Knowledge that Transforms

Problem definition: Are customers loyal to a ride-hailing platform or they see this service as a commodity and multihome (i.e., check several platforms before booking a ride)? Using a large panel dataset on ride-hailing transactions, we investigate to what extent customers multihome. Our dataset offers a unique opportunity to study this question as we observe the repeated choices of riders for both Uber and Lyft. Our dataset comprises more than 1.4 million rides completed by 162 thousand riders in NYC in 2018. Methodology/results: We develop a comprehensive structural model that incorporates both operational (price and waiting time) and behavioral factors (e.g., platform stickiness) to explain riders’ choices. Our model also accounts for the dynamic interactions between customers and platforms by assuming that riders update their beliefs on price and waiting time in a Bayesian fashion. Finally, the riders’ propensity to multihome is modeled by incorporating the consideration set formation of customers into our framework. We find that riders’ choices are not fully explained by operational factors, hence indicating that customers view the platforms as differentiated service providers. While 83.4% of riders took rides with a single platform, our model shows that even the remaining 16.6%, who used both Uber and Lyft at least once, considered both platforms only 43.4% of the time. Managerial implications: It is crucial for ride-hailing platforms to capture this single (or multi)-homing behavior while designing price discounts. Specifically, personalized discounts may be ineffective if the platform is not part of the customer’s consideration set. Our results show that targeting customers earlier in their lifecycle can enhance the platform’s market share by 77.56% more than their current discounting strategy. We also find that targeting customers with low search friction results in a 24.78% increase in market share relative to targeting customers with high search friction.

Problem definition: In the platform economy, platforms often engage in both demand and supply competition. We consider two platforms competing through service prices and wages. Instead of exploring myopic stability, i.e., Nash equilibrium, which focuses on the immediate payoff change from a one-step deviation, the two competing platforms are farsighted and can anticipate their rivals’ reactions. They consider the chains of reactions following their initial deviation as well as the long-term consequences of their actions. Methodology/results: We model the competing platforms by a queueing system where customers and workers strategically decide whether to join and which platform to join. We first determine the equilibrium of the worker-customer subgame and then analyze the platforms’ service pricing decisions. To derive farsighted stability, we apply the concept of the von Neumann-Morgenstern farsightedly stable set (vNM FSS), which requires both internal and external stabilities. To find the vNM FSS, we first prove that any vNM FSS for the service platform competition is a singleton, and then show that a strategy profile being Pareto efficient is a sufficient and necessary condition to form a singleton vNM FSS. In this way, we identify all the vNM FSSs by determining the Pareto-efficient strategy profiles. Managerial insights: We show that while myopically stable outcomes leave both platforms zero profit, platforms can coexist in the market and earn positive profits at the farsightedly stable outcomes. Farsightedness enhances the profitability of the platforms by fully extracting surplus from both the demand and supply sides. However, it cannot achieve the efficiency of a merged system where two platforms are operated by a single entity. We further show that the advantage of farsightedness over myopia is independent of the platform’s structure (whether one-sided or two-sided), whereas the efficiency of farsightedness does depend on the platform’s structure.

Problem definition: In this paper, we examine how firms offering AI services can effectively acquire large volumes of training data from their consumers to improve the accuracy of the machine learning (ML) models that drive these services. Since consumers often incur privacy costs when sharing sensitive information, it is essential to design data-sharing mechanisms that balance data acquisition needs with consumers' privacy. Methodology/results: Inspired by practice, we examine two fundamentally distinct data-sharing mechanisms: manual data-sharing, where consumers control the amount of data they share, and algorithmic data-sharing, where the firm’s algorithm redacts sensitive segments of data before using it to train the ML model. We obtain revenue-maximizing mechanisms for each approach and compare their impact on firm revenue, consumer surplus, and the volume of data collected. Our analysis highlights the conditions under which each mechanism yields superior outcomes in terms of revenue and consumer surplus. Managerial implications: Based on the comparative performance of the two mechanisms, we provide managerial guidelines that help firms choose the preferred data-sharing mechanism for different types of AI services and consumer characteristics.

Problem definition: Early-stage patient-centered clinical trials aim to establish the toxicity profile of investigational drugs and determine safe dose ranges for future trials using cohort-specific consent. Patients considering participation face intricate benefit-risk tradeoffs, particularly regarding the timing of their enrollment: Early enrollment corresponds to low doses with potentially low efficacy and toxicity with high uncertainty around them, while late enrollment corresponds to high doses with potentially high efficacy and toxicity with reduced uncertainty, but also an increased risk of disease progression and trial termination. Methodology/results: This paper develops a Bayesian model where a patient determines the optimal enrollment time based on health status and evolving beliefs about dose toxicity and efficacy. The patient periodically learns about the benefits and risks of doses and enrolls in the trial at the optimal time to maximize the expected benefits. We show that the optimal policy is of a control-limit type, where the threshold is determined by when the benefit of taking the current dose exceeds the benefit of waiting for future doses. We also establish sufficient conditions for the existence of a control-limit enrollment time policy based on the patient’s health status and beliefs. Managerial implications: Our study offers valuable insights into the nuanced benefit-risk tradeoffs patients face in enrolling early-stage clinical trials, especially the dynamic interplay between their deteriorating health status and evolving beliefs about the drug’s efficacy-toxicity profile. The control-limit structure of the optimal policy is robust across most settings, e.g., the structure is preserved under response delays from previous trials, under other trial designs such as the continual reassessment method, and when trial exclusion criteria are imposed. Our case study using lung cancer clinical trial data further illustrates how factors like belief and observation uncertainties influence patient decision-making.

Problem definition: Order fulfillment plays a pivotal role in shaping the competitiveness and profitability of online retailers. However, the erratic nature of order arrivals at fulfillment centers leads to imbalanced workloads and soaring operational costs. To tackle this challenge, we investigate the practice of anticipatory packing as a potential solution. Anticipatory packing involves strategically preparing packages during non-peak periods to fulfill orders during subsequent peak periods. In the face of order arrival uncertainties, our objective is to optimize the selection of packages to be packed during these non-peak periods. Methodology/results: We develop a two-stage sample-average approximation model using recent order data to enable effective anticipatory packing. For the second-stage problem, which optimizes the usage of prepacked packages, we design a constant-factor approximation algorithm. The first-stage problem is shown to be NP-hard to approximate within a factor of [Formula: see text] asymptotically, where [Formula: see text] is the sample size and [Formula: see text] is a measure for the inherent order structures. To address this, we propose general approximation algorithms with an approximation ratio of [Formula: see text], where [Formula: see text] is the maximum size of prepackages. For orders with laminar structures, the approximation ratio improves to [Formula: see text]. To enhance practical performance, we introduce a refined integer programming reformulation and an efficient subgradient descent method for solving the associated Lagrangian dual. Experimental results with real-world data demonstrate that anticipatory packing can reduce operational costs by over 7% while significantly lowering fixed investment expenses. Managerial implications: We showcase the significant potential of anticipatory packing as an analytics-driven operational strategy. Its widespread adoption could lead to substantial reductions in both operating expenses and fixed investment costs.

Problem Definition: We study delay information disclosure policies for on-demand platforms serving two user classes (customers and providers) who seek matches using the platform. The platform's objective is to maximize the match rate by choosing the level of information—no information, binary information (indicating whether the wait is zero or non-zero), or occupancy information (indicating the expected delay based on the number of users currently in the system)—to disclose to each user class. Users of each class are strategic and decide whether to join or balk based on the delay information disclosed to them. Methodology/results: We consider two user types in each user class: patient users who are willing to wait for a match and impatient users who are not. We use continuous-time Markov chains to model the system as two-sided queues and employ equilibrium analysis to characterize users' joining behavior and the platform's match rate under each information disclosure policy. We show that the two-sided system decouples and disclosure decisions can be analyzed as two one-sided systems only if some level of information (binary or occupancy) is disclosed to both user classes. We find that disclosing binary information dominates disclosing no information or occupancy information to a user class when its patient users are sufficiently patient, while disclosing occupancy information dominates disclosing binary information to a user class when there are enough patient users. Numerical experiments show that disclosing occupancy information to both user classes, while often suboptimal, is typically not suboptimal by much. Finally, we find that compared to the platform's optimal disclosure choice, a user class may prefer more or less granular information for themselves or the other class. Furthermore, this utility analysis does not lend itself to decoupling. Managerial implications: Our findings hold crucial implications for platform managers: carefully evaluating the chosen information-sharing strategy is imperative, and guidelines from the one-sided literature are generally inadequate for making disclosure decisions

Problem definition: As the operational boundaries of two-sided platforms gradually blur, we explore pricing schemes and strategic decisions on platform compatibility. In particular, a dominant platform with awareness and valuation advantages might open an interface for a competitive niche platform, which may choose to set unified or differentiated prices on the compatible channel compared with its exclusive channel. Methodology/results: We develop stylized game models to characterize the co-opetition interaction between the dominant and niche platforms. First, we find that compatibility generally intensifies competition, leading to price reductions. However, under certain conditions, platforms may raise prices on their exclusive channels when compatibility occurs, with the underlying rationale differing between the unified and differentiated pricing schemes. Second, compatibility can be sustained as an equilibrium strategy only when the level of awareness of the niche platform is high and the valuation increment from compatibility with the unified pricing scheme is low. In the differentiated pricing scheme, the niche platform’s pricing power erodes the dominant platform’s profits, making compatibility less viable. Furthermore, by comparing the effects of the two pricing schemes, we find that the differentiated pricing scheme allows for a greater concentration of profits and demand in the compatible channel than the unified pricing scheme does. Finally, compatibility can enhance social welfare when the niche platform’s valuation increment is small. Managerial implications: Our study provides guidelines on the conditions under which platform firms can realize compatibility. We demonstrate that the essence of compatibility lies in the redistribution of surplus between platforms. We also identify the co-opetitive mechanisms between platforms under different pricing schemes, highlighting how these mechanisms evolve depending on whether platforms adopt unified or differentiated pricing. Managers can learn how key features such as awareness and valuation advantages affect platform firms’ compatibility under different pricing schemes.

Problem definition: Worker absenteeism poses operational challenges across industries. Although extensively studied in traditional workplaces, its impact on online-to-offline service platforms remains overlooked. Unlike team-based settings, where colleagues can cover absences, customers on these platforms are particularly vulnerable to service provider no-shows because each customer is matched with one independent contractor. The service cannot proceed if the provider fails to appear. This paper addresses this gap by examining how service provider no-shows affect customer attrition and investigating the unique challenges and opportunities in designing effective mitigation strategies within the service platform context. Methodology/results: Using detailed transaction data from a European home-cleaning platform, we find that no-shows increase customer attrition probability by 25.6%, especially among loyal customers, and reduce customer lifetime value by 1.55% despite their low prevalence (3.51%). These notable adverse effects call for strategies to combat no-shows. We find that predicting gig workers’ no-shows ex ante is challenging, highlighting the need for post–no-show remedy strategies. Our findings suggest that frustration discounts after no-shows can help reduce attrition, whereas finding a substitute to restore service later the same day does not mitigate attrition. There is suggestive evidence that restoration is effective only if provided promptly (within one hour). We then build an analytical model that takes these empirical insights as inputs and delivers profit-maximizing optimal remedy strategies for service platforms tailored to customer and platform characteristics. Managerial implications: Our study sheds light on the significant adverse impact caused by no-shows in online-to-offline service settings, despite their rarity. We provide a framework that enables online-to-offline service platforms to design optimal remedy strategies to combat no-shows based on their own data, addressing the unique challenges of worker absenteeism in the gig economy settings. History: This paper was selected as part of the 1RR initiative between the M&SOM Journal and the MSOM Society.

Problem definition: We consider a price-setting newsvendor problem in which the demand distribution is unknown. We assume that the retailer exercises only a few price points with a sample set of demand realizations for each exercised price. Given this limited demand information, we define an ambiguity set and study two robust optimization models that maximize the worst-case profit (maxmin profit) and minimize the maximum regret (minmax regret), respectively. Methodology/results: These two robust models are reduced to easy-to-solve optimization problems. Compared with the maximal profit under complete demand information, we find that with only a few price points, we can achieve more than 90% of the maximal profit on average and around 70% of the maximal profit in the worst case. Our method is purely data driven and model free; that is, we do not assume that the demand model follows any specific form. This approach has the advantage of avoiding model mismatches in practice. Managerial implications: We show that this new method outperforms traditional methods, such as regressions, and other model-specific methods. We also propose demand learning methods with guaranteed convergence rates when the number of exercised prices increases.

Problem definition: Yield uncertainty is a ubiquitous and serious issue in supply chain management. A common mitigation strategy for buyers is dual sourcing: ordering the same good from multiple suppliers to hedge against uncertainty in the suppliers’ ability to fulfill orders. In the literature, it is typically assumed that there is a single buyer that knows or can learn the reliability of each supplier, which is assumed to be exogenous to the buyer’s ordering behavior. However, in supply chains with a large number of buyers, this assumption is limiting. Methodology/results: We study a stylized supply chain model consisting of n buyers and two suppliers. The buyers do not observe their suppliers’ reliabilities a priori but can learn about them over time based on their interactions with the suppliers (in the learning setting) or through supply chain visibility (in the upstream visibility setting). We allow for the suppliers’ reliabilities to be impacted by the buyers’ ordering behaviors; when suppliers receive larger order volumes, it diminishes their ability to meet a buyer’s order in full, and thus, their perceived reliability falls. As a result, buyers—who cannot distinguish whether shortfalls are caused by low supplier capacity versus order congestion—may increase order quantities to hedge against shortage risk. We study the long-run ordering dynamics that emerge from this model in both settings. Managerial implications: We find that without visibility, “overordering spirals” can emerge, whereby buyers continue to inflate orders to the suppliers because they perceive shortages. This phenomenon has been observed when sudden, unexplained shortages occur, such as in the semiconductor or drug manufacturing industries. Interestingly, this can occur even in cases where shortages would not arise under truthful ordering but are created by inflated orders. Upstream visibility can be used as a tool to combat these spirals and can benefit both the buyers and the suppliers. History: This paper was selected as part of the 1RR initiative between the M&SOM Journal and the MSOM Society. This paper was part of the 2024 MSOM Supply Chain Management SIG Conference.