High logistic performance and low costs in order fulfillment are crucial for competitiveness in manufacturing. The increasing accessibility of production data enhances the potential for machine learning (ML) in the continuous improvement of production planning and control. However, the success of ML applications is highly dependent on the available domain knowledge due to complex cause-effect relationships in production logistics. This paper illustrates the importance of transparent domain knowledge as a basis for successful ML applications using the Cross-Industry Process for Data Mining. Transparency about domain knowledge in production logistics is created through model-based derivation of simple logistic principles (logistic Do’s and Don’ts). Their contribution to the success of ML applications is proven with various use cases.