Blood type substitution is an effective policy for improving the blood supply chain resilience in responding to shortages at hospitals. However, while compatibility of certain blood types provides an opportunity to better manage volatility in supply (donation) and demand (transfusion), in practice, it remains challenging to decide whether to use a compatible blood unit from the on-hand inventory (substitution) or to place an urgent or emergency order. Available evidence indicates that, in Australia, current practices for deciding on blood type substitution over emergency orders result in a significant imbalance in supply and demand for the most widely compatible units: O-negative blood units. This challenge is particularly critical in blood inventory management and represents a significant research opportunity for the operations management (OM) community. It can be addressed using state-of-the-art methodologies in data-driven decision making and perishable inventory under stochastic supply and demand. This study proposes a stochastic optimization model based on sample-average approximation (SAA) to aid the blood-type substitution decisions at hospitals. We take into account the historical demand data, the type and age of blood units in the inventory, as well as the shelf-life variability of the units arriving at hospitals. We compare the SAA results with several policies, including a single-period myopic model, no-substitution, over-order practice, and an empirical guideline. Results indicate that the proposed SAA model outperforms all the above policies. It thus creates significant opportunities for improvement in the management of blood units, as it reduces the imbalance in the supply and demand of O-negative blood units, allowing practitioners to benefit from substitution while avoiding the over-ordering of O-negative units.

Popularity bias reflects the positive impact of popularity information on consumer choices. There are two common display formats for popularity information on online retail platforms: the total-based cumulative sales format, where the total sales of a product are displayed since its launch, and the period-based cumulative sales format, where only sales within a specific recent period are shown. Both types of popularity information are continuously updated. This paper focuses on the multiproduct dynamic pricing problem with popularity bias. We employ the widely used multinomial logit model to investigate the impact of popularity bias on consumer choices. In particular, we examine how popularity bias affects marginal revenue, pricing decisions, and market shares. Moreover, we highlight that ignoring popularity bias can lead to a suboptimal outcome. As the multiproduct dynamic pricing problem suffers from the curse of dimensionality, we propose a semi-myopic pricing policy, which is computationally tractable, and demonstrate its asymptotic optimality under both formats. Our numerical simulations further indicate that ignoring popularity bias can result in substantial revenue losses, while the semi-myopic pricing policy consistently outperforms other heuristics under both formats. Finally, empirical tests on real data provide a comprehensive procedure for identifying the most appropriate choice models, which offer practical insights for implementation.

We study multi-period risk-averse inventory control in a data-driven setting. In this problem, a risk-averse retailer makes periodic decisions on inventory levels based only on historical demand observations without full knowledge of the demand distribution. We adopt the popular nested formulation for risk-averse programs to formulate this multi-period problem and its data-driven counterpart under a coherent risk measure. Our objective is to study the sample complexity bound such that with high probability, the data-driven policy is near-optimal, i.e., the relative error of risk under the data-driven policy compared with the optimal risk is arbitrarily small. Analysis of this problem is inherently challenging, because the multi-period nature requires solving the risk-averse program and its data-driven version recursively backward in time, while the (empirical) risk-to-go functions in this process do not have closed-form derivatives for most risk measures, which renders existing first-order methods for the risk-neutral newsvendor model invalid. In this study, we develop a zeroth-order framework to establish the complexity bound on sample sizes to guarantee nearoptimality of the data-driven policy with given accuracy levels. Instead of using first-order derivative information on the risk-to-go function, our analysis directly examines the class of functions that underpins each cumulative risk function and derives maximum inequalities for this functional class by computing the covering numbers. Finite-sample complexity bounds are then used to establish asymptotic properties of the estimated risk, including consistency and convergence rate. Computationally, the time complexity for solving the data-driven policy, which is essentially an empirical dynamic programming (EDP) estimator of the optimal policy, increases exponentially in the length of the planning horizon. To speed up computation, we propose an approximation scheme that recursively approximates the empirical cumulative risk function with a convex piecewise linear function and then minimize it to obtain a modified data-driven inventory policy. We show that with proper control for approximation error, the modified data-driven policy is also near-optimal, and it has the same order of sample complexity bound as that for the original EDP policy.

The literature has highlighted the value of the solution business model for manufacturing firms. In this research, we investigate the dark side of providing customer solutions and challenge an assumption held by extant research: That shifting from selling individual products to providing customer solutions enhances buyer–supplier relationships. Drawing on the frame problem view and expectation-(dis) confirmation theory, we examine the impact of provision of customer solutions on customer default. We also explore contingent factors that either magnify or mitigate this effect. Using a longitudinal dataset of 2,218 observations from 446 Chinese listed firms between 2012 and 2020, we find that provision of customer solutions increases the likelihood of customer default. Moreover, we show that this positive impact is stronger in industries with higher turbulence and for firms with greater advertising intensity, but weaker for firms with stronger operational capability. These findings suggest that provision of customer solutions can damage buyer–supplier relationships by reducing the likelihood of meeting customer expectations and increasing transaction disputes.

Artificial intelligence (AI) is poised to reshape operations across industries. Yet its real-world impact reveals a jagged and uneven implementation frontier. To make sense of this emerging landscape, we develop a foundational framework that synthesizes research and practice at the intersection of AI and Operations Management (OM), anchored in three interdependent pillars: (1) AI for OM, (2) OM for AI, and (3) Human–AI Interaction. First, AI for OM analyzes how AI enhances core operational processes, including design, procurement, production, and delivery. Second, OM for AI argues that scaling AI safely and effectively stands to benefit from core OM principles, including workflow design, capacity management, process control, drift detection, and continuous improvement, all of which are central to AI development and deployment. Third, Human–AI Interaction emphasizes the role of trust, incentives, and organizational design in mediating how humans and machines learn from and collaborate with each other. This triadic framework provides a foundation for organizing research on AI and OM and offers practical guidance for integrating AI into business and societal systems.

Telemedicine has transformed healthcare delivery, yet its impact on the quality and consistency of clinical decision-making remains underexplored in operations management. While heralded for efficiency and access, it is unclear whether virtual care systematically alters adherence to clinical practice guidelines. This study investigates whether the mode of care, telemedicine versus in-person, affects clinicians’ guideline-recommended prescribing of anticoagulants for atrial fibrillation, where treatment adherence is crucial to prevent severe outcomes. Using electronic medical record data from 16,603 patient encounters at a major academic health system in the United States (2020–2023), we employ a two-way fixed-effects linear probability model to compare prescribing patterns. We find that telemedicine visits are associated with a 5.2-percentage-point lower probability of guideline-recommended anticoagulant prescription than in-person visits. The gap is driven by clinicians with historically lower adherence; top-quartile clinicians show no difference across modalities, exhibiting behavioral robustness (adherence resilient to modality). We propose that telemedicine’s leaner information channel heightens perceived uncertainty, activating cognitive biases: For some clinicians, telemedicine exacerbates ambiguity aversion and omission bias , favoring inaction in virtual settings. Linking service design to variation in professional judgment, we show how operational context moderates decision quality. Practically, healthcare organizations should not treat telemedicine as a simple substitute for in-person care; they should implement targeted operational interventions, such as enhanced decision support and training on cognitive biases, to maintain high standards of care so efficiency and access gains do not come at the cost of adherence. In a supplementary analysis, we estimate that, at the population level in the United States, a 5.2-percentage-point telemedicine-related gap in anticoagulant prescribing could result in more than 150 preventable strokes and roughly $50 million in avoidable lifetime costs annually for each newly diagnosed cohort, underscoring its substantial clinical and economic significance.

Ensuring fair platform practices has become a critical challenge in Operations Management. Increasing anecdotal evidence suggests that digital platforms may leverage consumer data to engage in data-driven price discrimination (DDPD)—charging discriminatory prices to customers based on inferred willingness to pay. Yet, despite widespread concern, no platform has publicly acknowledged the operation of DDPD, and academic research has thus far lacked the means and empirical evidence to quantify its magnitude in practice. Against this backdrop, the first objective of this study therefore is to provide direct evidence on the presence of DDPD. To this end, we conducted a field experiment with a leading e-retailing platform in China. The results reveal that customers for whom the platform possesses more data are subject to a higher level of price discrimination. Recently, various regulations have been introduced to protect consumers from unfair use of their data, but there is a lack of evidence on platform compliance to these regulations. The second research objective is then to investigate whether and to what degree platforms complied with these regulations. We leverage a unique natural experiment in which the Chinese government implemented a new regulation banning DDPD in 2020. The findings reveal that while DDPD practice did not disappear entirely, the level decreased significantly post-regulation. As the first empirical study to present direct evidence of DDPD presence and platform compliance, the findings have significant implications for policymakers, platforms, and customers.

A developer's popularity plays a crucial role in their success within open source software (OSS) communities and their access to sponsorship opportunities. This study seeks to answer the question: which signals have the most predictive power for popularity and sponsorship volume on social coding platforms? Using algorithm-supported abductive theory generation supplemented by qualitative insights from observations and interviews, we arrive at a theory of peer evaluation in OSS communities. We examine a large number of signals and categorize them. The two categories are signaling via self-disclosure through profile signals and signaling via contribution quantity and quality through behavioral signals. The large amount of data available to us allows us to use machine learning techniques to arrive at top-ranking predictors within each category. We generate our theory by finding robust patterns and test our theory using a hold-out sample. Our findings indicate that easily observable credibility-enhancing and approachability-related developer profile signals hold greater predictive importance in shaping popularity. However, harder to observe and more complex behavioral signals show greater predictive importance for sponsorship volume. These results signify that OSS social coding platforms are not meritocratic, as developer self-disclosure significantly influences popularity. In contrast, sponsorship decisions, due to their high cost and irreversibility, depend on within-platform contribution-related signals. This research contributes to a deeper understanding of popularity and sponsorship within peer-to-peer followership networks in OSS communities. Through our research, platforms are better informed about the predictors of popularity and sponsorship and can introduce measures to enhance the meritocratic nature of these communities. Developers who seek influence and sponsorship on the platform can be more strategic about information disclosure and their contributions.

Pirated video clips, a new form of piracy, have spread rapidly in recent years with the vigorous development of short‐video social media platforms such as TikTok and Douyin . This has led to a huge loss of revenue for the film and TV drama industry where the negative economic impact trickles down from legitimate video platforms to content creators. As video platforms fight back against copyright infringement, they need to decide how much effort to invest in detecting and taking down pirated video clips that emerge on social media platforms. In addition, they need to decide whether a lump sum or unit price contract should be used to secure the playing rights for a video from the content creator. A content creator has no incentive to engage in piracy surveillance together with the video platform after signing a lump sum contract but has an incentive to do so after signing a unit price contract because he/she would receive royalties each time someone watches the copyrighted video. Using an analytical model framework, we find that the video platform should use a unit price contract when production of the video is less costly for the same quality (e.g., music‐based, horror, and romance movies) but a lump sum contract when production is costly (e.g., Sci‐fi, animation, and fantasy genres). We show that in equilibrium, the total surveillance effort between the video platform and the content creator is smaller and the video platform charges customers less to watch the copyrighted video under the lump sum contract compared to the unit price contract. As a result, the lump sum contract not only achieves higher consumer surplus compared to the unit price contract but also leads to a higher percentage of customers legally accessing the video program. Finally, the content creator's revenue is the same under both contracts, but the content creator may prefer the lump sum contract because it attracts more (legal or illegal) viewers and the increased visibility could be beneficial in the long run.

Online healthcare consultation services (OHCS) hold immense potential to transform the healthcare industry. This study seeks to investigate the impact of doctors' participation in OHCS on offline outpatient and inpatient costs, addressing a critical yet underexplored area in the healthcare IT literature. Utilizing an exhaustive dataset from a high-level comprehensive hospital in China over a period of 42 months, we find that doctors' OHCS participation increases outpatient costs by 63.2% (¥268.510) and inpatient costs by 22.2% (¥2,365.092). These results remain robust after addressing endogeneity concerns through instrumental variable (IV) approaches and propensity score matching (PSM). Notably, mechanism tests reveal that OHCS participation enhances doctors' service volume expansion, cost-upgrading within services and patient mix changes, leading to induced demand and higher offline medical costs. Moreover, these effects are amplified for doctors with higher professional status. However, the COVID-19 pandemic and higher department crowding levels tend to mitigate these cost increases. This study makes several significant theoretical contributions. First, it expands the healthcare IT literature by uncovering the "dark side" of OHCS participation, showing that it raises medical costs by altering supply-demand dynamics, contrary to the cost-saving potential often associated with traditional healthcare IT. Second, it advances research on OHCS by identifying the mechanisms—service volume expansion, cost-upgrading within services and patient mix changes—through which OHCS participation influences offline costs. Third, it explores the heterogeneous effects of physician, environmental, and departmental characteristics on medical costs. Practically, our findings suggest that hospitals and policymakers should promote widespread OHCS participation to balance supply-demand dynamics, advance treatment standardization and differentiated payment models, establish a multi-dimensional physician performance evaluation system, and leverage AI technology to improve doctor-patient matching and resource scheduling—enabling the realization of online healthcare's potential without inadvertently driving up medical costs. This study highlights the nuanced implications of OHCS on offline healthcare costs, offering insights for improving its implementation in healthcare systems.