Predictive models are increasingly deployed within smart manufacturing for the control of industrial plants. With this arises, the need for long-term monitoring of model performance and adaptation of models if surrounding conditions change and the desired prediction accuracy is no longer met. The heterogeneous landscape of application scenarios, machine learning frameworks, hardware-restricted IIoT platforms, and the diversity of enterprise systems require flexible, yet stable and error resilient solutions that allow the automated adaptation of prediction models. Recommendations are provided for the application and management of predictive models in smart manufacturing. Typical causes for concept drift in real-world smart manufacturing applications are analysed, and essential steps in data and prediction model management are highlighted, to ensure reliability and efficiency in such applications. For this purpose, recommendations and a reference architecture for model management are provided. In addition, experimental results of two model adaptation strategies on an artificial dataset are shown.