AI cash forecasting: when machine learning beats the 13-week spreadsheet

Global reference

• Scope: Corporate treasury and finance teams
• Topic: AI and machine learning in short-term cash forecasting
• Evidence: TIS and EuroFinance study, July 2026

The state of play

• 46% of treasury teams are evaluating AI, more than any other technology.
• 38% see cash forecasting as the most suitable AI use case.
• Only 4% have AI processes in production.
• More than half of treasurers see data quality as the biggest drag on forecast accuracy.

When machine learning wins

• Many small, repeating flows: retail receipts, card payments, regular customer payments.
• Customer payment timing: AI learns which customers pay late and by how much.
• Several years of clean bank and ERP history.
• Daily or weekly forecasts where speed matters.

When the 13-week model wins

• Large one-off items: tax payments, dividends, capital spending, debt repayments.
• Contract-driven flows with known dates.
• New businesses, acquisitions or changes in payment terms with no history.
• Board and lender reporting that needs clear, explainable assumptions.

The best of both

• Use AI for the high-volume operating flows.
• Keep known and one-off items in the 13-week model.
• Combine both into one forecast and track error by line.

What to do now

• Measure your current forecast accuracy by week and by flow type.
• Clean bank transaction categories before training any model.
• Pilot AI on one entity or one cash flow type for a quarter.
• Use our 13-Week Cash Flow template as the baseline to beat.

Our take

AI should earn its place by beating your current forecast on measured error. Most treasuries will end up with a hybrid, and that is the right answer.