This paper reports the preliminary results of analyzing recorded voice conversations between customers and human agents in a help-desk organization. The key to translating such raw records of experiences into reusable knowledge is finding and assigning appropriate indices for them. We have applied conversational case-based reasoning techniques to them and found that script-based indexing effectively expresses various customer situations and problems. In several conversation records, a new type of tacit knowledge played an important role when solving customers’ problems. It is called cue questions given as advice by senior agents to junior agents in trouble answering customer inquiries. We have experimentally implemented this idea as an extension of script-based indexing.