A Closed-Loop Financial Decision Support System With Explainable AI for Manufacturing

Traditional financial forecasting systems often operate as black boxes, lacking the transparency required for regulatory compliance and strategic decisions. To address this, the author proposes a closed-loop decision support system that integrates accurate forecasting with built-in explainability for manufacturing finance. The end-to-end framework, centered on an enhanced temporal fusion transformer plus model, automates data ingestion, feature distillation, predictive inference, and adaptive feedback. The system provides visual explanations through attention heatmaps, enabling managers to understand model decisions, and incorporates a self-correcting mechanism that triggers model hot-swapping upon performance drift. Empirical results from 120 A-share manufacturing firms show the system achieves a mean absolute percentage error of 12.1% and reduces budget review time by 39%. This study demonstrates a template that transforms AI from a passive forecaster into an interpretable and adaptive partner for financial management.