Production networks which require quick internal automation and flexibility because of constantly changing workloads or participation of multiple jurisdictions and business areas require a reasoned system of solving decisions problems. Recent progress in machine learning and optimization provide effective domain-specific problem solving heuristics however, domain indifferent industrial environments warrant their use. This paper presents a new approach for assessing and enabling the transfer of learned decision heuristics across domains in heterogeneous production networks. The proposed system combines domain embedding, meta-learning, and adaptive reinforcement tuning to enable the automated application of heuristic decision policies developed for one production environment to different domains without full retraining. Simulations in different industrial scenarios that included both discrete assembly and continuous processes show that the approach can achieve up to 88% of the target performance in the adapted domains and provides over 40% decrease in adaptation effort. Results showed that drift aware re-tuning improves convergence time and stability in the presence of high variability. These results confirm the possibility of cross-domain heuristic transfer and mark an advancement in the development of models for universal production intelligence of Industry 5.0.