Accuracy is part of the forecasting problem.
Cash flow forecasting is often discussed as a search for greater accuracy. That is understandable. Treasury teams want better predictions, CFOs want reliable liquidity planning, and finance leaders want fewer surprises. But accuracy is only one part of the problem. A forecast is useful only when finance trusts the inputs behind it. If the data is incomplete, late, inconsistent, or difficult to explain, even a technically sophisticated model will struggle to create confidence.
A forecast is a process output
This is why forecasting cannot be treated purely as a modelling exercise. A forecast is the outcome of a process. It depends on who provides the input, how categories are defined, how often data is updated, which assumptions are used, how variances are explained, and whether actuals are compared back to earlier submissions. When that process is weak, the forecast becomes a number that people debate rather than a decision tool they trust.
The importance of this issue is clear across recent treasury research. The EACT Treasury Survey 2025 places cash-flow forecasting among the top priorities for corporate treasurers, while the Nomentia Treasury Trends Report 2026 shows that inaccurate or inconsistent forecasting inputs remain a major challenge, particularly for mid-market companies. This tells us something important: forecasting is not a mature solved problem. It remains one of the most visible pressure points between treasury, finance, subsidiaries, and the CFO office.
Where forecast trust often breaks down
In many organisations, the problem starts with input ownership. Forecast data often comes from local entities, business units, accounting, accounts payable, accounts receivable, sales, procurement, tax, and treasury itself. Each source may understand its own numbers, but not always the impact on group liquidity. Some inputs are submitted late. Some are submitted in the wrong category. Some are based on optimistic business assumptions. Some are copied from last month with small adjustments. Some are kept outside the formal process until someone asks for them. The result is a forecast that may be mathematically complete but operationally fragile.
A second problem is category discipline. If entities use different definitions for customer receipts, supplier payments, tax, payroll, internal funding, or extraordinary items, the consolidated forecast becomes hard to compare. One entity may classify a movement as working capital, another as financing, and another as a one-off. When categories are inconsistent, variance analysis becomes unreliable. Finance then spends time explaining definitions rather than interpreting the business impact.
A third problem is timing. Forecasting is highly sensitive to when data is captured. A forecast submitted before a large customer payment delay will look different from one submitted after the update. A forecast that includes approved payment runs will differ from one based on open invoices only. A weekly forecast may show risks that a monthly forecast hides. This does not mean one horizon is always better than another. It means finance needs clarity about the timing and status of each input.
Why forecast confidence matters
This is where forecast confidence becomes as important as forecast accuracy. Forecast accuracy asks whether the forecast matched reality. Forecast confidence asks whether the organisation understands why a forecast is reliable or not. A forecast may be inaccurate because the business changed, because an assumption was wrong, because data was missing, because an entity submitted late, or because the model did not capture a pattern. These are different problems. They require different fixes.
How AI supports better forecast discussions
AI-supported forecasting can help, but only when the input foundation is strong enough. An AI-generated reference forecast based on historical cash flow data can provide a useful baseline. It can help treasury compare manual submissions against historical patterns and identify unusual deviations earlier. It can challenge assumptions, highlight outliers, and support better conversations with entities. But AI cannot fix poor process discipline by itself. If the data history is inconsistent or the forecast categories are poorly managed, the model may simply learn from noise.
The best use of AI cash flow forecasting is therefore not to replace treasury judgement. It is to strengthen the discussion. When treasury has a system-generated reference forecast, a manual forecast, and actual outcomes in one process, the team can ask better questions. Why is this entity expecting a much higher inflow than the historical pattern suggests? Why does payroll move differently this month? Why are supplier payments lower despite known commitments? Why did last month’s forecast miss actuals? These questions improve accountability without creating blame.
Trust in forecasting also depends on feedback loops. A forecast process that only collects inputs and produces a number will not improve quickly. A better process compares forecast to actuals, explains variances, tracks recurring errors, and improves assumptions over time. This helps finance teams see where data quality is strong, where input behaviour needs attention, and where the model needs refinement. Over time, forecasting becomes less about chasing numbers and more about improving the reliability of the whole liquidity planning process.
What CFOs need from a trusted forecast
For CFOs, this matters because forecast trust directly affects decisions. Funding plans, investment choices, working capital actions, debt drawdowns, intercompany lending, hedge timing, and liquidity buffers all depend on the forecast. If leadership does not trust the inputs, decisions become more conservative. Cash may be left idle. Funding may be arranged earlier than necessary. Business plans may be challenged late. Opportunities may be missed because the organisation lacks confidence in its own view of the future.
For treasury teams, a practical starting point is to audit the forecasting process, not only the forecasting output. Which entities submit late? Which categories show the largest variances? Which assumptions are repeated without review? Which data points are manually adjusted? Which parts of the forecast could be supported by historical reference forecasts? Which stakeholders see the variance analysis? These questions reveal where forecast accuracy is being limited by process quality.
The useful forecast is the one finance can act on
A useful forecast is not the most complex forecast. It is the forecast that finance can trust, explain, and use. That requires reliable inputs, clear ownership, structured categories, transparent assumptions, and a feedback loop between forecast and actuals. Better models can improve forecasting. But better inputs and stronger accountability make the forecast usable.
Read more blogs of this series:
#4 From cash visibility to decision visibility: What CFOs expect from treasury now
