Abstract
Digital enterprises increasingly depend on agile IT service management to sustain operational continuity, yet incident resolution processes remain vulnerable to inefficiencies that escalate economic costs and undermine strategic governance. This study proposes an integrated Predictive Governance Framework that synergizes Long Short-Term Memory (LSTM) networks with economic optimization objectives to enhance decision-making in IT incident administration. Utilizing the enriched event log from the ServiceNow™ platform (UCI Repository #498; 141,712 events, 24,918 incidents), we engineer a temporally aware deep learning architecture capable of forecasting resolution trajectories and SLA compliance risks with exceptional precision. Through rigorous temporal validation and cost-sensitive learning, our framework achieves 98.2% classification accuracy in predicting SLA breach outcomes surpassing conventional machine learning baselines by a statistically significant margin (p < 0.001). Beyond predictive performance, the model embeds interpretable attention mechanisms and counterfactual simulation capabilities, enabling administrators to evaluate intervention strategies under explicit economic constraints. Empirical results indicate that adopting this framework could reduce expected operational expenditures by 21.4% while maintaining service-level commitments. This research advances the scholarly convergence of artificial intelligence, process-aware information systems, and managerial economics, offering a scalable, transparent, and economically rational approach to digital enterprise governance.