A scheme for building a decision model of short-term liquidity analysis from domain experts is presented, which combines the features of both process tracing approach and output analysis approach. The scheme consists of process tracing, output analysis, and model review component. The process tracing component applies the Concurrent Verbal Protocol Analysis to build an initial decision model by tracing through decision procedures from domain experts. The output analysis component applies Probability Neural Network to build a decision model based on the predictions of the initial model. The model review component investigates the cases where the predictions of the two models differ, and feeds the findings back to the process tracing component for further improvement. This scheme retains the explanation capability of the Protocol Analysis, and, at the same time, provides an opportunity for researchers to rectify some of the inherent problems associated with it.