Enhance Your Workforce Management Skills with Our Expert Training in Amsterdam This June!
Let’s Talk about Model Accuracy

Let’s Talk about Model Accuracy

We all know the importance of forecast accuracy. If the forecast is wrong, the outcome of the whole planning process is wrong. But the same holds for the other steps. If you calculate the required staffing incorrectly, no matter how accurate the forecast is, you will still schedule the wrong number of agents. Staffing usually consists of using the Erlang formula and then adding shrinkage. How good is this? As part of a PhD student and CCmath team member Siqiao Li’s research project, we dived into this subject with surprising results!

Challenges in Measuring Accuracy

But first, we had to tackle some practical difficulties. Forecast accuracy is easy to measure; you compare the forecast with the actual. But how about model accuracy? Your SL prediction will always be off because the forecast was wrong, or the realized was different than required because of unforeseen events that you couldn’t predict… which part of the “service level error” is due to the model? We found a way to differentiate the different reasons for having a wrong service level prediction and isolated the error from the model. Then, we compared different models and analyzed the impact of different model choices. In short, taking abandonments into account makes a big difference. Thus, Erlang C should be “abandoned”. Paid breaks are concentrated at certain moments of the day; thus, taking the resulting variability in shrinkage into account has a significant impact—finally, agent variability. Even under experienced agents, there is quite some variability in average handling time. Taking this into consideration dramatically improves the model accuracy.

Key Takeaways and Next Steps

The bottom line: taking these aspects into account will, together with improving the forecast accuracy, significantly reduce the need for real-time management and or reduce your outsourcing costs. All staffing methods and scheduling tools should take these aspects into account.

Learn More

Are you interested in learning more and seeing how CCmath can improve your SL accuracy? Contact us via our Contact page, or directly at ger@ccmath.com or siqiao@ccmath.com.

You can find the full paper on Arxiv: https://arxiv.org/abs/2402.19209

Leave a Reply