What are the prerequisites for AI-based cash flow forecasting?

The key to AI-supported cash flow forecasting is the data base. This must go back at least three years and contain sufficient details on historical cash flows so that they can be recognised by the algorithm and the cash movements can be assigned to the correct planning category. In order to create accurate forecasts with AI, the relevant expertise is also required. If this knowledge is not available within your team, you will need external consultants or a software solution such as Nomentia Cash Flow Forecasting.

How much data history do I need for AI to make sense?

It is recommended to have at least three years of history in order to be able to make accurate forecasts and analyses. A common misconception here is that if you have a forecast period of X time units, you need 3 times X as history. This is incorrect. Even if you are only forecasting 13 weeks, for example, you need a period of three years history to be able to identify seasonality and outliers. As an extreme example: If you want to forecast week 4 based on 3 weeks history, you can immediately see that this does not make sense because there is no intra-month seasonality available from previous years and the amount of data is far too small.

Where does the historical cash flow data for the forecast come from?

Ideally, the data can be derived from the ERP system through document chain analysis. This is the most effective option, as the data already contains the necessary details required for categorizing payment transactions. However, especially in a heterogeneous ERP landscape or when working with ERP systems that do not classify transactions, document chain analysis is often not feasible. In such cases, a secondary approach must be used, where the data is taken from bank statements. With this method, payment references can provide some insight into the category of the cash flow, but the level of detail is significantly more limited compared to document chain analysis.

Can AI recognise the payment behaviour of individual creditors, or is this done on a general basis?

AI can be used to recognise the payment behaviour of each individual creditor. However, when data volumes are low, similar creditors are often clustered according to country, etc.

How do you select the right models for forecasting?

Different models can yield very different results on the same historical data, depending on the data itself. That’s why it’s crucial to select models based on accuracy, speed, and transparency. Nomentia automatically preselects suitable models based on your data and requirements, but users can also add or exclude models manually. The reasons why a model has been chosen - like for example assumptions like seasonality or holiday effects - are clearly outlined in a report. The definite best model is chosen by running the forecast with the preselected models and comparing the results against actual historical data.

How to prevent "overfitting" of the models?

The risk of a model learning the training data ‘by heart’ and thus being unable to generalise to new data is prevented by validating the models with historical data and selecting the most suitable model. Another key factor here is the selection of parameters for the models, which are determined using a dedicated neural network.

How long does it take to calculate a forecast with AI?

The calculation time depends on the amount of data and the complexity of the models. However, it usually only takes a few minutes per company and category.

Does the use of AI for cash flow forecasting pose a data protection risk?

The type of AI used in cash flow forecasting – also known as predictive analytics – is based on selected historical payment data and algorithms that recognise patterns in this data. If AI-based planning is carried out using a software solution, the algorithms are pre-programmed in the system and the liquidity forecasts are calculated within the system itself. Sensitive cash flow data is therefore processed in the secure environment of the solution and is not passed on to external third-party providers.

How transparent are forecasts generated using AI?

Some providers of AI-based cash flow forecasting, including Nomentia, offer graphical representations that break down the factors included in the forecasts. This creates transparency about how the forecasts were calculated and makes them more transparent. Nomentia also provides a detailed report for each forecast on the internal and external factors taken into account, which can highlight observations and trends that a treasurer might not be able to identify with the naked eye, thus ensuring a deeper understanding of the various influencing factors.