Accurately predicting the unit costs of products enables an enterprise to estimate the future profits on which conducting financial or production planning can be based. However, performing such predictions is difficult because of the uncertainty of the unit cost learning process. A few fuzzy collaborative forecasting methods have been proposed in the past to address this difficulty. However, these methods exhibit some limitations such as a time-consuming collaboration process, the existence of low-quality experts, and insufficient global optimality of the solution. To resolve these problems, this study designed an improved fuzzy collaborative forecasting system. The proposed system exhibits the following innovative features. First, software agents substitute real experts to automate the collaboration process for enhancing efficiency. A quadratic programming approximation approach is then applied to improve the global optimality of the solution. The proposed methodology was tested with a real case from a dynamic random access memory factory. The experimental results showed that the proposed methodology can improve the forecasting accuracy and precision efficiently by systematically adding agents to the fuzzy collaborative forecasting system. In addition, the effectiveness of the proposed methodology is also robust to the existence of a low-quality expert.