Your Cash Forecast Is Built on Due Dates

A founder pulled up his 13-week cash flow forecast in our first meeting. Clean spreadsheet, week-by-week receipts and disbursements, ending balance out to Q2.

I asked how the receipts were modeled.

"Invoice due dates," he said. "Net 30."

That's not a forecast. It's a due date report.

Due dates aren't payment dates

When a customer owes you money in 30 days, the question isn't when the invoice is due. It's when that specific customer actually pays. Those numbers are different for every customer in your book, and ignoring the difference compounds across every week of the forecast.

Visa published findings last month from their 2026 Growth Corporates Working Capital Index, surveying nearly 1,500 CFOs and treasurers. Companies that adopted AI for working capital management cut cash flow uncertainty from 68% to 17%. Real numbers, not projections. But the improvement came with a condition: the AI needs historical payment pattern data to work from.

Companies running Net 30 terms often have customers who pay in 38 days, 52 days, or 18 days. The distribution matters. A customer who reliably pays in 52 days isn't late. They're predictable. Treating them as a 30-day payer builds a structural error into every weekly cash model you produce.

What better forecasting actually requires

Accurate cash flow models aren't built from invoice terms. They're built from payment history.

Pull your AR aging data from the last 12 months. For each customer, look at actual days to payment — not their contractual terms. Calculate the average, and the variance. Some customers are consistent. Others are all over the map. The consistent ones can be forecast reliably. The variable ones need to be discounted or widened into ranges.

That analysis takes a few hours in a spreadsheet. Most teams don't do it because it requires pulling historical data, not running a standard report. Standard reports show what's currently open. They don't show when that customer paid the last 15 invoices.

Once you have payment pattern data, even a basic model gets materially more accurate. Not because you've added AI, but because you've replaced fiction with observed behavior.

AI cash forecasting cuts uncertainty dramatically. Companies that see those results have payment pattern data going in. Companies that don't are feeding the model due dates and wondering why it's still wrong.

Where to start

Take your top 20 customers by revenue. For each one, calculate average days to pay over the last six months. Flag anyone with high variance. Build weekly receipt projections from those actuals, not your standard terms.

The model will look different. Some customers you've been treating as Net 30 are actually Net 45. Others pay faster than expected. The overall forecast will likely show cash arriving later than your current model suggests.

That's uncomfortable. It's also accurate. Accurate, even when the news is slower cash, is better than a forecast that makes next quarter look fine when it isn't.

The tool problem is easier to solve than the data problem. Most companies at $5M-$20M haven't solved either.