Optimizing Restaurant Customer Flow and Revenue with Real-Time Coupon Allocation: A Deep Reinforcement Learning Approach

To adjust customer numbers in restaurants in real-time, thus reducing customer churn and improving total revenue, this study diverges from traditional methods such as queuing theory and service capacity adjustments. Instead, we propose a novel approach that utilizes coupons of varying amounts to influence customer dining choices between dine-in and packing. First, we developed a lightweight deep learning model to predict customer flow, followed by the employment of deep reinforcement learning (DRL) to dynamically allocate coupons based on real-time restaurant conditions. An entropy-based mechanism was also integrated to enhance the diversity of coupon allocation, ensuring a strategic balance between exploration and exploitation. We conducted off-line experiments using data from a large chain of fast-food restaurants, which demonstrated that our model can reduce the daily losses of the restaurant by 515 RMB. This study also has implications for the dynamic adjustment of offline store personnel and sequential coupon distribution.