By Sergey Naumov📧, and Rogelio Oliva
In European Journal of Operational Research, 2025, 324 (3): 855–870. https://doi.org/10.1016/j.ejor.2025.02.010
Traditional queuing models mostly leave human judgment and decision making outside the scope of the system, ignoring their role as determinants of system performance. However, empirical evidence has shown that human behavior can substantially alter the system’s output. In this paper, we develop a hybrid approach that improves our understanding of the interplay between individual heterogeneous human agents and aggregate system behavior. We formulate human behavioral responses as feedback control processes, explicitly capturing the agent’s objectives and available information about the system’s state, accounting for delays and possible distortions. Our modeling approach taps into a behavioral modeling tradition that values realism and representativeness, making the formulations flexible and easily adaptable to specific situations. We illustrate our approach by considering a queuing system with delay announcement, commonly found in service and manufacturing settings. We find that the system continuously cycles between periods of low and high utilization, creating a suboptimal mode with predictable periods of high and low congestion and fewer customers served overall. By structuring the effect of behavioral responses as feedback loops, we formally analyze the observed system behavior and map it to behavioral decisions. The proposed modeling and analysis framework can guide system design and improve performance in scenarios where key dynamics are driven by both feedback structure and stochasticity. It provides generalizable structural explanations of the impact of human behavior in queuing systems.
Keywords: Agent-based simulation; System dynamics; Behavioral operations management; Queuing theory
Drug shortages have been persistent in the United States for over a decade, posing serious threats to public health and the healthcare system. While previous research has investigated the causes and effects of drug shortages, there is a dearth of research exploring potential solutions to mitigate this problem. Using a system dynamics model of the US generic drug market, we evaluate the long-term effectiveness of two existing policy interventions (expediting drug approvals and nudging manufacturers to ramp up their production) and the “quality reward” initiative that is being actively explored by the FDA and industry. Our results indicate that while the existing interventions can be helpful in addressing shortages, their long-term effect seems limited. In contrast, quality reward can mitigate drug shortages in a sustainable way. However, a caveat of quality reward is the potential emergence of a monopolistic supply market with negative consequences. We suggest that a carefully designed quality disclosure mechanism can address this issue. To the best of our knowledge, this is the first study to quantitatively and comparatively evaluate the long-term effectiveness of quality reward and other interventions on drug shortages and provide structural explanations for their performance.
E-commerce deliveries have grown significantly in the last decade, generating increased environmental impacts. While e-commerce grows, there is a need to integrate sustainability into its operations, especially concerning last-mile deliveries. Our goal is to understand the factors influencing consumer decision-making when selecting delivery alternatives. We examine consumers’ preferences using stated preference data collected in the United States. Respondents evaluated delivery alternatives, including operational and behavioral attributes. Interestingly, disclosing the environmental impact of each option does not influence consumer decisions. However, disclosing the delivery vehicle type does matter. The analyses reveal that consumers’ willingness to use sustainable options depends on their receiving additional benefits, varying across socio-economic profiles. Furthermore, elasticities and substitution rates confirm that consumers are willing to pay for convenient deliveries and are reluctant to make behavioral changes. These insights should be used to incentivize the use of sustainable alternatives for last-mile deliveries. We close the paper with recommendations to key stakeholders in urban areas.
Problem definition: We examine how to manage turnover among drivers delivering parcels for last-mile platforms. Although driver attrition in these platforms is both commonplace and costly, there is little understanding of the processes responsible for this phenomenon. Methodology/results: We collaborate with a platform to build a structural model to estimate the effects of key predictors of drivers’ decisions to leave or remain at the platform. For this estimation, we apply a dynamic discrete-choice framework in a two-step procedure that accounts for unobserved heterogeneity among drivers while circumventing the use of approximation or reduction methods commonly used to solve dynamic choice problems in the operations management domain. Drivers are compensated using a combination of regular payments that reward their productivity and subsidy payments that support them as they gain experience on the job. We find that regular pay has a greater effect on drivers’ retention. Furthermore, the marginal effects of both regular and subsidy pay diminish with drivers’ tenure at the platform, but the latter diminishes faster than the former. Additionally, we find significant heterogeneity among drivers in their unobserved nonpecuniary taste for the jobs at the platform and a significantly greater probability of retention among drivers with greater taste for these jobs. Managerial implications: Platforms can leverage our results to improve driver retention and design more profitable payment policies. We perform counterfactual analyses and develop a modeling framework to guide platforms toward this goal.
Today’s business environment is characterized by intensified competition, rapid changes, and shorter product lifecycle. These conditions push companies across industries to place a greater emphasis on building operational flexibility that can enable them to rapidly innovate new products, change product mixes, and adjust capacity. However, in the quest to build operation flexibility, many companies are disappointed and frustrated by the unfulfilled promise of flexibility. Our research reveals that this underperformance frequently occurs because companies direct considerable resources toward strategic planning with little in the way of execution, rendering them unprepared for complexities brought by a larger number of decision variables involved in managing flexible systems. In this white paper, we describe techniques developed by a team of researchers at Penn State Smeal College of Business in conjunction with Dow AgroSciences (DAS) to enable the benefits of operational flexibility. The paper discusses DAS operational flexibility strategies, the characteristics of operational flexibility paradoxes, and how such paradoxes can be effectively attenuated by an informed implementation approach.
Online retail platforms are increasingly challenged by the proliferation of low-quality products, which may damage their reputation and sales. To address this problem, we propose a system architecture to proactively identify products that are likely to go “out of favor”. Our approach uses historical data to extract useful information from customer ratings and textual reviews. Available data are fed into a state-of-the-art deep learning sequence model to forecast future ratings. We then analyze rating trends, extracting hyperparameters that a binary classifier uses to label products as “out of favor” or not. We tested this system on an Amazon dataset comprising nearly 800,000 observations across 2,826 electronics products. Our results show that the Long Short-Term Memory (LSTM) model excels in forecasting future product ratings compared to other benchmarks. Ablation analysis shows sentiment-related features significantly improve rating forecasts by up to 40%, with review topics adding 10% and other review characteristics 4%. Counterintuitively, topic extraction from reviews does not provide substantial benefits, despite the heavy computational resources it requires. Finally, the two-stage classification process, which leverages time-series data and rating trends, offers a more stable and robust performance than conventional single-stage methods. We provide considerations for system architecture development through robustness checks ensuring its resilience to stressors. Our experiments indicate that rating trends can change in subtle ways over time, leading a promising “star” product to turn into a liability (“dog”). E-commerce platforms can use the proposed system architecture proactively to identify and remove potentially dubious products instead of waiting to take reactive action.