Predictive Controlling: Automated Projection in the Group

Why automation and what is predictive controlling?

The advantages of automation are obvious: Thanks to the newly gained information density, planning processes can be optimized and made more efficient. The massive reduction in manual activities frees up capacities for new tasks, while the mathematical susceptibility to error approaches zero. Thanks to the automation of a monthly projection, it is also possible for controlling to create different scenarios in a timely manner, even in times such as the Covid 19 or supply chain crisis. These allow management to simulate uncertainties and react proactively to possible changes.

The need for coordination between parent and subsidiary companies (including high manual and time-intensive effort) can be significantly minimized: This is possible by using artificial intelligence (AI) to largely automate monthly forecasts at group level.

See below for brief insights into the content of the whitepaper:

1. hierarchical time series

Like balance sheets, income statement data has a hierarchical structure: up to eleven hierarchy levels are not uncommon. The simplified income statement hierarchy in Figure 1 combines a total of seven time series in three hierarchy levels. These are not independent of each other: EBT is, of course, the sum of EBIT and the financial result. Theoretically, all seven time series could be forecast separately. However, this approach would lead to the problem that, for example, the sum of the forecasts of EBITDA and depreciation and amortization would generally not correspond to the EBIT forecast - which, however, must necessarily be fulfilled by definition of the income statement structure.

This problem can be solved using so-called matching methods: Essentially, this involves the question of ...

1.1 Adjustment methods

1.2 Data quality and data cleansing

2. algorithms

The most common theoretical forecasting models are presented below. The choice of the best model(s) is also discussed. Generally speaking, the goal of each model is to produce a forecast. Thereby h stands for the forecast horizon (in the present project 2-13 months) and T for the last observed forecast ...

2.1 Autoregressive Integrated Moving-Average (ARIMA)

2.2 Exponential Smoothing (ETS)

2.3 Stabilization

2.4 More advanced algorithms

2.5 Selection of the method(s)

2.6 Forecast of individual time series - base forecast

2.7 Temporal adjustment

2.8 Hierarchical alignment