Based on accounting informatization, this paper constructs a financial risk prediction system by applying the CNNs (Convolutional Neural Networks)- BiLSTM (Bi-directional Long Short-Term Memory)-Attention model to accurately identify and classify various risk types in enterprise FM (financial management), and improve the accuracy and efficiency of financial risk prediction. CNN was used to extract local features in financial data, BiLSTM was used to capture time dependencies, and finally the importance of financial indicators was weighted and fused through the Attention mechanism. During the training process, the Adam optimizer and cross entropy loss function are used for optimization, and appropriate learning rates and training rounds are set to ensure the stability and performance of the model. The experimental results show that when the epochs is 50, the accuracy of risk classification is 98.9% and the loss value is 0.012. In the analysis of each data level, the average response time of the proposed system and the traditional system is 1.80s and 7.17s respectively. The system in this paper shows obvious advantages in response time and prediction accuracy. The response time is greatly shortened, and it can provide effective support in real-time decision-making. This paper model has significant application prospects in financial risk prediction, and can provide enterprises with efficient and accurate risk warnings, which has important theoretical significance and practical value. Keywords: Enterprise Financial Management, Risk Classification, Financial Data, Accounting Informationization, Convolutional Neural Networks