top of page

The Practicality of Complexity in Call Center Data Analysis

Call centers are the battlegrounds where customer satisfaction is won or lost, often in minutes. A critical factor in managing these centers effectively is understanding why customers hang up before their call is answered. The Institute for Integrating Statistics in Decision Sciences laid out a sophisticated framework for this in their 2009 paper, "Bayesian Analysis of Abandonment in Call Center Operations" by Tevfik Aktekin and Refik Soyer. The Bayesian approach is robust, incorporating prior knowledge and offering a detailed analysis of customer wait times. However, in the fast-paced environment of call centers, when is this complexity more than is practical?

Bayesian models are thorough. They allow managers to see what is happening and weigh the likelihood of various outcomes based on past data. This thoroughness is a double-edged sword; the models require heavy computational resources and can be slower to update with real-time data. In contrast, more straightforward methods like logistic regression can quickly highlight critical factors affecting customer patience. Our use of logistic regression on simulated data showed us plainly that the chance of a call being abandoned increases as wait times increase.

For managers in the trenches of daily operations, the choice between a Bayesian model and a more straightforward one hinges on several factors. How quickly do decisions need to be made? What computing power is available? What level of detail is necessary for action? For instance, this graphic illustrates the relationship between wait times and customer abandonment—a snapshot that's easy to digest and can inform quick decisions:

Jupyter Notebooks and simple Python code offer an excellent compromise between detailed analysis and speed. They offer a platform where complex models can be run more efficiently, and updates can be made more agilely. They're perfect for situations where the power of a Bayesian model is needed, but the speed of logistic regression is also desired.

In sum, Bayesian methods give us a comprehensive view, but simpler models provide quick insights that are easily translated into action. It is crucial to find the right tool for the job. When resources or time are limited, a less complex model can be the key to responsive and effective call center management. As the analytics field evolves, so do the tools at our disposal. This combination of the right models and tools will drive the next generation of operational excellence in contact centers.

We can learn from papers such as Aktekin and Soyer's work, as they provide a foundational understanding of the Bayesian approach. However, there is merit in finding a balance between in-depth analysis and operational practicality for everyday use. By using tools like Jupyter Notebooks for rapid model testing and implementation, call center professionals can tap into advanced analytics without getting bogged down by the process. Ultimately, it's about choosing the path that leads to understanding and actionable insights that enhance customer service.

At WFM Labs, we discuss topics like these and apply them in a practical sense. A parallel concept we discuss in the community is forecast accuracy; while we all need a solid forecast, at what point is great accuracy not worth the time and effort? To learn more about these concepts, join the conversation at our community!

Citation: Aktekin, T., & Soyer, R. (2009). Bayesian Analysis of Abandonment in Call Center Operations. Technical Report TR-2009-3. Institute for Integrating Statistics in Decision Sciences, The George Washington University.



bottom of page