Of all the challenges faced by treasury teams around the world, cash flow forecasting is one of the most enduring. Accurate cash forecasting is essential when it comes to making informed decisions about debt, investments and working capital – but in practice, forecasting is often hampered by manual processes, poor-quality data and inconsistencies in the forecasting methods used by different areas of the business.

As Marianna Polykrati, Group Treasurer at Greek aquaculture company AVRAMAR, observes, “Cash flow forecasting has consistently been among the top priorities of treasury throughout my career, and I do not see that changing in the future. Regardless of how advanced technology becomes, forecasting will always involve a degree of uncertainty – our personal Odyssey.”

And Kemi Bolarin, Head of Treasury – Europe at GXO, says that while the company has strong liquidity and robust processes, “forecasting cash has become more challenging rather than less. Customer payment behaviour continues to evolve, supply chains remain dynamic, and economic conditions can change the assumptions underpinning a forecast remarkably quickly.”

The trouble with forecasting

James Kelly, former FTSE 100 treasurer and co-founder of AI implementation advisory firm Your Treasury, outlines some of the challenges that treasurers face on this journey. “In a decentralised process, for example, the group forecast may depend on dozens of finance leaders applying different assumptions, different levels of conservatism and different interpretations of risk,” he says.

“The whole exercise then moves at the speed of the last person to submit their numbers, leaving finance with precious little time to challenge the result before it is circulated.”

While treasurers have been grappling with these challenges for decades, changing market conditions have only ramped up the level of difficulty.

Thomas Mehlkopf, General Manager and Head of Working Capital Management CoE, SAP SE, notes: “Most models are still built for a deterministic world that no longer exists. Standard forecasts based on historical averages or contractual due dates assume payment behaviour is stable and predictable.

“In reality, flows are increasingly driven by real-time, condition-based triggers – whether that’s a milestone delivery, an invoice approval or a payment released automatically once operational conditions are met.”

Inflection point

Against this backdrop, the cash forecasting landscape is evolving rapidly. James Mitchell, Head of office of the CFO at FIS, says forecasting has evolved into a strategic capability, with treasury using data points such as behavioural data to enable faster, more informed decisions.

“Organisations have moved away from a single monthly forecast and now use multiple forecasting periods to support everything from daily liquidity management through to strategic planning,” he says.

Today, treasurers are looking for ever greater levels of accuracy – and they also have access to an increasingly sophisticated array of forecasting tools, including AI-powered solutions that can analyse historical data, identify patterns and streamline longstanding manual processes.

Enrico Camerinelli, Strategic Advisor at Datos Insights, says that with businesses expanding beyond single-location operations and increased transaction velocity, spreadsheets simply cannot scale.

“Datos Insights research confirms this inflection point,” he says. “Treasurers managing complex operations recognise that operations have become too intricate for manual processes to provide the real-time visibility required to meet daily obligations and optimise cash positions simultaneously.

“This shift has driven aggressive adoption of forecasting tools. Ninety-three percent of the mid-market businesses (US$20m to US$100m in annual revenue) we surveyed currently use or express interest in cash forecasting tools.”

Meanwhile, treasurers’ expectations of accuracy have also shifted. Camerinelli says that five years ago, a forecast accurate within a week was acceptable. “Today, treasurers operating with tighter margins and less access to emergency financing need daily – sometimes intraday – precision.”

At the same time, sophisticated forecasting technology has become more accessible to a wider range of organisations, thanks to developments such as cloud infrastructure and API-based integration. According to Camerinelli, financial institutions expanding their mid-market offerings are increasingly bundling forecasting with cash management, “positioning it as a native capability rather than an add-on.”

As a result, treasurers no longer need substantial IT investment or multi-month implementations to access forecasting functionality. And as Camerinelli notes, “Deployment timelines have compressed from quarters to weeks.”

Beyond the numbers

So what exactly should treasurers be aiming for as they strive for better cash forecasting? “Thankfully, cash forecasting is moving beyond simply searching for some mythical perfect number,” says Kelly. “No model can completely remove uncertainty from a complex business, nor should that be the goal.

“What matters is whether treasury and the wider finance team can see changes coming, understand what’s driving those shifts, and act before the consequences hit the bank account.”

As such, he says the “technology story” is shifting away from spreadsheet consolidation and retrospective variance analysis. Instead, it’s moving towards transaction-level classification, connected operational data and forecasts that refresh as conditions change.

Kelly says that while automation is undoubtedly part of that shift, it isn’t a magic wand. “A forecast can look reassuringly accurate at group level while concealing serious movements underneath,” he observes. “For example, a £1m net variance might reflect a £10m receipt arriving late and a £9m payment slipping into the following period. The maths says the forecast was close – but the underlying story says the business has two issues it needs to understand.”

Evidence behind the variance

Where forecasting is concerned, explainability matters just as much as accuracy. Treasurers need to know which entities, counterparties and transaction types are creating a variance. As Kelly points out, this distinction becomes particularly important when a major receipt is delayed.

“A missing purchase order number, and a customer withholding payment because it is unhappy with delivery, may look identical in the forecast – yet they require completely different responses,” he says. “One calls for an administrative intervention; the other may demand senior commercial attention, revised external messaging and a reassessment of whether the cash will arrive at all. Technology needs to expose the evidence behind the variance, not simply flag that the expected cash failed to appear.”

Kelly observes that one of the most useful advances is technology’s growing ability to connect financial forecasts with the operational events that shape them, such as project delays, unresolved customer complaints, missing approvals or resource shortages.

“These events might be visible elsewhere in the business weeks before they affect revenue recognition, invoicing or cash collection,” says Kelly. “Finance experiences these events as surprises because the knowledge was trapped upstream.”

Condition-aware forecasting

Mehlkopf says that best practice in cash forecasting has shifted away from static, spreadsheet-driven processes and towards “condition-aware” forecasting.

He adds: “The benchmark treasurers should be aiming for is dynamic accuracy: forecasts that automatically ingest live operational data and adjust as underlying conditions change.”

This is where AI changes the equation – provided it has access to the right data. “Invoice status, PO information and logistics milestones reside natively within the ERP, not the bank layer, enabling treasurers to leverage the deep integration they already own to dismantle data silos and pinpoint optimal cash flow outcomes,” says Mehlkopf.

“AI models embedded in the ERP can probability-weight the likelihood of specific trigger conditions being met, continuously refining timing and volume predictions.”

The practical benefit is a forecast that reflects real-time operational reality, giving treasurers earlier warning of liquidity pressures and the confidence to deploy cash productively – “whether that involves reducing reliance on external funding or optimising working capital positions instead of holding cash defensively.”

Upper limit

So how accurate is accurate enough? As Camerinelli points out, the forecasting challenge is one that will never be fully solved. “Legitimate uncertainty exists at the edge: unexpected customer defaults, economic shocks, regulatory changes, supply chain disruptions,” he notes. “No model, whether human or machine, eliminates that tail risk.”

He argues that the practical upper limit on useful accuracy lies around 90-95% for five-day forecasts and 80-85% for 30-day horizons, depending on the business model. “Beyond these thresholds, marginal accuracy improvements cost disproportionately more than they return. Treasurers must accept residual uncertainty and size liquidity buffers accordingly.”

Rather than chasing a perfect accuracy percentage, Mitchell says treasurers should focus on continuous improvement by comparing forecasts against actuals and testing improvements through small pilots.

“It is also important to challenge whether the data is fit for purpose, because understanding the behaviours behind it often has a greater impact on forecast quality than simply adding more data.”

Direction of travel

In the future, forecasting needs to become even more tightly coupled to operational systems, says SAP’s Mehlkopf. “As we enter the era of agentic payments, where AI agents will soon be able to autonomously negotiate terms and settle transactions, forecasting must account for autonomous financial decisions alongside cash flow predictions,” he notes. “Treasurers who build that ERP-native forecasting capability now will be far better placed to safeguard and control in an increasingly autonomous, real-time world.”

Mitchell predicts that forecasting will become far more connected and continuous, “with real-time data and improved connectivity allowing forecasts to update more frequently and helping treasury move from being responsive to predictive.”

Organisations will bring together data from across treasury, accounts receivable, accounts payable and ERP systems, he says. This will create a connected view of future cash flows across the office of the CFO finance function and give treasury a stronger foundation for forecasting, liquidity management and scenario planning.

“We are also likely to see much greater use of predictive analysis and scenario modelling, with treasury asking not just what the cash position looks like, but what happens if interest rates change, if a major customer pays late, or how an acquisition affects liquidity,” Mitchell adds.

Getting started

So where should treasurers start when seeking to make real improvements to their cash flow forecasting?

Mitchell advises starting with the basics by understanding where the current forecasting process is breaking down, and whether the issue is data quality, timing, ownership, process discipline or access to the right information. “Once that is clear, improvements become much easier to target,” he says.

“Good-quality data does not necessarily mean more data; it means data that is fit for purpose, relevant to the decision being made, available when needed and trusted by the business.”

As such, treasury should measure forecast performance against actuals and review the variances. Key to this, says Mitchell, is not just knowing whether the forecast was right or wrong, but understanding why and using that insight to improve over time.

“Finally, don’t try to forecast everything,” he advises. “Focus on the cash flows that have the greatest impact on liquidity and decision-making, because a simpler, well-governed process will often deliver more value than a complex model nobody fully trusts.”