In many companies, rolling forecast projects are on the rise as the understanding of the need for faster planning cycles is increasing. This is due in part to the numerous crises and changes that have made the business environment increasingly unpredictable in recent times.
A rolling forecast is not only important for being able to quickly respond to changes in turbulent times, but also to improve planning quality through continuous variance analysis and plan-to-actual comparisons.
Simply repeating the traditional annual planning is not enough to achieve the desired effect. Often, forecast models are too complex and opaque, so the information is already outdated by the time it is completed. This necessitates freezing the numbers. Statements such as "That was calculated based on the old personnel planning figures on investment level 3" should be a thing of the past in the age of big data and automated processes. In addition to accuracy, new parameters for the quality of a planning system should be considered in light of crises and faster changes in the economic environment: robustness, resilience, and speed. In all these aspects, many planning systems, especially during the COVID-19 pandemic, have completely failed. The path to a robust forecasting system – one that delivers meaningful and informative results under all conditions at all times – is long, but important.
It is particularly important to check drivers for causal logic. Unforeseeable events that are not reflected in historical data can only be represented in planning models if causal relationships are correctly represented. Models are too often adjusted too precisely to current conditions and no longer work under major changes in circumstances. It is advisable to keep the number and complexity of planning parameters low. Unnecessary details increase planning effort and also increase the uncertainty in relationships. Therefore, the number of accounts and cost centers should be reduced.
In addition, a comparison and linkage to actual figures should always be maintained to continuously work on planning accuracy and conduct variance analyses. Modern technologies and automation should be used to create forecasts in hours instead of days. This leaves more time to engage with the results. To find a good starting point, one should first try to replicate the past with the rolling forecast model. Only when good results are achieved should the driver logic be transferred to the future.
Through consistent improvement of assumptions and driver logics, as well as fast and automated data processing, a forecasting system can become very robust, meaning it delivers meaningful results under all circumstances. The robustness can be measured by stress-testing the forecasting model. Individual parameters or combinations of parameters will deviate unusually strongly from the expected data, and the planning will be carried out accordingly. Experts can then assess whether the real company would behave the same way in an exceptional situation. In this way, the causal relationship between drivers and results is checked, and the company can learn a lot about its own interactions.