Keongtae Kim

With the rapid expansion of online food delivery (OFD) platforms, subscription models have emerged as a strategic tool to retain users and shape consumption behavior. Despite their growing prevalence, the broader impacts of these subscriptions—both within the OFD sector and across downstream consumer activities—remain underexplored. Leveraging a difference-in-differences approach and a large-scale individual-level transaction data set, we identify a significant shift in consumption from a competing platform to the focal platform following the introduction of a subscription service. This shift did not result in cannibalization; instead, overall food consumption within the OFD market increased. Geospatial analysis shows that subscribers were more likely to engage in offline spillovers, including increased local dining and activity in nearby noncompeting economic sectors. These patterns suggest that subscriptions in physical services, like OFD, not only reshape digital engagement, but also, they influence consumer mobility and stimulate localized economic ecosystems. Mechanistic testing supports the interpretation that these effects are driven by increased variety-seeking behavior for offline restaurants that predominantly serve non-OFD cuisines within a proximal location, suggesting that such behavior is constrained by transportation costs. Overall, this study offers novel insights into how subscription models in physical service platforms can expand market demand, foster crosschannel engagement, and generate positive spillovers in adjacent sectors. The findings offer actionable guidance for platform designers and policymakers to enhance consumer welfare and support local economic vitality. They also provide new evidence for restaurants that additional platform usage costs may be offset by latent economic benefits from heightened consumer engagement. This paper was accepted by D. J. Wu, information systems. Funding: F. J. Costello acknowledges funding by Fundação para a Ciência e a Tecnologia (UID/00124/2025, UID/ PRR/00124/2025, Nova School of Business and Economics DOI: https://doi.org/10.54499/UID/00124/2025 ), UID/PRR2/00124/2025: DOI: https://doi.org/10.54499/UID/PRR2/00124/2025 and LISBOA2030 (DataLab2030 - LISBOA2030-FEDER-01314200). Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2025.00954 .

We investigate the impact of mobile location-based recommendation systems (MLBRS) on the performance of local businesses. Leveraging a comprehensive dataset of 16,720 restaurants and quasiexperimental variation generated by the official launch of a leading MLBRS, we employ a 2-Stage Least Squares instrumental-variable framework to estimate the effect of algorithmic exposure on consumer engagement. Our analysis reveals that a 1% increase in exposure is associated with a 0.078% increase in clicks; separately, a 1% increase in clicks is associated with a 0.053% increase in visit intentions. Importantly, the estimated exposure effect is larger for geographically disadvantaged restaurants, but primarily at the discovery stage (clicks). This pattern highlights a boundary condition of MLBRS: algorithmic visibility can aid, but it does not by itself resolve the uncertainty involved in acting on a recommendation. Downstream conversion is stronger when restaurants provide richer digital information and reputation signals, including consumer reviews and owner-curated information such as photos and menus. Auxiliary credit-card sales data from a restricted Seoul subsample provide consistent evidence of a positive association between algorithmic exposure and revenue, suggesting that MLBRS can shape the connection of online recommendations to offline economic outcomes.

Digital marketplaces concentrate consumer attention among highly visible sellers, creating persistent visibility inequality. We study whether live streaming commerce (LSC), a real-time selling format that concentrates consumer attention into short broadcast windows, reallocates attention across heterogeneous sellers and generates asymmetric competitive spillovers. Using proprietary panel data from a leading Korean e-commerce platform, we examine LSC broadcasts in two product categories from 2020 to 2021. A stacked difference-in-differences design with coarsened exact matching shows that LSC generates large but short-lived sales gains for broadcasting sellers. These gains are accompanied by sharp increases in product page views and are disproportionately larger for lower-tier sellers, consistent with the idea that temporally concentrated attention has higher marginal value when baseline visibility is constrained. We then examine broader market consequences using store-week and segment-week panels. LSC intensity is associated with heterogeneous spillovers among non-broadcasting sellers: same-tier LSC activity is negatively associated with traffic and sales for the smallest sellers but less adverse, and often positive, for higher-tier sellers. At the segment level, LSC intensity is positively associated with aggregate sales and page views, suggesting that attention reallocation and attention expansion may operate jointly. Finally, LSC activity is associated with greater product-positioning similarity among peer sellers, particularly among lower-tier sellers, consistent with competitive convergence in response to attention-driven demand signals. The findings position live streaming commerce as a platform-enabled attention-reallocation mechanism with asymmetric consequences for seller competition.

As AI becomes increasingly embedded in organizational decision processes, the need for effective explanations of algorithmic outputs grows critical. Existing explainable AI (XAI) approaches emphasize technical transparency through visual feature attribution, yet these methods often fail to support meaningful human interpretation. We introduce GenXAI, a conversational explanation layer that translates model outputs into interactive natural-language explanations. Drawing on Cognitive Fit Theory and Construal Level Theory, we argue that explanation effectiveness depends on alignment between linguistic explanation style and the cognitive demands of decision tasks. Across two randomized experiments using a loan default prediction task, we show that conversational GenAI explanations significantly improve users' understanding of algorithmic predictions and decision confidence relative to static XAI visualizations. A second experiment isolates the mechanism by manipulating explanation style: concrete explanations—characterized by numerical specificity, benchmark comparisons, and structured reasoning—produce stronger improvements than abstract narrative explanations. Behavioral telemetry from chatbot interactions reveals that users exposed to concrete explanations process information more efficiently while spending more time deliberating, a pattern consistent with active reasoning rather than passive AI reliance. These findings advance XAI research by demonstrating that explanation effectiveness depends not only on technical transparency but on representational alignment between explanation style and task cognition. More broadly, the study highlights linguistic explanation design as a critical lever in conversational AI systems. For practice, organizations can improve human oversight of AI by complementing static transparency dashboards with conversational explanations that enable users to interrogate model reasoning.

Crowd-generated rating systems are ubiquitous on online platforms, yet little is known about their impact on investment decisions in crowdfunding. This study examines the introduction of a crowd-generated rating system on a leading South Korean platform that adopted ratings for equity crowdfunding while leaving reward crowdfunding unchanged. Using investment data and a difference-indifferences design, we identify the causal effects of ratings on investor behavior. We find that ratings increase funding frequency and amounts, accelerate investment timing, and reduce herding, with stronger effects among less experienced investors. Beyond the system's presence, we document participatory bias: investors who rate campaigns are significantly more likely to invest in them, and to invest earlier and in larger amounts. At the campaign level, ratings weaken the predictive power of conventional quality signals, suggesting that they substitute for traditional indicators. We also find that rating scores and evaluator volume indirectly improve startups' longer-term outcomes by increasing capital raised during crowdfunding. Overall, our findings demonstrate that crowd-generated ratings meaningfully shape investment behavior and campaign dynamics, providing an effective mechanism to support decision-making under uncertainty and information asymmetry.

With the rapid expansion of online food delivery (OFD) platforms, subscription models have emerged as a strategic tool to retain users and shape consumption behavior. Despite their growing prevalence, the broader impacts of these subscriptions—both within the OFD sector and across downstream consumer activities—remain underexplored. Leveraging a Difference-in-Differences (DID) approach and a large-scale individual-level transaction dataset, we identify a significant shift in consumption from a competing platform to the focal platform following the introduction of a subscription service. This shift did not result in cannibalization; instead, overall food consumption within the OFD market increased. Geospatial analysis shows subscribers were more likely to engage in offline spillovers, including increased local dining and activity in nearby non-competing economic sectors. These patterns suggest that subscriptions in physical services like OFD not only reshape digital engagement but also influence consumer mobility and stimulate localized economic ecosystems. Mechanistic testing supports the interpretation that these effects are driven by increased variety-seeking behavior for offline restaurants that predominantly serve non-OFD cuisines within a proximal location suggesting such behavior is constrained by transportation costs. Overall, this study offers novel insights into how subscription models in physical service platforms can expand market demand, foster cross-channel engagement, and generate positive spillovers in adjacent sectors. The findings offer actionable guidance for platform designers and policymakers to enhance consumer welfare and support local economic vitality. They also provide new evidence for restaurants that additional platform usage costs may be offset by latent economic benefits from heightened consumer engagement.

This study investigates the short-term impact of generative AI, exemplified by the introduction of ChatGPT, on user contributions in a coding Q&A platform. We find that the introduction of ChatGPT led to a reduction in the number of high-quality answers provided by users, particularly among highly engaged contributors, despite an overall increase in answers. We identify two key mechanisms: (1) increased perceived question sophistication despite no actual change in content and (2) reduced motivation of loyal users in providing answers in the face of AI-generated alternatives. The findings suggest that while generative AI can facilitate value creation on user-generated content (UGC) platforms, it also poses challenges in retaining core contributors and managing content quality. The paper contributes to the literature on the impact of AI adoption on platforms and suggests practical implications for UGC platform management, such as the need for AI content disclosure measures to retain engaged users.

While machine-generated information is increasingly prevalent, how it is used as a basis for human ratings is not well-explored. Using the context of online professional ratings of initial coin offerings (ICOs) projects, this study examines how increased social exposure of human ratings and different experiences impact human experts' ratings relative to machine-generated ratings (MGRs). Leveraging an interface design change on an ICO rating platform, we find that increased social exposure leads experts with advisor experiences to lower ratings and rate below and closer to MGRs. Additionally, increased social exposure leads human experts with team member experiences to rate closer to MGRs, without significantly affecting their rating levels. These suggest that human experts with advisor experiences may strategically rate above MGRs to overrate and impress project teams, while those with team member experiences do not. Overall, increased social exposure drives human experts to conform to MGRs, possibly correcting humans' rating biases.

This study investigates the impact of digital piracy on illegal gambling activities in the context of live sports broadcasting. When previously free content becomes restricted, users may resort to illegal channels to access the content. This research analyzes the relationship between digital piracy and illicit gambling by examining a specific event in South Korea where a media company introduced a subscription model for English Football League broadcasts. The study finds a positive correlation between increased visits to illegal sports broadcasting websites and higher engagement in illegal gambling. These findings contribute to our understanding of the consequences of digital piracy and provide insights for industry stakeholders and policymakers.

Large language models (LLMs) like ChatGPT are reshaping content creation and knowledge sharing on user-generated content (UGC) platforms. While they enable AI-assisted content generation and potentially increase interactions among users, concerns arise regarding possible discouraging effects on user participation through increased competition for visibility and eroded trust stemming from an influx of low-quality content. This study presents timely evidence on changes in user contributions following the introduction of ChatGPT. Conducting a difference-indifferences analysis on a coding Q&A platform, our results reveal a reduction in user effort, marked by a decline in contributions from active users. Despite no immediate decline in overall quality, the reduced participation, possibly caused by increased question complexity, calls for cautious consideration. Furthermore, the quality loss for answers from junior users also sets an alarm for platforms. Our findings underscore the necessity of monitoring LLM impacts on UGC platforms and formulating harmonious human-AI collaborations. Overall, this study offers critical insights into the emerging influence of generative AI on UGC platforms.