Cash forecasting with AI: accuracy starts with your accounts receivable

Thursday, 4:00 p.m. Management asks how much cash will be available in eight weeks to fund an inventory purchase. Treasury opens its spreadsheet, adds up yesterday’s bank balances and accounts receivable where some payments are still waiting to be applied.

The figure provided at the end of the day is already outdated. And everyone knows it.

According to the 2025 EACT survey, conducted among approximately 275 group treasurers, cash forecasting is the number one priority for the next 12 to 24 months: 30.5% of treasurers rank it first.

Why is cash forecasting back at the top of the priority list?

Because management demands more accurate figures, more often, in a context where every investment or borrowing decision depends on the cash position. The Strategic Treasurer and TIS (2025) survey measures the trend: management expectations regarding forecasts have increased in 68% of companies, and 76% demand greater accuracy.

Spending is following suit. Forecasting leads the list of technology spending increases, cited by 43% of respondents, ahead of investments in fraud prevention.

The same trend is sweeping through accounts receivable. Quadient (2026) reports that more than 60% of finance departments planned to increase their investments in financial automation, and ranks real-time reconciliation among the dominant trends of 2026.

Is AI making cash forecasting more accurate?

Yes, on one condition: that the cash receipt data feeding the model is complete, applied, and up to date. A predictive model learns from your customers' actual payment habits. If payments remain unapplied for days, it learns from noise.

Adoption is moving fast. According to Strategic Treasurer and TIS (2025), 60% of treasury teams use or plan to use AI for their forecasts, compared to 27% in 2021. That proportion has doubled in four years.

Vendors promise accuracy rates of 90% and higher. These are vendor figures, not a guaranteed average: the accuracy achieved depends first and foremost on the quality of the input data. We see the same principle with agentic AI in ERP financial processes.

Key figures to remember:

• Treasurers' number one priority for the next 12 to 24 months: cash forecasting, ranked first by 30.5% of respondents (EACT, 2025).
• 60% of treasury teams use or plan to use AI for their forecasts, compared to 27% in 2021 (Strategic Treasurer and TIS, 2025).
• 68% of companies report increased management expectations regarding forecasts, and 76% demand more accuracy (Strategic Treasurer and TIS, 2025).
• 43% of respondents place forecasting at the top of technology spending increases (Strategic Treasurer and TIS, 2025).
• 40.4% of Canadian finance and insurance companies use AI, compared to 19.2% across all sectors (Statistics Canada, 2026).

Why do your forecasts fail before the algorithm even kicks in?

Because a forecast inherits the quality of the accounts receivable that feed it, and the cash receipts line is almost always the most fragile. A customer settles 40 invoices with a single transfer, deducts two credit notes, and sends no payment advice. As long as that payment remains unapplied, your actual cash and your forecasted cash are telling two different stories.

In the accounts receivable mandates PlanAxion manages, this manual sorting delays payment application by several days each cycle. Unapplied payments inflate DSO, distort aging reports, and deprive forecasts of their raw material. We detailed the mechanism in our article on DSO and unapplied payments.

A cash flow forecast is only as good as the accounts receivable data that feeds it.

This is why automating cash application must come before the predictive layer. The cash application solutions PlanAxion implements target this exact friction point: applying payments to the correct invoices as soon as they are received, so that forecasts are built on clean data.

Where should you start for reliable cash flow forecasting?

First, ensure reliable cash application, then measure the variance between your forecast and reality over a full cycle, and finally, add the predictive layer. The order matters: reversing these steps simply automates inaccuracy.

This is the approach of our four-week AI workshop: prepare, identify, prioritize, and validate data, then decide and deliver. Every idea passes through the same filter: problem to solve, expected value, available data, and effort required. The investment varies based on scope and is confirmed during a brief exploratory call.

What should you do before the end of the quarter?

Measure two figures this week: the amount of unapplied payments at month-end and the average variance between your forecasts and actuals over the last 13 weeks. If the former exceeds a few days of billing, the priority isn't an algorithm. It’s your accounts receivable.

Frequently asked questions about cash flow forecasting and AI

What is a 13-week cash flow forecast?

It is a weekly projection of cash inflows and outflows over a three-month horizon, updated every week. It is used to anticipate financing needs, plan investments, and identify cash shortfalls before they occur. Its reliability depends first and foremost on the accuracy of customer receipts.

Can AI predict customer payment dates?

Yes. Models learn from each customer's actual habits, such as average delays, seasonality, and partial payments. According to Strategic Treasurer and TIS (2025), 60% of treasury teams are using or planning to use AI for their forecasts. However, the accuracy achieved depends on the quality of payment application data.

Why do unapplied payments distort forecasts?

Because they create a gap between actual cash and accounts receivable. The money is in the bank account, but the invoices remain open in the ERP. The forecast then counts incoming payments that have already arrived, which inflates DSO and distorts aging reports.

Do you need a modern ERP to improve cash flow forecasting?

No. The first step is to ensure reliable cash application and measure the variance between forecasts and reality, which can be done with your existing ERP. AI agents from major vendors are becoming available in existing licenses, but their value still depends on the same clean data.