Estimating R&D cost at an early new product development stage using neural networks: a case study in the automotive industry
The development of shared architectures, platforms and construction kits enables OEMs to build several derivatives based on one of these concepts. Each new project should comply with the OEMs financial targets and therefore a profitability analysis has to be performed at an early new product development (NPD) stage. R&D costs are part of this analysis. Aim of this study is to transfer the neural network approach from product cost estimation to R&D cost estimation and test its applicability in terms of accuracy and processing time. Therefore this study proposes a model, which is tested with real world data from the brake development department at Porsche AG (PAG), gathered through a case study. The results show that R&D costs can be estimated at this point very accurate without having detailed knowledge on cause-effect relationships of cost drivers. Especially when using the Bayesian Regulation algorithm. A good input structure is a prerequisite. The processing time is relatively fast, compared to the actual cost estimation method, but leaves room for improvements. Concluding, a neural network approach is suited for early R&D cost estimation, if a good input structure can be retrieved and historical data for these inputs can be gathered.