Modeling call centers agent client interaction performance

This research proposes to model the performance of the interaction agent/client during conversations in call centers. The call centers represent strategic CRM tools for a majority of companies spending big budgets in their management without really making data mining to measure their agents' efficiency face to their clients. Contrary to previous researches based on questionnaires and declarative data, we use a rich dataset of conversations between agents and clients and measure interaction performance based only on the conversation content. These talks have been recorded and manually fully transcribed (we also worked on automatic transcribed data by speech recognition and compared the results). We decided to focus on the reality of the dyadic interactions "talks" without any interpretation or rationalization from none of those who take part of it. We first split the conversations in different phases (open, identification, treatment, proposal, and closure). We then coded the conversations in several quantitative variables (such as length, number of topics, number of talk rebound-linked to a commercial proposal done by the agent-, commercial proposal acceptance..) and qualitative variables based on sentiment analysis (using text mining) and emotions measure. Based on literature on recovery satisfaction, we defined three types of performance: procedural, commercial and interactional performance and finally measure a global performance of the conversation. Using structural equation models, this research provides new insights on the measure of the performance of the interaction agent/client in call centers, improving knowledge in the area of call centers data mining.