Abstract: The traditional BP neural network has some application problems. For example, the network structure parameter is too dependent on experience and easy to fall into local solution. In order to improve the application defects of BP neural network model, the optimization GA-BP algorithm to optimize BP neural network topology and the selection process of network initial parameter value is proposed. In order to verify the feasibility of the model, marketing customer historical data of a bank short-term financial products as the research object is used to validate the model which could more accurately predict the customer compared with BP neural network model. The test results show that the model could be applied to analysis financial product marketing data and more accurately predict the future marketing results.