The carbon-in-pulp (CIP) gold extraction process is widely used in the gold beneficiation field. However, due to the inherent characteristies of the process, there is a lag in metrics control. At present, metries adjustment is typically carried out through post-analysis of issues, which significantly increases the uncertainty in process control. To effectively address this issue, a three-layer BP neural network model was constructed, with the number of neurons in the input, hidden, and output layers set to 5, 8, and 1, respectively, for predicting the tailings grade in the CIP process. The network was trained using actual production data from the Dongliang Gold Mine, and the results showed good network fitting performance. The trained network was then used to simulate and verify 30 sets of on-site production data, achieving a simulation accuracy of 85.52 %, effectively resolving the issue of lagging tailings grade in the CIP process. In addition, pro-cess optimization was performed using the BP neural network, and the optimal process parameters were identified. After optimization, the tailings grade was reduced to 0.08 g/t, significantly improving the utilization rate of gold resources.