In today`s economy, call centers play a very important role, serving as the main customer contact channel in many different enterprises. Call centers are highly labor intensive. Typically, 60%-70% of the overall operating expenses of call centers are derived from agents’ employment costs. Reducing the number of agents handling calls, without degrading service level, is thus of interest and importance. Enabling customers self-service is one of the basic means for doing so, with Interactive Voice Response (IVR) systems being one of the main self-service channels.
The goal of our research is to improve and enhance IVR systems, as a special case of self-service systems. To do so, we model and analyze customers flow within an IVR system. The model features were established and inspired by an Exploratory Data Analysis (EDA) of real IVR transactions in a call center of a large Israeli bank, based on more than one year of data, including millions of calls.
An IVR system usually offers several services. Customers enter the IVR and then follow a series of menus in order to reach a desired service or services. We represent the IVR system as a rooted tree and model customers flow within it as a stochastic search. We model the search of group of customers, which are characterized by their perceived rewards and costs, their success probabilities, their service time distribution, and their patience distribution. The goal of our search is to find the optimal path on the IVR tree, which will result in maximal expected discounted revenue for customers within each group. We show that, at each stage, an index can be assigned to each feasible option, and the optimal policy is to choose the option with the highest index at each stage.
One of the main observations derived from the EDA was that some customers leave the IVR system without getting any relevant information. These customers may either leave the call center or join the agents queue (opt-out) to receive the desired service. In both cases, we say that these customers abandon the IVR service. When customers are self-served, finding whether their service was successful or not is not an easy task. The subject of identifying abandonments from self-service systems, such as IVR, is thus of interest and we are addressing this issue in our work. We also discovered that there is a learning process, which means that as customers gain more experience within the system, their response time is getting shorter. This fact was incorporated into our model.
Our model enables the comparison between alternative IVR designs, both from the customer point of view and from the enterprise point of view, thus supplementing existing research from other fields such as Human-Factor-Engineering and Telecommunication Engineering. Although this research focuses on IVR systems, we believe that both the theoretical model and some of the methods presented in our EDA can be easily implemented to other self-service systems, which becomes more and more relevant these days, such as Internet websites.