Inquiry handling is widely used in most large companies to compensate for problems in business processes. However, it remains a manual, labor-intensive process and lack of effectiveness and accuracy. This thesis proposes a systematic framework based on data mining techniques to help automating inquiry handling and predicting potential risks according to inquiries.A target-oriented feature weighting model is applied to pre-process raw inquiry data, and the neural networks are constructed to cluster inquiries into patterns. Since inquiry handling results are also learned during clustering, a processing recommendation can be made when a new inquiry is classified. And a significant change in inquiry patterns, which implies potential risks in the transaction processing system, can be identified by deviation analysis.