Prediction of Investment Costs

Challenge

Estimating and planning investment costs is a crucial factor in controlling. For our client, an international automotive corporation, ensuring the smooth operation of controlling processes is essential but time-consuming. This includes calculating the annual expected expenditures based on the total budgets of individual investment projects. To allocate resources optimally, both the amount and timing of expenditures are critical.

Approach

To create the data foundation for modeling, historical investment expenditures for various projects were consolidated and enriched with additional information, such as the production site or product class of the projects. To reduce the complexity of the forecasting problem, both the time and cost dimensions were standardized. After intensive benchmarking of various classes of machine learning models, the forecast accuracy of a boosting model was optimized through backtesting for a time horizon of up to ten years. Based on the latest information on total budget and project characteristics, the model now effortlessly generates the cost plan on an annual level, thereby data-drivenly supporting the client's previously lengthy manual planning process.

Result

Forecasting the distribution of the project budget over the next 10 years enables our client to achieve significant time and cost savings. The data-driven automation of forecasting allows budget adjustments to be seamlessly incorporated into the annual planning.