Papers & Projects
Capacity Planning for Resource Turnaround Operations
Under Review at Manufacturing & Service Operations Management (Thrid Round---After Major Revision)
Authors: Buyun Li , Vince Slaugh
Many shared resources, such as hotel rooms or rental cars, require cleaning, charging, or some other operation to turn around the resources between successive customer uses. We study staffing and shift planning decisions for the turnaround service capacity to minimize the sum of customer waiting and staffing costs. Random customer departures, random customer arrivals, and worker shifts with breaks add to the managerial challenge. Using the frameworks of diminishing returns, submodularity, and M-convexity, we demonstrate analytical properties for capacity decisions in three staffing scenarios, including our primary model that focuses on shift planning. We propose a solution heuristic that efficiently provides near-optimal solutions. We illustrate the value of our model for hotel housekeeping operations using data from a large city-center hotel. Reallocating some room attendants to different shift start times, especially later in the day compared to the current practice, can effectively eliminate guest waiting after the posted check-in time. Hotels can reduce room attendant idleness and room readiness issues by departing from the common industry practice of all workers starting at 8:00 am. Simply having two shift start times in the morning may virtually eliminate waiting and help in recruiting and retaining workers.
Dynamic Scheduling with Bayesian Updating of Customer Characteristics
Under Review at Manufacturing & Service Operations Management
Authors: Buyun Li , Xiaoshan Peng, Owen Wu
In many service industries, decision making about service scheduling often relies on assessing and prioritizing customer needs and value using professional judgment and customer data. Traditional scheduling models assume perfect knowledge of customer service rewards and delay costs, which is unrealistic. This paper considers the optimal scheduling problem in a multi-class queueing system where the system manager learns the reward of serving customers dynamically. We model the scheduling problem as a restless multiarmed bandit (RMAB) problem, with each customer class representing an arm characterized by queue length and the manager's belief about the reward distribution. We derive the Whittle index for each customer class. The resulting Whittle index scheduling policy which prioritizes the class of customers with the highest Whittle index. We prove that the Whittle index offers an optimal solution for a system with two customer classes-one with perfect information and one with unknown parameters-and show that it is near-optimal for more general settings numerically. Our results show that the incentive to serve a class of customers with unknown rewards increases with service rate, higher belief in rewards, arrival rate and length of wait, which contrasts with traditional models. This finding highlights that as queues grow longer, the priority for serving them increases due to extended busy periods. Furthermore, for a fixed product of service rate and reward, we find that customer classes with higher service rates provides higher incentives for learning. By understanding these dynamics, managers can better allocate resources, ensuring that longer queues, which imply greater potential delays and customer dissatisfaction, are addressed more promptly.
Infection Aware Nurse Staffing Using Random Graphs with Hidden Health Status
In Preparation for Manufacturing & Service Operations Management
Authors: Buyun Li , Jonathan Helm, Kurt Bertthauer
During outbreaks of infectious diseases, hospitals encounter significant challenges in making well-informed nurse staffing decisions. The dilemma during these outbreaks involves a simultaneous increase in inpatient admissions during disease outbreaks and a notably elevated rate of nurse absenteeism caused by infections. The unobservable nature of nurse infection time, incubation period, and number of nurses infected but yet to show symptoms adds complexity to understanding when and how nurses are infected. Lack of this critical information restricts hospital managers from implementing effective and informed operational strategies and staffing plans, limiting their ability to proactively address the staffing crisis during an outbreak. We develop a dynamic random graph model with hidden nurse health status to examine the interplay between staffing policies and infection transmission dynamics. Our model extends existing random graph frameworks by incorporating nurse health status (healthy, incubation, symptomatic) as a latent variable that is endogenously linked to the evolution of disease transmission networks. Within this framework, we design an estimation procedure that maps nurse characteristics to disease transmission rates across patient-to-nurse, nurse-to-nurse, and community-to-nurse interactions. This approach enables dynamic tracking of infection sources, locations, and timing. Using data from the IU-Health hospital system during the COVID-19 pandemic, we perform counterfactual analyses to assess the effectiveness of mitigation and staffing policies aimed at protecting nurses from infections. We find that hospitals can reduce nurse absenteeism due to infection by up to 25% through improved staffing levels and workload management. Furthermore, when establishing dedicated units for the care of infectious patients, simply isolating infected patients is insufficient; it is crucial to assign a fixed group of nurses exclusively to these patients to minimize cross-infection.
Early Intervention for Chronic Pain: Reducing Surgeries, Healthcare Utilization and Costs
In Preparation for Management Science
Authors: Buyun Li , Mohammad Zhalechian, Christopher Chen
Chronic pain, affecting approximately 25% of U.S. adults and costing $560–$635 billion annually, poses significant challenges to healthcare systems due to high utilization and fragmented care. This study investigates whether early referral to pain management specialists enhances treatment effectiveness and reduces unnecessary surgical interventions for chronic pain patients. Utilizing national insurance claim data from 140,000 patients and 35 million claims, we employ a difference-in-differences framework and a two-stage regression approach, leveraging a 2016–2022 zero-copay policy for physical therapy/chiropractic visits in the South Atlantic division as an external shock. Results indicate a 56.5% increase in the odds of early pain management intervention, leading to significant reductions in healthcare costs, outpatient visits, and a 12.0% decrease in unnecessary surgeries, alongside a 7.7% reduction in post-surgical visits and costs. Despite these benefits, the limited capacity of pain management specialists (3,500 nationwide versus 75 million patients) highlights access bottlenecks. Ongoing research explores optimal allocation strategies for pain specialists to maximize clinical and operational outcomes under capacity constraints.