Quickest detection in practice in presence of seasonality: An illustration with call center data

TL;DR

This paper explores the application of quickest detection algorithms, including CUSUM and machine learning methods, to identify imminent staffing needs in call centers affected by seasonal patterns, using real insurer data.

Contribution

It demonstrates how quickest detection techniques can be effectively applied in practical, seasonally influenced scenarios like call center staffing, with real data illustration.

Findings

CUSUM algorithm is relevant for detecting staffing needs.

Machine learning methods offer competitive detection performance.

Application to real insurer data validates practical usefulness.

Abstract

In this chapter, we explain how quickest detection algorithms can be useful for risk management in presence of seasonality. We investigate the problem of detecting fast enough cases when a call center will need extra staff in a near future with a high probability. We illustrate our findings on real data provided by a French insurer. We also discuss the relevance of the CUSUM algorithm and of some machine-learning type competitor for this applied problem.