPTOM
If your biggest customer paid four weeks late tomorrow, could you tell me within two minutes whether payroll still clears?
There is a file somewhere called cashflow. It was built for the bank, or a board pack, or in a bad week, and it has not been touched since. When a large invoice slips, you feel it as a knot in your stomach rather than seeing it as a number in a column. Cash conversations happen at the moment of the problem, and they are usually started by someone else: the bank, a supplier, your bookkeeper.
If two or more of those are true, this gap is open.
THE CAUSE
Profit is not cash, and in many growing businesses the mismatch is structural. You do the work weeks before you invoice. You invoice weeks before you are paid. Your outgoings do not wait for either. A handful of customers dominate your receipts, so one slipped payment moves the whole picture, and VAT and PAYE land in lumps on fixed dates that have no interest in your pipeline.
That explains why cash is hard. It does not explain why the file dies. The file dies because it was built as a document rather than a routine. A forecast made once, to answer a question, starts decaying the day after it is made, and nobody goes back to a decayed file. Monthly granularity finishes the job: a month that looks fine in total can contain a week that does not clear payroll, and a monthly view will never show you which week.
So the fix is not a better forecast. It is a weekly operating routine with a file attached.
THE MOVE
In order, with the decisions already made.
1. Direct method, weekly, thirteen weeks. Money in and money out by week, from the bank's point of view. Not a P&L projection, not monthly, not a year. Thirteen weeks is one quarter: far enough out to act, near enough to be honest.
2. Start from the bank. Your opening balance is today's cleared balance across every business account, as one number. Not what the ledger says. What the bank says.
3. Enter receipts by name for the first six weeks. List the actual invoices: customer, reference, amount, and the week you expect the money. Expected, not due. If a customer pays thirty days late every time, forecast them thirty days late. Weeks seven to thirteen can be patterned on typical activity.
The due date is a hope. The expected date is a fact about their behaviour, and you already know it.
4. Enter payments from standing reality. Payroll. PAYE and NI on the 22nd. The VAT quarter. Rent. The suppliers who chase. Loan and finance payments. Then everything else as one honest line. Do not itemise the small stuff; the lumps are what sink you.
5. Set your floor. The minimum balance you are willing to see in any week. If you do not have a number, use one month's payroll. The spreadsheet flags any week that goes below it.
6. Run the Monday ritual. Twenty to thirty minutes, every week, no exceptions. Enter last week's actuals. Look at where they differed from what you forecast, and ask why. Roll the forecast forward one week, so you are always looking at thirteen. Then answer one question: which week is my lowest, and what single thing moves it. That question is the entire point of the model.
7. Do it yourself for the first four weeks. After that you can hand the data entry to a bookkeeper. But the judgement about when each customer will actually pay is the asset being built, and it has to be yours first.
THE ARTEFACT
The spreadsheet that comes with this Fix. Five tabs, and three numbers to enter before it works.
Start here. Week 1 commencing, your opening bank balance, your floor. Every date and every balance in the model rolls from those three cells, so you set them once. It also carries the colour key: blue is where you type, black is calculated, green is pulled from another tab.
Invoices. Where receipts get named. Customer, reference, amount, and your expected payment date. The forecast reads from here automatically, so the discipline of step three is built into the structure rather than left to willpower.
Forecast. The thirteen-week grid. Receipts, payments, net movement, opening and closing balance, your floor, and the headroom above it. Named invoices flow in from the Invoices tab automatically. Any week where the closing balance drops below the floor turns red, and one cell at the end of the headroom row shows your lowest week across the quarter, which is the number the whole model exists to produce.
Actuals. The same grid, filled in each Monday for the week just gone.
Variance. Calculated for you: actual against forecast, line by line. This tab is where you learn. A receipt that landed a week late is not a data entry chore, it is information about a customer, and next week's forecast should use it.
Two design decisions matter more than the rest. Expected dates, not due dates. And named receipts, not a revenue line. Both exist to kill the same error: smoothing.
Cash does not arrive evenly. It arrives in lumps, late.
Download the spreadsheet
THE CHECK
A number and a behaviour, with dates.
By week four: your receipts forecast for the week just gone lands within ten percent of what actually arrived, most weeks. And you can name your lowest week and what drives it without opening the file.
By the end of the quarter: no cash surprise you did not see coming at least four weeks out. And every cash conversation in that quarter, with the bank, with a slow customer, with a supplier, was started by you, early, because you saw it in the model. Not by them.
If week four arrives and your variances are wild, the model is still working. It is telling you your assumptions about payment behaviour are wrong, which is worth more than not knowing.
THE AI LEVERAGE
Once the model is live and updated weekly, you can safely hand AI real work in this area.
Give it your thirteen weeks and ask what happens if your largest customer pays four weeks late. Ask which single receipt your quarter actually depends on. Ask it to rank your overdue invoices by impact on the lowest week and draft the chase emails in that order. Ask it to run a scenario where you delay the hire by a month. These are good uses of the tools, and they are safe now because every number in the model has a source you own.
Here is what you would have got wrong before. If you had asked AI for a thirteen-week cashflow without this in place, you would have received one. Confident, well formatted, and built on smoothed guesses: revenue spread evenly across the weeks, payment timings assumed, tax dates approximated. Smoothing is exactly the error this Fix exists to kill, and the output would have looked authoritative enough that you never checked. You had no way to check.
Now you do. When an AI answer disagrees with your model, the disagreement is information, and you can find out which one is right. That is what closing the gap buys. Not the spreadsheet. AI still gives the answer. You now know whether to believe it.
PUBLISHED
LAST REVISED