5월 3주차

While agile software development (ASD) promises rapid, iterative delivery, agile teams often face temporal demands—such as organizational reporting schedules, quality requirements, and resource availability—that challenge their ability to meet this promise. These conflicting temporal demands create what we call temporal misfits. Based on an in-depth study of five software development projects, we found that a temporal misfit disturbs an agile team’s work by creating delays and undermining software quality every time it occurs. Because a given temporal misfit reoccurs at each sprint until resolved, work disturbance escalates over successive sprints. Teams respond in different ways. They may sacrifice the speed of delivery and comply with demands that jeopardize it. They may preserve the agile rhythm, sometimes shielding the team from external temporal requirements. Finally, they may mobilize people or tools—digital or not—to play the role of a differential gear, therefore allowing conflicting temporal demands to be met simultaneously. Our work invites ASD teams to consider both the immediate and the longer-term effects of temporal misfits and their responses, highlighting how decisions made within each agile sprint can impact the entire project.

As social media platforms deploy large language model (LLM)-powered agents to help influencers manage social relationships with users, it remains unclear how this delegation impacts user engagement. Automating interactions provides scalability and efficiency for influencers, but it may weaken the influencer-user relationship if the agents fail to serve as effective social delegates. To explore this question, we empirically investigate the impact on user engagement when influencers delegate social interaction tasks, such as replying to comments, to a social artificial intelligence (AI) agent, an LLM-powered proxy that responds on behalf of an influencer. Leveraging the rollout of a social AI agent feature on a major social media platform, we use a staggered difference-in-differences design to compare engagement behaviors between users who received an AI reply (i.e., a reply from an influencer’s social AI agent) and those who did not. Our results show that receiving an AI reply significantly increases user commenting on subsequent influencer posts, particularly when AI replies amplify an influencer’s social presence, as reflected in content relevance, stylistic alignment, and reply timeliness. We also find heterogeneous effects based on influencer-user relationships: engagement gains are stronger among loyal followers but weaker for commercialized influencers and those in the technology domain. Additionally, reply scarcity amplifies the effect: engagement increases more when influencers rarely replied previously or when fewer AI replies appear under the focal post. The engagement boost extends to both sponsored and nonsponsored posts, as well as user reposting behavior, whereas influencers themselves also post more frequently after adopting AI agents. This study contributes to the literature on AI delegation and influencer engagement by highlighting when and how delegating social relationship management to social AI agents can enhance user engagement. History: Jeffrey Parsons, Senior Editor; Pallab Sanyal, Associate Editor. Supplemental Material: The online appendices are available at https://doi.org/10.1287/isre.2025.2270 .

This study introduces a novel, human-centered framework for evaluating the holistic intelligence of large language models (LLMs), using behavioral theory and experimental benchmarks drawn from human intelligence. Through extensive online experiments, the framework reveals that GPT-4 outperforms humans in cognitive, emotional, and creative intelligence, but falls short in social intelligence, especially in social interest, self-efficacy, and understanding mental states. Beyond theoretical insight, the study validates this framework by assessing GPT-4’s impact across diverse job roles, finding results consistent with established labor market research. It also offers a reusable tool for firms and policymakers to evaluate LLM intelligence and forecast job-level impacts. This enables informed decisions about where and how to integrate LLMs, match models to specific job requirements, and identify risks in socially intensive roles. The framework provides a foundation for responsible LLM deployment, ensuring alignment with human-centered structures and supporting strategic workforce planning.

This study introduces an efficient specialized artificial intelligence (AI) tool and the SEPTE model—a comprehensive framework for evaluating healthcare service quality—to help healthcare platforms and hospitals better understand what drives patient demand for online consultations. By analyzing physician reviews from one of China’s largest telehealth platforms, our small language model (Doc-BERT) uses the SEPTE framework to accurately identify key aspects of service quality, such as medical effectiveness and empathy, that matter most to patients. Unlike traditional large language models, our approach is cost-effective and can be readily implemented in real-world healthcare settings. We find that higher service-quality scores, especially in effectiveness and patient-centeredness, lead to greater patient demand for online consultations. These insights offer actionable guidance for healthcare providers and administrators seeking to improve patient experiences, optimize physician performance, and inform platform design and policy. Our work demonstrates that targeted, domain-specific AI—guided by the SEPTE model—can deliver both efficiency and impact for digital health services.

Whereas online reviews have become a primary data source for understanding customer requirements in both research and practice, using such information alone to guide product design can be unreliable. Our research investigates whether and how consumers’ preferences expressed through online reviews (words) align with their actual purchase decisions (deeds). Our analysis shows that features praised in online reviews do not necessarily translate to market success. This inconsistency between what consumers say and what they do poses significant challenges for manufacturers in product development decisions. We empirically identify the existence of word–deed inconsistency in consumer preferences. Some features are silent in online reviews yet significantly drive purchase decisions, whereas others are frequently praised but have limited influence on actual purchases. Building on these insights, we propose an innovative dual-weights model that extends existing two-dimensional customer requirement analysis by integrating both prepurchase choice drivers and postpurchase satisfaction determinants. Using this model, we classify features based on their importance for satisfaction versus purchase decisions and offer actionable product improvement strategies for different types of features.

Equity appeals are increasingly used by digital fundraising platforms, nonprofits, and public institutions to direct attention and resources toward disadvantaged communities. However, it remains unclear whether and when equity appeals actually increase giving. We examine this question in the context of educational crowdfunding, where platforms explicitly focus on reducing funding disparities across schools, particularly for students from racial or ethnic minority and low-income backgrounds. Leveraging large-scale data from DonorsChoose, one of the largest educational crowdfunding platforms in the United States, and exploiting arbitrary cutoffs in the platform’s deployment of equity appeals based on the student composition of benefitting schools, we show that equity appeals increase fundraising when they highlight student disadvantage in terms of poverty while providing little to no measurable benefit when they highlight student disadvantage in terms of race. These differential effects reflect how donors interpret disadvantage. Many donors appear to view poverty as a legitimate and actionable barrier to learning, making poverty-based appeals effective. In contrast, perceptions of race as a structural barrier to educational opportunity are more heterogeneous and politically sensitive, limiting the impact of race-based appeals. For platform designers and policymakers seeking to reduce educational fundraising disparities, our findings highlight the importance of how equity appeals are framed. More broadly, our results contribute to understanding under what conditions behavioral nudges can meaningfully reduce inequality versus when alternative approaches may be necessary to achieve equitable outcomes.