The focus of this research is on supporting a resource-efficient economy to minimize post-consumer plastic waste by incorporating secondary materials into existing product design and thereby closing material cycles in plastic usage. In addition to lower primary resource demand and lower energy consumption in the production phase, the potential opportunities include lower overall environmental impacts. An important challenge arises from batch-dependent variations in the quality of recycled plastics, influencing product development in the design phase. Finite element analysis (FEA) plays a crucial role in the early stages of product development, demonstrating a concept’s potential for further development at a low cost and with minimal production effort. Detailed modeling of material properties within FEA is essential to ensure the utmost validity of calculation results and to pinpoint any weaknesses in the early design stages. The present paper discusses various approaches and applications for using neural networks (NNs) in constitutive modeling, particularly when dealing with heterogeneous material behavior resulting from the recycling process. The use of virtual training data derived from a phenomenological constitutive model and its advantages and potential applications to recyclates are highlighted. This leads to the proposal of using inverse surrogates of the phenomenological constitutive model as one method for obtaining suitable constitutive laws from experimental data in the future.