The 13-week cash flow forecast: why the old instrument still wins

Before I ran finance at a startup, I ran treasury operations at a payments company with roughly 40,000 merchant accounts on the books. Every treasurer I worked with, without exception, forecasted cash in the same unit: 13 weeks, rolling, updated at least weekly. Not monthly. Not quarterly. Thirteen weeks, one week at a time.

That's not a superstition. A month is too coarse a unit to catch the thing that actually kills a company: a specific week where two large outflows land before a large inflow clears. A monthly forecast can show a perfectly healthy month on average while hiding a week in the middle of it where the account would have gone negative if a customer had paid three days later than usual. Average the month and the trough disappears. Zoom into weeks and it's the whole story.

Why 13 weeks specifically

Thirteen weeks is roughly one fiscal quarter, which matters for two practical reasons. First, it's long enough to see a full payroll and rent cycle play out more than twice, so seasonal or cyclical patterns show up rather than looking like noise. Second, it's short enough that near-term weeks can be built from confirmed, contractual data — signed invoices, scheduled payroll, known rent — while only the far end of the window relies on projection. A 52-week forecast is almost entirely projection past week 6 or 8; a 13-week forecast keeps roughly a third of its horizon anchored to things that are already true.

The discipline is in updating it every week, not building it once a quarter and referring back to it. A forecast built on Monday and never touched again is a snapshot of what you believed a week ago, not a live picture of where you're headed.

Where the spreadsheet version breaks

Almost every finance team we talk to before they adopt Dibein already has a 13-week forecast. It's usually a spreadsheet, usually maintained by one person, and usually rebuilt from a combination of a bank export, an AR aging report, and memory of which customers tend to pay on time. It works, until three things happen at once, which they eventually do:

• The person who maintains it goes on leave, or leaves the company, and the tribal knowledge about which customers are reliable leaves with them.
• The business grows past the point where one person can hold the payment behavior of every customer and vendor in their head, and the forecast quietly reverts to a straight-line guess.
• A board member or investor asks for a confidence range instead of a single number, and there's no way to produce one without redoing the whole exercise three times with different assumptions.

None of these are spreadsheet problems exactly — Excel and Google Sheets can technically do all of this. They're process problems. Nobody has time to rebuild a payment-behavior model for 80 customers and 40 vendors every Monday morning, so the forecast degrades into a straight-line extrapolation of last month, which is precisely the kind of forecast that misses the week that matters.

What changes when it's a live system instead of a weekly chore

Cash Flow Copilot exists because we ran into this exact failure mode ourselves, twice, before we built it. The mechanical shift is small: instead of a person rebuilding the forecast from a snapshot once a week, a system rebuilds it continuously from the same underlying data — every cleared payment, every new invoice, every payroll run — and holds a separate, per-counterparty model of payment timing instead of one blended assumption for the whole business.

The more important shift is what that makes possible. Once the forecast is cheap to regenerate, it stops being a static report and starts being a place to ask questions: what if this customer pays two weeks late, what if we draw the credit line now instead of in six weeks, what does the tightest week of the quarter actually look like today versus what it looked like a month ago. That's the entire premise behind Scenario Studio — it's only useful because the baseline underneath it is trustworthy enough to fork.