Your 13-Week Forecast Hides the Day You Go Negative

A founder I work with checked her cash forecast on a Monday and felt fine. The tool showed every week of the next quarter in the black, the tightest one still netting positive by $90K. On Wednesday her payroll run bounced. The week wasn't short. Payroll cleared Tuesday, and the $400K receivable that made the week whole didn't land until Friday. Her forecast was right about the week and useless about the three days in the middle of it.

Thirteen weeks is a treasury habit, not a law

Thirteen weeks became the treasury standard for good reasons. It's long enough to see a quarter coming and short enough to stay grounded in real receivables and real bills. AI made it sharper this year, and that part is genuine. AFP's 2026 treasury technology survey found 52% of US corporate treasurers piloting or running AI for cash forecasting, nearly double two years ago. Manual forecasts land around 60% accurate at the 13-week horizon. The AI tools claim 88 to 92%, pulling from live bank feeds and your own payment history.

So the curve got tighter. Here's the problem nobody flags at the demo: accuracy at the week level is the wrong test for a company your size. A $400M enterprise with a $40M cash cushion cares about the quarterly shape. A $9M company cares about whether Thursday clears.

Netting is where the danger hides

A weekly bucket does one thing that quietly breaks for smaller balances. It nets every inflow against every outflow inside the same seven days, then shows you the leftover. Money that arrives Friday gets credited against a bill that cleared Monday, as if timing didn't exist.

Run the founder's week. Starting cash $120K. Payroll of $180K clears Tuesday. A $400K customer payment posts Friday. Net the week and you're up $340K, a healthy line on any dashboard. Walk it day by day and Tuesday afternoon you're sitting at negative $60K, four days before the money that covers it shows up. The week was never the risk. One Tuesday was.

A weekly forecast tells you whether the money exists. It doesn't tell you whether it exists on the day the payment leaves. For a growing company those are different questions, and only one of them bounces a check.

This gets worse, not better, as the tools improve. A more accurate weekly number reads as more reassuring, so the founder relaxes the buffer she used to keep on instinct. The forecast earned more trust and gave back less of the one signal that protects her.

Forecast the trough, not the total

The fix isn't a better model. It's a better unit. Pull the forecast to a daily grid for at least the next four weeks and find the single lowest point, not the period balance. That number, the worst Tuesday, is your real floor. Size your operating cushion to it.

Then check the sequence behind it. Most intra-week troughs trace to two or three predictable events colliding: payroll, rent, a tax payment, all landing before a big receivable you don't control. Once you can see the collision, you can move it. Shift a vendor payment three days. Ask the anchor customer for net 15 on the invoice that always lands late. Time the deposit run.

Your forecast can be 90% accurate on the quarter and still walk you straight into an overdraft. Stop reading the week. Read the day the cash leaves.