Knowledge that Transforms

Fake news negatively impacts business operations across various sectors, including healthcare, retail, and finance. One promising way to mitigate this problem is through automated content moderation using machine learning (ML). However, existing methodologies have struggled to identify fake news due to its inherent challenges—namely, class heterogeneity, concept drift, context dependency, and adversarial attacks. To address these challenges, we propose a novel strategy that focuses on the heterogeneous intentions behind fake news. Our approach builds on insights from prior social science research indicating that the intent behind creating fake news may differ from that behind disseminating it. This intent-driven variation influences the syntactic structures of fake news, leading to its heterogeneous nature. Accordingly, we argue that developing an ML model focused on the intent dimension of fake news can help overcome the issue of class heterogeneity, which often prevents effective model convergence. Moreover, since intent-driven linguistic patterns are generally less susceptible to temporal shifts and manipulative alterations, we expect this focus to enhance robustness against the other key challenges—concept drift, context dependency, and adversarial attacks. Grounded in a design science approach, we therefore propose a novel ML framework for fake news detection that centers on identifying and leveraging the intent behind fake news. Specifically, we first develop a domain adaptation model to infer the intent behind fake news, specifically targeting contexts where sufficient training data is unavailable—a common limitation in intent detection. Then, based on the intent detection model, we redefine fake news detection by shifting from binary (fake vs. true) to ternary classification (deceptive fake news vs. non-deceptive fake news vs. true news), thereby reducing the computational burden on ML models tasked with processing heterogeneous information within the fake news category. When evaluated against state-of-the-art baselines, our approach demonstrated significant improvements in both intent detection and fake news classification. Robustness checks further demonstrate that our method is less sensitive to distributional shifts in data and is more computationally efficient than baselines. Moreover, post-hoc analyses reveal the underlying mechanisms through which incorporating intent into fake news detection enhances model performance, addressing the key challenges associated with fake news.

Resource pooling is often introduced in service systems to cope with the variability in customer demand. The primary motivation behind creating a more flexible system is to utilize resources efficiently—namely, assigning customers whose dedicated resources are fully occupied to available non-dedicated resources (referred to as off-service placement). In such a setup, the service system manager expects to serve more customers within a fixed timeframe. However, recent empirical evidence shows that, due to limited server capability, the service time at non-dedicated providers can be significantly longer than that at dedicated ones. In this study, we develop a two-server stochastic model to examine how server capability levels and other factors—such as overall workload and demand asymmetry—affect pooling configurations. We derive conditions that specify the optimal system flexibility configuration across these parameters. Our findings reveal that a partially flexible system can outperform a fully flexible one, particularly in asymmetric scenarios with low server capability. This advantage is also pronounced when considering a range of system costs, including server capability, cross-serving, and other related costs in stochastic service systems with and without buffers. These insights from our two-server model offer guidance on designing more efficient flexibility in complex multi-class, multi-server service systems.

Sponsored-products (SP) advertising is a popular way to promote products on Amazon. Etailers who have a large catalog of products often create SP ad groups for products with similar attributes. An SP ad group consists of a set of products that share a same keyword set used for product search. In addition to SP ads, etailers may link to external websites for advertising their products, which is called off-Amazon (OA) ads. This study focuses on the optimization of sequential SP and OA (abbreviated as SSPOA) ads decisions for etailers. We model the SSPOA optimization as a controlled Markovian multi-armed bandit (MAB) process. When the mean sales volume per unit time (i.e., sales rate) for each product is known, we characterize the etailer’s optimal SSPOA policy for products in an ad group. When the parameters of the sales rates are unknown, we develop a Thompson-sampling-based algorithm that couples the SP and OA ads decisions. We prove that the regret bound of the proposed algorithm is <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mrow> <mml:mover> <mml:mi>O</mml:mi> <mml:mo stretchy="false">~</mml:mo> </mml:mover> </mml:mrow> <mml:mo stretchy="false">(</mml:mo> <mml:msqrt> <mml:mi>T</mml:mi> </mml:msqrt> <mml:mo stretchy="false">)</mml:mo> </mml:math> , where <mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML" display="inline" overflow="scroll"> <mml:mi>T</mml:mi> </mml:math> is the total horizon length. Compared with existing literature, our problem additionally considers the regret from applying the estimated control policy and the impacts of choosing non-optimal keyword sets on subsequent states. We also conduct numerical experiments that validate our theoretical results. Moreover, we extend the base model in several directions, that is, considering unknown transition rates between different sales rate levels, incorporating correlated keyword sets, and learning the optimal policy using Posterior Sampling for reinforcement learning under a discretized setting.

Taxi drivers face intensified competition in the private transport industry following the entry of independent contractor drivers on ride-hailing platforms. To secure demand, taxi drivers increasingly join platforms, yet differ from ride-hailing drivers in that they can serve both platform orders and street orders (e.g., roadside hailing or taxi stands) and switch between them. The availability of street orders as an outside option creates a decision problem for taxi drivers and compels platforms to carefully design incentives-such as commissions and subsidies-to attract participation. A key distinction is that while ride-hailing platforms typically do not disclose trip destinations before pick-up to prevent cherry-picking behaviors, many platforms provide trip information to taxi drivers to compete against street-order opportunities. Consequently, taxi drivers need to choose between platform orders with known trip information but positive pick-up time and street orders without advance information but zero pick-up time. This article studies how taxi drivers optimally choose between platform and street orders and how platforms should design incentive schemes to maximize profit. To this end, we develop a game-theoretical model and characterize equilibrium strategies. We show that allowing taxi drivers to serve both platform and street orders is optimal for the platform, whereas enforcing exclusivity limits platform profit because street orders exist as an outside option and protect drivers' surplus. Furthermore, disclosing trip distance reduces the platform's subsidy burden relative to nondisclosure. These results remain robust across alternative settings, including homogeneous orders, zero commissions and subsidies, distance-based subsidy schemes, and heterogeneous pick-up times.

Most retailers offer trade-in programs that allow customers not only to trade-in a used product in part payment for the purchase of a new one, but also to return the new product purchased via a trade-in. In this context, we study how a jewelry trade-in program affects sales and returns of trade-in eligible versus ineligible new products, and how such programs can be better designed to improve profitability. We conduct an empirical analysis using data from a national jewelry retailer that increased the number of stores offering the trade-in program. Leveraging this expansion, we show that the trade-in program impacts sales and return rates, and find that this impact substantially differs across trade-in eligible versus ineligible products. For trade-in eligible products, sales increase by 11.4%, and so do return rates—by 4.3 percentage points. For trade-in ineligible products, sales increase by 2.7% and there is no impact on return rates. The provision of a trade-in option reveals a novel trade-off for retailers: It leads to higher sales, but also results in greater returns, which can be detrimental to profitability. Retailers therefore need to carefully assess this trade-off to manage trade-in eligibility of different products. A counterfactual analysis suggests that a selective trade-in program accounting for this trade-off can enhance program profitability by 19%.

Blockchain technology holds promise for improving access to financing within supply chains, especially for small and under-financed suppliers. Yet, the specific ways in which blockchain technology mitigates financing frictions and the features that contribute to successful platform implementations remain unclear. Following a theory elaboration approach, we close that gap by studying which financing frictions the blockchain-enabled supply chain financing (BCF) solutions aim to address, which blockchain features they use, and the association between the frictions and features. In our analysis, we examine 312 documents with unstructured text describing 11 BCF solutions, both successful and failed. Using AI-based large language models, we identify patterns connecting seven types of financing frictions and three key blockchain features. We find that the transactional friction is the most prominent, despite receiving limited attention in the academic literature, while bankruptcy costs and taxes—frictions that are commonly discussed in the literature—are rarely associated with blockchain features in our sample. Among blockchain features, tokenization is used sparingly; and, unlike other features, it appears in successful BCF solutions only. Moreover, connections between frictions and features are not random: while transactional and hidden actions frictions are linked to all three blockchain features, other frictions are typically associated with only one. Our findings offer a deeper understanding of the mechanisms through which blockchain adds value in supply chain finance, suggesting that aligning blockchain features with specific frictions may enhance the chances of success. We also demonstrate a method for using AI to evaluate large amounts of unstructured data in operations management research.

Government environmental agencies are increasingly partnering with large, foreign corporate buyers in the energy efficiency improvement process to encourage investments at small manufacturers that can enhance the value of local manufacturing. Using a stylized supply chain model, we show that while buyer collaboration enhances energy efficiency investments, its impact on the value of local manufacturing varies. Specifically, buyer collaboration improves the value of local manufacturing only when both the cost of the buyer’s outside sourcing option and the environmental impact of energy are either high or low. In cases where one factor is high and the other is low, buyer collaboration reduces the value of local manufacturing. Moreover, we find that the higher penetration of renewable sources in the energy mix can either amplify or mitigate the beneficial impact of buyer collaboration. When the manufacturer is more competitive in the global market, a greener energy mix can negate the beneficial impacts of buyer collaboration on the value of local manufacturing. Our study provides valuable insights for designing environmental initiatives and policies to boost both energy efficiency and local manufacturing, underscoring the importance of supply chain interactions in synchronizing climate change policies.

In digital marketplaces, product ratings play a critical role in shaping consumer demand and seller success, yet their relationship with logistical factors remains unexplored. This study investigates how deviations from promised delivery dates, whether early or delayed, influence product ratings. Using transaction-level data from a major e-commerce platform encompassing over 11M product purchases and 500K customers, we estimate the causal effect of deviating from the promised delivery time on both rating incidence and valence. To overcome the challenges of matching imposed by an unbalanced treatment and control distribution, we used a machine learning-based estimator, R-learner, to estimate the effects of interest. We find that both delayed and early deliveries increase customers’ likelihood of leaving a rating but reduce rating valence. Delayed deliveries lower rating valence by an average of 0.4, with some categories seeing a drop of up to 0.6. Early deliveries also reduce rating valence, with an average decrease of 0.2 and reductions of up to 0.5 depending on the product category. The negative effect of early delivery is consistent across all categories, except for food and beverages, where experienced customers show a small positive response. This is due to reduced uncertainty from reordering previously purchased products, a pattern that does not occur as often in other categories. We contribute to the service and logistics performance literature by establishing the causal link between delivery performance and product ratings by quantifying the direction and magnitude of these effects across various product categories and customer segments. We propose managerial implications for mitigating customer dissatisfaction arising from both early and delayed deliveries.

Firms are increasingly confronting a fundamental organizational choice: whether to retain human control over operational and marketing decisions or to delegate decision authority to agentic artificial intelligence systems. While recent advances in generative and autonomous AI enable real-time pricing, inventory allocation, and demand coordination, firms exhibit substantial heterogeneity in how much autonomy they grant these systems—ranging from full automation to extensive human oversight. This raises a central operations management question: when should firms delegate pricing and inventory decisions to agentic AI, and how should such delegation be governed? We develop an analytical model of AI delegation at the operations–marketing interface in which a firm jointly determines pricing and inventory under demand uncertainty and chooses among human control, full AI autonomy, or human-in-the-loop governance. Agentic AI improves responsiveness by enabling state-contingent decisions, but also introduces new forms of operational exposure by reducing buffers and accelerating execution. Our analysis yields several key insights. First, we identify a demand-variance threshold above which delegating decisions to agentic AI becomes optimal, even when AI is imperfect. Second, we show that partial delegation can strictly dominate both full autonomy and full human control, providing a theoretical foundation for hybrid governance structures widely observed in practice. Third, when AI investment is endogenous, adoption and autonomy become distinct decisions, generating a three-region equilibrium in which firms may invest in AI while deliberately restricting its authority. We further show that learning, service-level asymmetry, stochastic lead time, endogenous human oversight, and organizational scale fundamentally reshape delegation incentives, often in counterintuitive ways: faster learning can delay early autonomy; improved pricing coordination can increase inventory imbalance; and larger organizations may rely on autonomy even under moderate uncertainty. Together, our results demonstrate that AI delegation is not a technological inevitability but an economically contingent organizational choice shaped by uncertainty, risk asymmetry, and structural complexity. The paper provides a unified theoretical framework for understanding AI governance in operations and offers guidance for firms navigating the transition toward autonomous decision-making.