The existing forecast remains valuable, but it no longer has to stand alone
Cash flow forecasting often combines ERP data, subsidiary input and treasury judgement. That business view contains information historical data cannot provide and remains important for planning.
Nomentia Predictive Forecasting adds an AI-generated forecast from historical actual cash flow data. It can sit alongside existing liquidity and subsidiary forecasts, creating a second view treasury can compare, challenge and, where appropriate, use directly.
The result is a more flexible forecasting process rather than a single mandated source.
Historical actuals provide the training basis
The AI-generated forecast is built from historical liquidity transactions already available to the forecasting process. The data can include amounts, dates, cash flow categories, companies, bank accounts and currencies.
Using actual cash flows gives the model a behavioural basis: what has really happened over time rather than what was previously planned to happen.
The quality and length of that history matter. Nomentia makes data quality and forecastability part of the process because a model should not encourage users to treat every dataset as equally predictable.
Different cash flow patterns can use different models
One forecasting method rarely fits every time series.
Nomentia Predictive Forecasting evaluates multiple statistical and machine-learning approaches and selects the best-performing model for each time series. A recurring payroll pattern may behave differently from customer collections, taxes or less regular supplier payments.
Automated model selection reduces the need for treasury to choose and maintain the statistical method manually across large numbers of entities, accounts and cash flow categories.
Seasonality and forecastability become visible
Historical cash flows often contain seasonal patterns, structural trends and irregular movements.
The modelling process can detect seasonality and provide forecastability and accuracy metrics for individual time series. Treasury can therefore see which parts of the forecast have a strong historical signal and which require more caution.
That visibility helps users decide how much weight the AI-generated result should carry in the planning process.
The forecast can cover short-, mid- and long-term horizons
The AI-generated output can support short-, mid- and longer-term planning depending on the available data and configuration.
This is useful where detailed manual forecasting is strong in the near term but becomes thinner further out. Historical-data-based forecasting can extend the forward view without requiring every subsidiary to produce the same level of manual detail across the full horizon.
For treasury, this can reduce blind spots in the liquidity outlook.
Comparison creates an independent challenge to the business forecast
The most immediate use case is side-by-side comparison.
If the subsidiary forecast and the AI-generated forecast agree, treasury gains another piece of evidence that the current expectation is plausible. If they diverge, the difference gives the team something specific to investigate.
The goal is not to declare that the model is right and the business is wrong. It is to identify where the two sources of information tell different stories before a liquidity decision is made.
Treasury can decide when the AI-generated forecast becomes the working forecast
For some predictable cash flows, the AI-generated result may be strong enough to use directly.
That gives treasury the option to reduce manual forecasting effort, particularly for entities or categories where historical patterns are stable and local business input adds limited additional value.
For irregular or event-driven cash flows, the AI-generated forecast may remain primarily a reference. The user can apply a different level of human involvement to different parts of the forecast rather than forcing one process across everything.
Forecast coverage can expand without expanding the treasury team
A common constraint in group forecasting is the number of entities treasury can realistically cover in detail.
Historical transaction data already flowing through Liquidity can be used to generate forecasts for a broader set of entities and cash flows, including areas that may currently sit outside the manual process.
This is particularly useful for central treasury teams managing many subsidiaries with limited local forecasting resources.
The output returns to the liquidity reporting process
The AI-generated forecast is consumed in Nomentia Liquidity Reports alongside actuals and other forecast materials.
This keeps the result inside the same planning context where treasury reviews cash positions and makes funding or investment decisions. The forecast does not need to become another separate file or data-science output that has to be translated back into the treasury process.
Scheduled reruns can also keep the AI-generated view current according to the organisation’s forecasting cycle.
Better forecasting comes from using the right source for the right cash flow
Business input, historical data and treasury judgement each contribute something different.
Nomentia Predictive Forecasting gives finance teams a structured way to add statistical evidence to that mix. It can challenge existing forecasts, extend horizons, fill coverage gaps and reduce manual forecasting effort where the data supports it.
The practical outcome is a forecasting process that gives treasury more evidence and more choice before it acts on the future cash position.
Frequently Asked Questions (FAQ)
What does Nomentia Predictive Forecasting use as input?
It uses historical actual liquidity transaction data, including dimensions such as amounts, dates, cash flow categories, companies, bank accounts and currencies.
Does the AI-generated forecast have to remain a reference?
No. Treasury can compare it with existing forecasts, use it as an independent reference, or use the predicted result directly for suitable cash flows and entities.
How does the system choose a forecasting method?
Different statistical and machine-learning approaches are evaluated, and the best-performing model is selected for each time series.
Why are data-quality and forecastability indicators important?
They show treasury where the historical basis is strong enough to support the forecast and where irregular data or limited history means more caution and human judgement are required.