Silent abandonment in service systems: Performance and patience estimation
Customer abandonment and willingness to wait for service are regarded as important measures of service systems. However, we find that they are often miscalculated in contact centers due to the phenomena of silent abandonment. Specifically, data that we have reveals that only a third of the abandonment is acknowledged by the organization as such. This may create a problem as the enterprise may not understand the source of customers’ complaints, since they believe that most of their clients got service while in reality they did not. In addition, such missing information biases estimation of customers’ willingness to wait (patience). This results in wrong staffing, concurrency and routing decisions, that impact efficiency and revenue.
We define a silent abandonment customer as a customer that leaves the system without indicating doing so. Therefore, the system is unaware of this happening until an agent tries to contact that person and gets no reply.
We discuss the phenomena of silent abandonment, and build classification models to estimate the real proportion of abandonment. The model uses textual features as well as meta-data of the conversation. We then build an Expectation Maximization (EM) algorithm to estimate customers patience. This algorithm is needed to face the challenge of uncertainty (as in the above classification) as well as data censoring. The algorithm accuracy is evaluated using simulation. In addition, a queueing model is used to show the importance of taking into account the phenomena of silent abandonment explicitly.