Scalable and low-cost artificial intelligence (AI) assistance has the potential to improve firm decision making and economic performance, particularly in emerging markets. However, running a business involves a wide range of open-ended problems, making it unclear whether and how recent advances in AI can help business owners around the world make better decisions. In a field experiment with Kenyan entrepreneurs, we evaluated the impact of AI advice on small business revenues and profits by randomizing access to a GPT-4-powered AI business assistant. Although we are unable to reject the null hypothesis of no average treatment effect on firm revenues and profits, we find that the effect for entrepreneurs who were low performing at baseline is over 0.20-standard-deviations lower than for initial high performers. Subsample analyses show that low performers did nearly 10% worse because of the AI assistant, whereas high performers may have benefited by over 15%. This differential impact does not appear to result from differences in the questions posed to the AI or the advice that it provided but rather, from the advice that entrepreneurs chose to implement. More broadly, these results show that generative AI is already capable of impacting real-world business performance—although in uneven and sometimes unexpected ways. This paper was accepted by Anita McGahan, business strategy. Funding: This work was supported by the Weiss Fund, Harvard Business School, Berkeley Haas, the Cora Jane Flood Endowment at Berkeley Haas, the Agency Fund, the South Park Common Social Impact Fellowship, and the Digital Data Design Institute at Harvard. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.06909 .
We return to the theoretical foundations of absorptive capacity and test the idea that personal experience in a field makes it easier for a firm’s inventors to recognize and build upon knowledge in that field from other local firms. We propose a new empirical model of localized knowledge diffusion, which (1) measures a firm’s absorptive capacity by its inventors’ prior experience in a field, (2) uses a death instrument to exogenously vary the availability of knowledge of the same collaborative patent in different regions, and (3) estimates the difference in citation likelihood from all subsequent inventors across both regions as a function of a potentially citing inventor’s prior experience in the field. Consistent with the original theory of absorptive capacity, firms whose inventors have prior experience in a field are more likely to use locally available interpersonal knowledge from other firms; furthermore, this effect declines monotonically with distance. Although the effects strengthen for more recent experience and multidisciplinary knowledge, the greatest benefits accrue to firms in the interaction—those whose inventors have more recent experience and that seek to absorb multidisciplinary knowledge. Interpersonal absorptive capacity within firms does not appear to localize. This paper was accepted by Anita McGahan, business strategy. Funding: The authors acknowledge financial support from the Alfred P. Sloan Foundation and the National Science Foundation [Grant 1360228]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.06147 .
Consumer experiences are inherently dynamic. When engaging in a sequence of activities, consumers are influenced by past experiences in two ways: negatively by their objective quality, through quality contrast; and positively by their subjective appreciation of them, through affect assimilation. How should experience curators sequence activities to maximize consumer satisfaction in the presence of such intertemporal effects? We formulate an experience curator’s problem as a dynamic optimization program. We show that, because of affect assimilation, the best activity may be scheduled at the beginning or in the middle of an experience, in contrast to the common peak-end rule—this provides a rationale for the saying that “first impressions matter.” Under uncertainty, it may be valuable to save the best activity as a “wild card” to recover from bad outcomes. We calibrate our model to four distinct experiential contexts (namely, watching movies, reading books, visiting touristic attractions, and eating out) and consistently find the presence of both quality contrast and affect assimilation. Through a counterfactual study in the context of touristic tours, we show that experience curators may significantly benefit from offering different fixed sequences to different types of consumers, but they tend to gain little from dynamically adjusting them. This paper was accepted by David Simchi-Levi, operations management. Funding: This work was supported by the Agencia Estatal de Investigación [Grant PID2020-116135GB-I00 MCIN/AEI/10.13039/50110001]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2023.02150 .
Explainable artificial intelligence (AI) models have been proposed to mitigate overreliance and underreliance on AI, which reduce the effectiveness of human-AI collaborative tools. Yet, empirical evidence is mixed, and the impact of explainable AI on the cognitive effort and fatigue of a decision maker (DM) is often overlooked. This paper offers a theoretical perspective on these issues. We develop an analytical model that incorporates the defining features of human and machine intelligence, capturing the limited but flexible nature of human cognition with imperfect machine recommendations. Crucially, we represent how AI-based explanations influence the DM’s belief in the algorithm’s predictive quality. Our results indicate that explainable AI has varying effects depending on the level of explainability provided. Although low explainability levels have no impact on decision accuracy and reliance behavior, they lessen the cognitive burden on the DM. In contrast, higher explainability levels enhance accuracy by improving overreliance but at the expense of increased underreliance. Further, the relative impact of explainability is higher when the DM is more cognitively constrained, when the decision task is sufficiently complex, or when the stakes are lower. Importantly, higher explainability levels can escalate the DM’s cognitive burden (and hence, overall processing time and fatigue) precisely when explanations are most needed (i.e., when the DM is pressed for time to complete a complex task and doubts the machine’s quality). Our study clarifies how explainability affects decision outcomes and cognitive effort, informing the design of effective human-AI systems across decision environments. This paper was accepted by Jeannette Song, operations management. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.07450 .
When designing product rankings, online retailers and platforms choose which outcome to maximize: revenues from commissions or markups, the number of transactions, or consumer welfare. These objectives need not align, creating potential trade-offs. This paper studies how rankings differ between objectives and quantifies the resulting trade-offs. I provide descriptive evidence showing that lower-priced and high-utility alternatives gain more demand when ranked higher, suggesting that ranking them higher increases transactions and consumer welfare but may decrease revenues. To quantify these trade-offs, I develop and estimate a structural demand model in which consumers search and discover products, and construct rankings for each objective. The results show that these counterfactual rankings all increase consumer welfare, transactions, and platform revenues relative to a neutral benchmark and the status quo and that trade-offs between these rankings are limited. This paper was accepted by Jean-Pierre Dube, marketing. Funding: R. P. Greminger received financial support from the Nederlandse Organisatie voor Wetenschappelijk Onderzoek [Research Talent Grant 406.18.568]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2025.00782 .
How does generative artificial intelligence (GenAI) reshape the skills that organizations seek as they adapt to a new general-purpose technology? GenAI effectively retrieves data, performs analysis, and conveys information, so it can substitute for workers doing these activities and complement workers relying on them. A natural consequence is skill deprioritization, a systematic reduction in firms’ demand for human skills that GenAI can effectively address as organizations adjust the division of labor and integration of effort. We draw on a theoretically grounded classification of organizing skills—task division, task allocation, information provision, reward distribution, and exception management—and a queuing theory model of organizing efficiency to predict which skills firms will deprioritize first. Using a quasiexperimental design that leverages the introduction of ChatGPT as an exogenous shock, we analyze 1,820 publicly listed U.S. companies and track changes in their hiring demand patterns over a period of a ±12-month window surrounds the shock. We find significant declines in demand for monitoring (reward distribution), operational exceptions, and task division skills, with information provision also showing declines. Task allocation and conflict resolution showed greater stability, suggesting greater reliance on human judgment. These effects intensify following GPT-4’s release, indicating that capability improvements also drive adaptation. Our findings demonstrate that firms engage in immediate and selective skill deprioritization, raising questions about longer-term hollowing out of human expertise in automated domains. We contribute a novel taxonomy of organizing skills, extend queuing theory to the GenAI context, and provide early empirical evidence on how GenAI is reshaping organizational skill demands. This paper was accepted by Anita McGahan, Business Strategy. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2025.01859 .
We present one of the first systematic audits of cryptocurrency market data quality across leading vendors. We document pervasive mislabeling, identifier instability, and large cross-provider discrepancies in prices, market caps, and volumes. To address these issues, we develop an aggregation method that yields asymptotically correct data by autonomously identifying and filtering unreliable observations. Using this framework, we construct an index to measure data quality over time and a grading system to benchmark providers. Our findings show that data inconsistencies can materially distort empirical research and investment analysis. They highlight possible oversight gaps in the market for crypto data. This paper was accepted by Will Cong for the Special Issue on Digital Finance. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2025.00611 .
Adoption of artificial intelligence (AI) by authors has accelerated production of academic articles and increased submission rates to journals, thereby straining review capacity and hurting journal outcome metrics, such as turnaround time and decision accuracy. Academic journals face an imperative to improve review quality and productivity by incorporating generative AI tools in the review workflow. As a concrete first step toward this idea, we outline a particular workflow that deploys large language models as a structured, trained, frontline reviewer. The workflow underlines that AI-produced evaluations must be transparent and contestable by authors. We emphasize the need for AI-infused peer review to be journal specific, designed for efficiency, accuracy, and accountability, within a human-in-the-loop oversight framework. We build a mathematical model of journal operations to compare outcomes under a status quo no-AI workflow and the proposed AI-infused workflow. The model captures how governed AI reconfigures human reviewer effort and illuminates conditions under which the AI-infused workflow can improve vital metrics, such as turnaround time and decision accuracy. Our essential point is that journals need to adopt a deliberate and institutionally governed approach for using AI in the review process. We recognize that identifying an “optimal” AI-infused peer review workflow or even definitively projecting the consequences of one will require substantial experimentation to calibrate and configure the workflow across multiple alternative designs. This discussion was accepted by Christoph Loch. Funding: D. Zantedeschi acknowledges support for this research from the Muma College of Business Dean’s Research Fund. Supplemental Material: The online appendix is available at https://doi.org/10.1287/mnsc.2026.00184 .
Theoretical research suggests internal information asymmetry (IIA) between top managers and divisional managers can reduce productivity by distorting internal resource allocation. The net effect, however, is unclear because IIA also reflects valuable private information within divisions, and empirical evidence is limited because of the unobservability of managers’ private information sets. Using a measure of IIA that exploits differences in returns to insider trading by top and divisional managers, we find that higher IIA is associated with lower total factor productivity. A one standard deviation increase in IIA corresponds to a 2.6% decrease in productivity relative to industry peers. This effect is comparable in magnitude to other important internal informational frictions, such as internal control weaknesses, and is distinct from internal information quality. Cross-sectional analyses indicate that the negative association is stronger when curative measures are limited and when agency conflicts between top managers and shareholders are greater. Mechanism tests indicate that IIA reduces internal capital market efficiency and increases underinvestment. Finally, using brokerage mergers and closures as a plausibly exogenous shock, we find that increases in IIA lead to declines in productivity. Overall, we provide empirical evidence that IIA is an economically important determinant of firm-level productivity. This paper was accepted by Jan Bouwens, accounting. Supplemental Material: The code and data files are available at https://doi.org/10.1287/mnsc.2024.08526 .
Prior research indicates that social capital constrains opportunistic behavior and facilitates cooperation in bilateral relationships. We hypothesize that firms conducting business with high social capital customers are perceived by their banks as having lower supply chain risk and document a negative association between suppliers' loan spreads and the social capital of their major customers. Our cross-sectional analyses show that the negative association is stronger when (1) suppliers make more relationship-specific investments or have weaker bargaining power; and (2) customers' information environment is more opaque. Further analyses reveal that customers' social capital has a positive impact on their supplier firms' nonpricing loan terms. Taken together, our findings indicate that customers' social capital engenders a spillover effect along the supply chain, enabling suppliers to enjoy lower borrowing costs and more favorable nonpricing loan terms by mitigating the hold-up risks they face.
We study the relation between firms' environmental, social, and governance (ESG) performance and the aggregate stock market returns. Based on 38 individual ESG measures, we construct a market-level ESG index. With both the traditional predictive regression approach and two recently developed machine-learning methods, we find that the ESG index has strong and positive predictive power on the market both in-and out-of-sample, and both the cash flow and discount rate channels are the economic drivers of predictability. Our results are robust to a number of controls and set-ups. Our novel finding on the significant market-wide impact of the ESG provides support for the economy-wide importance of the ESG risk and for the central role played by governments.
This is a commentary by Dr. Hans-Paul Bürkner (Global Chair Emeritus and former CEO of The Boston Consulting Group) on Yang, Bauer, Li, and Hinz (2025) “My Advisor, Her AI, and Me: Evidence from a Field Experiment on Human–AI Collaboration and Investment Decisions.”
This study quantifies the impact of voluntary carbon goal announcements on cumulative abnormal stock returns (CAR). Using 188 announcements of U.S. publicly traded firms between 2018 and 2024 across a variety of industries, this event study offers new and economically relevant insights on when and how firms should announce their carbon goals. First, a voluntary carbon goal announcement is associated with 0.65% positive CAR for the main event window (-2,+2), which translates to an increase of about $490 million based on the average market capitalization of firms in our data set. Notably, despite the substantial financial commitments implied by carbon neutrality, this corresponds to an average gain of about $75 in abnormal return per current ton of CO2-equivalent emissions. Second, our analysis of announcement-related media coverage using topic modeling reveals that investors respond more positively when the media highlights the expected business impact of the carbon goal. Third, firms with higher current emissions in the scopes they commit to eliminate are evaluated more favorably by the stock market, suggesting that investors value the larger emission reduction volume associated with such commitments. Finally, the positive effects of emphasizing business impact and promising to eliminate higher current emissions diminish over time. We conduct a series of robustness checks to ensure that our results are reliable and unbiased. Overall, this study provides a strong empirical foundation for understanding the consequences of voluntary carbon goal announcements of firms and discusses implications for theory and practice.
Despite their potential to advance financial inclusion and improve business outcomes, digital payment solutions have low adoption rates among retail merchants in developing economies. Prior interventions have tackled prepurchase frictions such as awareness and affordability, showing limited success. We instead examine postpurchase frictions in the merchant onboarding journey, focusing on installation complexity and value uncertainty. We implement a randomized field experiment with 479 cash-only retailers in Mexico, testing interventions modeled on customer success management (CSM)—a marketing function widely used by business-to-business technology companies to onboard and retain clients, but not previously examined causally. In the first treatment group (T1: n = 159), we tackle installation complexity by introducing a CSM to support system integration for a provided digital payment solution. In the second treatment group (T2: n = 162), we further tackle value uncertainty by adding a second CSM to support value realization. Relative to the control group (n = 158), long-term adoption increased by 21.4 percentage points in T1 and by an additional 13.4 points in T2. Mechanism evidence shows that system integration primarily operates on the extensive margin of ensuring a functioning device in-store and substitutes for missing digital capacity, which constitutes merchants’ digital savviness and the local installed base of cardholders. By contrast, value realization operates on the intensive margin of usage and complements existing digital capacity. Cost-benefit analyses and spillover analyses support the scalability of our interventions. These findings offer insights into promoting the diffusion of digital payments. This paper was accepted by Raphael Thomadsen, marketing. Funding: This research was supported by the Stanford King Center on Global Poverty and Development, the World Bank’s Latin America and the Caribbean Gender Innovation Lab (LACGIL) with the Umbrella Facility for Gender Equality (UFGE), and the MasterCard Center for Inclusive Growth. Further support is acknowledged from the Leonard L. Berry Chair in Services Marketing at Mays Business School. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.04703 .
We propose a new Fast and Simple Elicitation procedure (FSE) for the measurement of decision models. Our procedure bounds the function of the decision model, iteratively halves the maximal distance between the bounds, and incrementally restricts the feasible space of associated parameters. It requires no distributional assumptions and relies on linear programming and approximating splines, improving tractability and descriptiveness. We apply FSE to elicit the probability weighting function in three studies: a simulation and two experiments. Our results demonstrate FSE’s ability to faithfully recover the function. Both in the laboratory and online, FSE reflects the choices made by a respondent more precisely than alternative elicitation procedures and standard functional forms. Importantly, FSE also predicts the out-of-sample choices of a representative online sample more accurately than its alternatives. Taken together, our results shed new light on the prevalence of possibility and certainty effects, and convey general implications for experimental modeling and design. This paper was accepted by Manel Baucells, behavioral economics and decision analysis. Funding: This work was supported by INSEAD R&D and by a grant from Sorbonne University. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2025.01348 .
We examine the role of second-tier stock exchanges in an entrepreneurial context from a global perspective. We construct a country-industry-year panel of global venture capitalist (VC)-invested startups, and find a significant increase in entrepreneurial activity following the launch of such markets. A causal interpretation is supported by additional tests for staggered difference-in-differences regressions and interacted regressions based on industry-specific dependence on second-tier stock exchanges. We propose a VC exit mechanism that receives empirical support: the launch of second-tier stock exchanges positively predicts (i) VC investment in startups, (ii) VCs’ valuation of startups, (iii) startups’ patenting records, (iv) startups’ initial public offering (IPO) likelihood, and (v) startups’ IPO valuation. We find that more startups are created in countries with more VC investment after the launch of such markets. We also find that the effect of second-tier stock exchanges on entrepreneurial ventures increases with financial resources, human capital, and intellectual property protection. This paper was accepted by Matt Marx, entrepreneurship and innovation. Funding: H. Gao acknowledges financial support from the National Natural Science Foundation of China [Grant 72472028]. P.-H. Hsu acknowledges research support from the National Science and Technology Council in Taiwan [Grants NSTC 113-2410-H-007-008-MY3 and NSTC 114-2410-H-007-016-MY3], the Mack Institute for Innovation Management of the Wharton School at the University of Pennsylvania, and the E.SUN Academic Award for financial and research support. Y. Wang acknowledges financial support from the National Natural Science Foundation of China [Grant 72402130]. This work was also supported by the Dalian Commodity Exchange. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.08619 .
This paper empirically investigates how exposures of an industry’s cash flows and stock returns to discount rate shocks are affected by the market concentration of its upstream and downstream industries. In the cross-section of U.S. industries, we find that industries’ cash flows and stock returns are more negatively exposed to fluctuations in the aggregate discount rate if their upstream or downstream industries are more concentrated. These industries also have higher expected stock returns and costs of capital. Our study highlights the role of vertical competition in determining firm risk exposure and expected returns. This paper was accepted by Lukas Schmid, finance. Funding: K. C. J. Wei received partial financial support from the Research Grants Council of the Hong Kong Special Administrative Region, China [Project 15510222]. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2024.04715 .
This article may be used only for the purposes of research, teaching, and/or private study. Commercial use or systematic downloading (by robots or other automatic processes) is prohibited without explicit Publisher approval, unless otherwise noted. For more information, contact permissions@informs.org. The Publisher does not warrant or guarantee the article's accuracy, completeness, merchantability, fitness inclusion of an advertisement in this article, neither constitutes nor implies a guarantee, endorsement, or support
The class of fixed-fraction rules for sharing sequentially triggered losses has been axiomatized using a set of axioms that include continuity. In this note, we show that continuity is redundant because it is implied by the other axioms. This allows for a more parsimonious characterization and thus reducing complexity. This paper was accepted by Manel Baucels, behavioral economics and decision analysis. Funding: T. Oishi acknowledges financial support from the Japan Society for the Promotion of Science [KAKENHI Grant-in-Aid for Scientific Research (C), Project 24K04913]. E. Gavilán and C. Manuel acknowledge financial support from the Complutense University of Madrid [Grant PR27/25-32473].
