Most startups build a cash flow forecast once, usually in the week after a board member asks for one, and never open it again. It is a twelve-month projection derived from the annual budget, it is wrong by the second month, and it is useless for deciding whether to sign a lease or delay a hire.
The forecast that companies actually use every week is shorter, more granular, and built from a completely different set of inputs. It is the 13-week rolling cash flow forecast, and it is the standard tool in restructuring and turnaround work for a reason: when cash is tight, it is the only forecast horizon where the numbers are both knowable and actionable.
Why 13 weeks
Thirteen weeks is one fiscal quarter. It is long enough to see a problem coming with time to do something about it — enough runway to accelerate collections, renegotiate a vendor payment, delay a hire, or start a bridge conversation. And it is short enough that you actually know most of the inputs.
Look at the horizon tradeoff directly. Over the next two weeks, you know nearly everything: payroll is scheduled, AP is in the system, the invoices you will collect are already issued. Over the next twelve months, you know almost nothing with precision — the customers who will generate half of next year’s collections have not been sold yet.
Thirteen weeks sits at the point where accuracy is still high and the horizon is still long enough to matter. Weeks one through four are close to certain. Weeks five through eight are well-informed estimates. Weeks nine through thirteen are directional. That gradient is fine, because you rebuild the forecast every week and the uncertain weeks keep converting into certain ones.
The word “rolling” carries as much weight as the number. Every week you drop the completed week and add a new week 13. The forecast never expires and never becomes a stale artifact from last quarter.
Direct method, not indirect
This is the distinction that determines whether your forecast is operationally useful or just an accounting exercise.
The indirect method starts with net income and adjusts for non-cash items and working capital changes to arrive at cash flow. It is how the cash flow statement in your financial statements is prepared, and it is correct. It is also useless for operating decisions, because it tells you cash changed by $180,000 last month without telling you which customer paid, when, or what you can do about next month.
The direct method forecasts actual cash receipts and actual cash disbursements, line by line, dated. It asks two questions: what money is coming in, from whom, on what date; and what money is going out, to whom, on what date.
Only the direct method produces an actionable output. When week 9 shows a negative closing balance, a direct-method forecast tells you it is because a $240,000 receipt from your largest customer lands in week 10 while payroll and a quarterly tax payment both hit in week 9. That is a solvable problem — call the customer, ask for early payment, or move the discretionary spend. The indirect method just tells you the quarter looks tight.
Building the model
The structure is simple. Thirteen columns, one per week, each week following the same block:
Opening cash — the actual bank balance at the start of week 1, and the prior week’s closing balance for every week after. Use your real available balance across all operating accounts. Exclude restricted cash, security deposits, and anything you cannot spend on Monday.
Receipts, forecast by source:
- Collections from existing accounts receivable — this is the largest and most forecastable line. Take your AR aging, and for each open invoice, estimate the week it will actually be collected.
- Collections from revenue not yet invoiced — work you will deliver and bill during the forecast period, with collection landing 30 to 60 days later depending on your real DSO.
- Non-operating receipts — investor funding, loan draws, tax refunds, R&D credits, grant disbursements, insurance proceeds. Forecast these only when they are contractually committed with a known date. A term sheet is not a receipt.
Disbursements, forecast by category:
- Payroll and payroll taxes, on actual pay dates
- Benefits, insurance, and employer contributions
- Accounts payable, by vendor, on the date you actually intend to pay
- Rent and facilities
- Debt service — principal and interest separately
- Sales, payroll, and income taxes
- Contractors and professional services
- Software, subscriptions, and recurring tooling
- Capital expenditures
- Everything else, itemized rather than lumped into a “miscellaneous” line
Net cash flow — receipts minus disbursements for the week.
Closing cash — opening cash plus net cash flow. This becomes next week’s opening balance.
Add a line beneath the model for your minimum operating cash threshold — the balance below which you cannot function. Seeing closing cash cross that line in week 8 is the entire point of the exercise.
Forecast the collection date, not the invoice date
This is the single most common modeling error, and it makes an otherwise sound forecast worthless.
If you invoice on net 30 and your customers actually pay in 52 days, forecasting collections at 30 days makes every week look better than it will be. The error compounds: each week you forecast cash that does not arrive, and each week you roll the shortfall forward into the next week, which now also looks fine. By the time reality catches up, you have lost a month of warning time.
Forecast from behavior, not from terms. Pull twelve months of payment history by customer and calculate actual average days to pay for each one. Your enterprise customer with a procurement department and a 45-day approval cycle should be forecast at 45 days, not the 30 printed on the invoice. Your customer who has been late on the last four invoices should be forecast late on the fifth.
For meaningful accounts, forecast individual invoices by name and expected date rather than applying a blended average. A single large receipt landing one week later than assumed can be the difference between a comfortable quarter and a covenant breach.
The variance review is the discipline
A forecast you build and file is a document. A forecast you compare against actuals every week is a management system. The comparison is what makes the model improve, and it is what makes the numbers credible when you put them in front of a board or a lender.
Run it the same day each week. Take the completed week, put forecast next to actual for every line, and calculate the variance. Then answer one question for each material gap: was this a timing difference or an amount difference?
A timing difference means the cash is still coming, just later or earlier than modeled. A customer paid in week 4 instead of week 3. The fix is to the forecast assumption, and the cash moves forward.
An amount difference means the cash is not coming, or more is going out than expected. A customer paid short. A vendor invoice was larger than estimated. An expense appeared that no one modeled. This is the more serious category, and it usually points at a process gap rather than a forecasting gap.
Track your forecast accuracy over time. Within a few cycles you should be within 5% on week 1 and within 10 to 15% on week 4. If you are not, the misses will cluster around a specific line — almost always collections timing or a category of disbursement that someone is spending without visibility.
Then roll it forward. Drop the completed week, add a new week 13, and update every assumption you learned something about.
Failure modes that will break your model
Forecasting from invoice date instead of expected collection date. Covered above, and worth repeating because it is the most common and most damaging error.
Missing the three-paycheck month. On a biweekly payroll, 26 pay periods do not divide evenly into 12 months. Twice a year, three payrolls land in one month. If you have built your forecast on a monthly payroll assumption and converted it to weeks, you will miss an entire payroll — which for most startups is the largest single disbursement they make. A weekly model built on actual pay dates catches this automatically. A monthly model divided by 4.33 does not.
Ignoring quarterly and annual true-ups. Payroll tax deposits, estimated income tax payments, sales tax remittances, annual insurance renewals, workers’ comp audits, 401(k) true-ups, software contracts on annual billing. These are lumpy, they are predictable, and they are routinely omitted from forecasts built by extrapolating a typical month. Sit down with the last eighteen months of bank statements and find every payment over a threshold that did not occur monthly. Put each one on the calendar.
Treating uncommitted funding as a receipt. A funding round in diligence, a loan in underwriting, or a grant under review is not cash. Model your forecast without it and treat it as a separate scenario. If the base case requires money you have not signed for, you do not have a forecast — you have a hope.
Building it in the accounting system instead of alongside it. Your general ledger is organized around accrual periods and GL accounts. A cash forecast is organized around dates and counterparties. Build it in a spreadsheet, pull the inputs from your accounting system, and keep them separate.
Delegating it entirely. Someone has to own the model, but the founder or CEO needs to read it every week. The forecast is the mechanism by which cash problems become visible early enough to solve. That visibility does not help anyone if it stays inside the finance function.
What it looks like when it works
The output is a single number per week — closing cash — and a line showing the point below which you cannot operate. Nothing else in your financial reporting answers the question the forecast answers, which is whether you can meet your obligations over the next quarter and what you would need to change if you cannot.
Companies that run this weekly stop having cash surprises. They still have cash problems, but the problems appear in week 9 of the model rather than in a Thursday phone call from the bank, and eight weeks of warning is usually enough to fix them.
The forecast tells you what happens over the next quarter. The related question — how many quarters you have left at your current rate of spend, and which levers actually extend that number — is a different calculation.
Pegacorn Group provides fractional CFO and back-office support to venture-backed startups, nonprofits, and public companies. If your cash forecast lives in a file nobody opens, we can help you build one that gets used.