AI-Driven Predictive Workload Management in Contact Centres: A Review of Forecasting, Staffing, Real-Time Optimization, and Human-AI Governance

Abstract

The contact center has grown from a “service queue” into an AI-driven Omni channel environment where the needs of the customers, the availability of your agents, service quality, automation, and real-time routing all dynamically interact. Predictive workload management is therefore critical as it is instrumental in managing the workload for a good balance of customer waiting time, service level compliance, operational cost, and agent utilization and employee wellbeing. This work is a literature review focused on key areas identified for AI-driven predictive workload management in a contact center—a few of which have particular relevance for this round—including workload forecasting, staffing optimization, queueing-based decision support, deep learning for time series prediction, automation channeling demand deflection, and responsible human-AI governance. The review reveals that classical queueing and workforce management modelling concepts continue to play a vital role in meeting demand forecasts, adapting staffing levels and service levels, but they are becoming increasingly inactive without the incorporation of adaptive forecasting, uncertainty considerations for forecasts and real-time feedback. In recent years, the application of machine learning, recurrent neural networks, attention-based models, and transformer networks has introduced enhanced capabilities to capture the complex and non-linear nature of workloads with multiple horizons and high frequency. But that's not enough - predictions have to be accurate. For AI-driven workload management to take place, the architecture must be integrated to enable data ingestion, forecast demand, optimize staffing, refocus reforecast in the middle of the day, control routing, monitor automation, and make sure that the automated results are explainable and fair. Some key research gaps highlighted by the review are: multichannel forecasting, concept drift, workload redistribution due to the use of a chatbot, agent well-being, privacy-preserving learning, and interpretable prescriptive analytics. It suggests that creating a new generation of workload management for contact centers must be a socio-technical system with the augmentation and enhancement of human workload operational judgment provided by AI.