Reinforcement Learning for Automated Workforce Scheduling in Retail Environments

Abstract

Optimizing scheduling remains a persistent problem in any retail setting due to demand surges, employee availability,
shifts of understaffing/overstaffing, and policies on labor usage. This research looks at automating workforce
scheduling using reinforcement learning (RL), which is capable of adapting to and performing better in changing retail
environments. The RL-based model proposed in this work treats scheduling as a multidimensional, sequential
decision-making process, permitting an agent to determine the best shift allocations with historical data, employee
requirements, and operational objectives. Employing a Q-learning algorithm in a simulation-based environment, we
assess schedule efficiency against labor costs, service levels, and employee satisfaction. Performance comparisons
against heuristic and rule-based scheduling techniques illustrate the marked cost-efficient flexibility of the RL
approach. In addition, performance evaluation shows improved schedule robustness during demand extremes. This
work demonstrates the remarkable unexploited capabilities reinforcement learning can offer to industrial-grade
workforce management systems in the context of guard-gated and on-demand scalable solutions for increasingly
complicated retail operations. RL's ability to act on real-time information and learn over time means that proactive
workforce planning can occur autonomously, to the detriment of managerial burden and the benefit of staff
satisfaction. Lastly, they set forth research opportunities focused on combining RL with real-time analytics, and multiagent
systems for wider application in enterprise resource planning systems.