Rehabilitation authorizations are a critical bottleneck in post-acute care transitions that impact patient outcomes, resource utilization, and operational efficiencies. In this study, we explore 1,787 rehabilitation authorization cases in three hospital campuses to identify performance patterns and build predictive models for the delay of authorization. We used multiple methods, such as exploratory data analysis, two-way ANOVA, Cox proportional hazards modeling, and artificial neural network (ANN) classification, to analyze authorization processes at 11 major insurance payors. Main process measures were Ready to Place (RTP) to decision time, length of stay, and authorization approval rates. The results showed a significant variation in authorization times was observed across payers (p<0.001) and campuses (p=0.0159). The developed ANN model achieved 96.1% accuracy in predicting authorization delays, with discharge planning coordination emerging as the dominant predictor (feature importance: 0.89). Cox modeling identified weekend submissions as the second-strongest predictor of delays (HR=0.55, p<0.001). Bottom-quartile payors were found to account for 22% of volume but contributed disproportionately to extended stays. Payors' approval rates are predictable based on payor, campus, and time of submission. By applying the confirmed ML prediction model, we can prospectively identify high-risk cases, allowing preemptive actions to be taken for bottom-quartile payors, and with weekend submission protocols, ML models offer early intervention. The consistency of results between different analytical methods is strong evidence for a systemic optimization.