How much data does AI cash forecasting need?

Models need 12 to 24 months of transaction history to learn seasonality. Payroll, receivables, and card settlement forecast well. One-off strategic items do not, and an acquisition or new entity resets part of the model's learning.

How Accurate Is AI Cash Flow Forecasting? Variance Benchmarks

AI cash flow forecasting cuts forecast variance by over 60% against manual spreadsheet methods. Manual teams commonly run 10% to 20% variance. Accuracy depends on categorisation quality, data history, and flow type, not on the model alone. Short-range, high-volume flows improve most.

Categorisation sets the ceiling

A forecast is only ever as good as the classification beneath it. Miscategorised intercompany sweeps, FX settlements, and debt flows mask the signals you need. Palm reports 90%+ categorisation accuracy, and that number matters more than any headline forecast claim.

Three factors decide the outcome

Data history. Models need 12 to 24 months of transaction history to learn seasonality.
Flow type. Payroll, receivables, and card settlement forecast well. One-off strategic items do not.
Business stability. An acquisition or a new entity resets part of the model's learning.

Measure variance the right way

Percentage thresholds mislead. A 100% variance in a small account may not matter. A 5% deviation in your largest account can make the whole forecast unreliable. Weight variance by impact, then fix the categories that move the number. Variance analysis: a treasurer's guide covers the method in full.

Separate timing from amount

A payment that landed on Thursday instead of Tuesday is a timing variance. A payment 40% larger than forecast is an amount variance. They have different causes and different fixes. Collapsing them into one figure hides both. See cash variance analysis.

What to ask a vendor

Request variance measured against actuals, broken down by forecast horizon and cash category. A single accuracy figure with no horizon attached is marketing, not a benchmark. How to choose an AI treasury platform lists the questions that get past the demo.

What it looks like in production

Palm reports variance reduction of over 60% in production. ON unlocked $350M in additional cash investments after tightening its forecast.

Want your own baseline first? Run a Forecasting Health-Check, and read where companies fall short with cash forecasting.

How accurate is AI cash flow forecasting?