At present, enterprise management effectiveness evaluation mostly relies on traditional statistical methods or expert experience, which are highly subjective, lack dynamic adaptability, and have difficulty in dealing with high-dimensional nonlinear relationships. Therefore, this study proposes an enterprise management effectiveness evaluation model based on deep learning, aiming to achieve objective and accurate effectiveness quantification through data-driven methods. Methodologically, a multi-source feature system including financial indicators (ROI, debt-to-asset ratio), operational data (inventory turnover, customer satisfaction) and environmental factors (market fluctuations, policy changes) is first constructed, and the sliding window method is used to align time series data; then a hybrid neural network architecture is designed, combining 1D-CNN to extract local spatiotemporal features, BiLSTM to capture long-term dependencies, and introducing the Attention mechanism to dynamically weight key indicators, and finally output the effectiveness score through the fully connected layer. The experiment uses 12,800 sample data of A-share listed companies over a five-year period, with MSE and R2 as evaluation indicators. The results show that the model test set has a minimum MSE of 0.146 and R2 of 0.923, and could effectively identify performance mutation points. The conclusion shows that the model significantly improves the evaluation accuracy and robustness, and provides a quantifiable analysis tool for intelligent management decision-making.