In this Chapter, we will discuss two further case studies, namely, problems that were proposed by a French bank. Our team was involved by the request of a French university, and we started to collaborate with their respective team, experts in management topics. The bank put up several models, but unfortunately the real-world equivalent of most models, the modeled system, remained unknown to us because of trade secrets. This made it difficult or impossible to determine the correctness of the models and fine-tune their settings in order to make their behavior more similar to reality. There was only one exception regarding the management system of the bank. The concepts of it were identified by themselves and the French specialist team, and experts have estimated the mutual influence strengths and directions of these components on each other. Models created this way unavoidably contain more or less subjective elements, but even if the models had been produced by machine learning, their thorough investigation is recommended before usage. Sometimes a subtle modification of connection weights or a different λ parameter in the threshold function can result in a model that behaves significantly different. This chapter describes a method to automatically investigate a model and detect if some of its components affect its dynamic behavior significantly. Two models were selected from those developed by the bank to illustrate the method and the results obtained with it.