A profitable company can fail on a Thursday. Revenue was recognised in March, the customer pays in June, payroll runs fortnightly and the tax instalment is due on the 28th. None of that appears in a profit and loss until it is far too late to act. The 13-week cash forecast is the instrument built specifically for that blind spot, and it is the single most useful thing most growing companies do not have.
Why thirteen weeks
Thirteen weeks is one quarter. It is long enough to see the obligations that will hurt — a tax instalment, an annual insurance renewal, a lease review, the payroll fortnight that lands three times in a month — and short enough that you can still name individual invoices rather than modelling averages.
Shorter than about eight weeks and you have a bank statement with ambition; you can see the problem but not far enough ahead to fix it. Longer than about sixteen and the granularity collapses, you start using ratios, and it stops being a cash forecast and becomes a slow-moving budget. Thirteen is the point where precision and usefulness intersect.
It is also the horizon a lender or an incoming investor will ask for by name when things get tight. Producing one on request, already built and already reconciled, says considerably more about management quality than any slide.
Direct beats indirect, and it is not close
There are two ways to forecast cash. The indirect method starts with forecast net profit and adjusts for non-cash items and movements in working capital. It is how the statutory cash flow statement is prepared, and for a 13-week operating forecast it is the wrong tool entirely.
The direct method forecasts the actual receipts and payments, week by week: this customer pays this invoice in week four, that supplier gets paid in week six, payroll lands in weeks two, four, six and eight.
An indirect forecast tells you the bank balance will be tight in week nine. A direct forecast tells you which invoice to chase and which payment to delay so that it is not. Only one of those is an operating instrument.
The indirect method also hides its own errors. Because it starts from profit and applies assumed debtor and creditor days, a wrong assumption produces a smooth, plausible, entirely fictional curve. A direct forecast is built from named items, so when it is wrong, you can see exactly which item was wrong and correct the underlying belief rather than the output.
The line-by-line structure
Thirteen columns, one per week, dated by week-ending. Rows grouped so the eye can find the answer without reading every line.
Two rows are more important than they look. Undrawn facility matters because your real liquidity is cash plus what you can draw, and a forecast that ignores an unused overdraft will trigger a panic you did not need to have. Total liquidity matters because it is the number a board should be governing to, not the bank balance in isolation.
Where each line comes from
The forecast is only as good as its inputs, and every input should have a named owner and a system of record. If a line is sourced from "the sales team's view", it will be wrong and nobody will be accountable for it.
| Line | Source of truth | Owner | Common failure |
|---|---|---|---|
| Named debtors | Aged receivables plus the collections log | Credit control | Using invoice terms rather than that customer's actual behaviour |
| Recurring collections | Billing platform schedule | Finance | Ignoring failed-payment and churn rates |
| New sales receipts | CRM, probability-weighted, plus a delivery lag | Sales lead | Booking the sale in the week it is won rather than paid |
| Payroll | Payroll system, by actual pay date | People / payroll | Averaging monthly instead of dating the fortnights |
| Tax | Obligations calendar per entity | Finance | Missing provisional tax and terminal tax dates entirely |
| Suppliers | Aged payables plus approved purchase orders | Accounts payable | Omitting committed spend that has not yet been invoiced |
| Debt service | Facility amortisation schedule | Treasury | Forgetting fees, and forgetting the covenant test date |
| Capital expenditure | Approved capex register | Operations | Deposit and final payment treated as one cash event |
Four errors that make it useless by week five
1. Forecasting collection on invoice terms
Your terms say 20th of the month following. Your largest customer pays on day 62 and always has. A forecast built on terms rather than observed behaviour will be wrong in the same direction every single week, and the error compounds. Build a collection profile per significant customer from at least twelve months of actual payment history, and use that. The gap between terms and behaviour is usually the single largest error in a founder-built forecast.
2. Monthly thinking in a weekly model
Payroll divided by 4.33. Rent spread evenly. Tax as one twelfth. Every one of these smooths away exactly the spikes the forecast exists to reveal. Three payroll runs land in some months. GST is quarterly, not monthly. Insurance renews once a year and it is a large number. Date every payment to the day it actually leaves the account, then allocate it to the week it falls in.
3. No variance loop
A forecast that is never compared with what happened cannot improve, and nobody will trust it by week five. Every week, put last week's forecast next to last week's actual, line by line, and write one sentence on anything material. Within a quarter you will know precisely which lines you systematically over-forecast, and the model gets better because the beliefs behind it get better.
4. One consolidated number for a group
Group cash is not one figure. It is cash by entity, by currency and net of what is trapped. The consolidated balance can look comfortable while the entity that has to pay Thursday's wages is short, and intercompany transfers are rarely instant, sometimes taxable and occasionally restricted by a lender's terms. Forecast each entity, then consolidate — never the other way round.
Ask whoever maintains the forecast: "What did we forecast for last week, and what actually happened?" If the answer takes more than thirty seconds, the variance loop does not exist and the forecast is decoration.
Running it weekly
Same day every week — Monday works because the prior week has closed and the coming week can still be changed. Roll the window forward one column so it always looks thirteen weeks ahead. Load last week's actuals, review the variance, update the beliefs that were wrong, and circulate one page: closing balance by week, total liquidity, the lowest point in the window, and the two or three decisions that need making.
That last part is the point. A 13-week forecast is not a reporting artefact. It is a decision instrument, and if nobody makes a different decision because of it, you have built the wrong thing.
XLCFOChartered Accountants building AI-native finance functions for founders and boards. We maintain 13-week cash for every client on the Control tier and above.
