Enterprise intelligent audit process optimization and automatic decision-making method based on reinforcement learning

Traditional auditing methods suffer from delayed responses, low resource utilization, and high misjudgment risks in dynamic business environments. This paper proposes an enterprise intelligent auditing process optimization and automated decision-making method integrating Reinforcement Learning (RL). The approach models auditing decisions as a Markov Decision Process (MDP), constructs a multi-objective reward function encompassing risk, cost, and efficiency, and designs a policy learning algorithm based on Deep Q-Network (DQN) to achieve dynamic allocation of auditing resources and path optimization. The system framework comprises a data perception layer, RL decision core layer, and application interaction layer, supporting real-time feedback and expert intervention to form a closed-loop mechanism of “continuous monitoring and self-evolution.” Experiments using real enterprise supply chain data demonstrate that this method significantly outperforms rule-based systems and supervised learning models in terms of precision (52.7%) and F1-Score (63.0%), while reducing audit resource consumption to 5.1%. This validates its advantages in balancing accuracy, efficiency, and cost. The research provides a practical technical pathway for the intelligent transformation of auditing, empowering enterprises to build proactive, efficient, and adaptive risk control systems.