D. J. Wu

Points are a common feature of digital learning platforms, but they are typically used as symbolic feedback rather than as redeemable assets. We examine whether converting points into a functional currency improves student learning behavior, and whether the design of the redemption mechanism matters. In a randomized field experiment conducted with a large K--12 education provider in China, students in all groups earned points for learning activities, while treated students could redeem points for non-monetary digital collectible cards through one of three mechanisms: store, gacha, or time-limited redemption. Using matched difference-in-differences analysis on 7,359 student-week observations from 388 ninth-grade students, we show that point redemption increases assignment completion and error correction relative to a points-only control. The gacha design produces the strongest and most consistent effects. Mechanism analyses suggest that redemption increases motivation by giving points functional value, and that gacha outperforms deterministic designs because uncertainty prolongs pursuit of preferred rewards. The gains are concentrated among lower-performing students, female students, and non-STEM subjects. Our findings highlight point redemption as a consequential design lever in gamified education and show that the effectiveness of gamification depends not only on whether rewards exist, but also on how they are redeemed.

In software markets, the sheer number of available applications make it rather challenging for any given new one to stand out and be noticed by consumers. Moreover, a push towards privacy by regulators and consumers is making it harder to target consumers. As such, firms have to rely on more non-targeted go-to-market strategies. We explore two popular strategies through which developers can catalyze adoption by helping consumers directly or indirectly learn the value of their products-seeding (free full-feature product giveaways to a subset of the consumer base) and time-limited freemium (TLF). Seeding, as a business strategy, existed for a long time. On the other hand, the feasibility to offer market-wide TLF became mainstream more recently, with the advent of the Internet and a plethora of digital tools. Thus, a natural question emerges-if TLF represents nowadays a feasible and easily implementable strategy for software applications, has seeding approach been rendered irrelevant in these markets? In this study, we provide managerial recommendations on when each of these strategies with a free full-feature-consumption component is optimal, based on social and self-learning dynamics, consumer priors, adoption costs, and individual product value depreciation. To that end, under a multi-period parsimonious unifying framework, we show that S becomes dominated as free trials enter the picture. We identify two specific market factors that, when present, can induce seeding to be optimal when consumers initially underestimate true product value - (i) user adoption costs and/or (ii) individual depreciation of value by usage. Moreover, we show that these two factors have a moderating effect on the impact of word-of-mouth (WOM) effects on the optimality of seeding. In the absence of these factors, stronger WOM effects alone cannot give seeding an edge against the other business strategies. However, once either depreciation or adoption costs are accounted for, strong WOM effects increase the relevance of seeding (enlarging its optimality region in the parameter space). Our results remain qualitatively consistent under a battery of robustness checks.

Despite the popularity of the phrase “wisdom of the crowd,” not all crowds are wise because not everyone in them acts in an informed, rational manner. Identifying informative actions, therefore, can help to isolate the truly wise part of a crowd. Motivated by this idea, we evaluate the informational value of investors’ bids using data from online, debt-based crowdfunding, in which we were able to track both investment decisions and ultimate repayment statuses for individual loans. We propose several easily scalable variables derived from the heterogeneity of investors’ bids in terms of size and timing. We first show that loans funded with larger bids relative to the typical bid amount in the market or to the bidder’s historical baseline, particularly early in the bidding period, are less likely to default. More importantly, we perform theory-driven feature engineering and find that these variables improve the predictive performance of state-of-the-art models that have been proposed in this context. Even during the fundraising process, these variables improve predictions of both funding likelihood and loan quality. We discuss the implications of these variables, including loan pricing in secondary markets, crowd wisdom in different market mechanisms, and financial inclusion. Crowdfunding platforms can easily implement these variables to improve market efficiency without compromising investor privacy. This paper was accepted by David Simchi-Levi, information systems. Funding: We gratefully acknowledge the financial support of the Business Analytics Center in the Scheller College of Business and the Ewing Marion Kauffman Foundation. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.02575 .

“Shoppertainment” live streaming, a blend of entertainment and product selling, is becoming increasingly popular. In shoppertainment sessions, streamers face bandwidth constraints so balance between providing product information and entertainment (such as singing, dancing, and storytelling). We conceptualize a shoppertainment live streamer as a novel type of online retail platform that maximizes total commissions from selling a range of manufacturer products. Our study investigates how streamers can leverage entertainment bandwidth as a long-term strategy to influence manufacturers’ pricing decisions and consumers’ attendance, as well as purchasing behaviors. Our model reveals several intriguing findings. First, we find that higher entertainment bandwidth encourages more manufacturers to adopt a demand-oriented, low-pricing strategy to attract buyers rather than a margin-oriented, high-pricing strategy that appeals only to buyers who perceive a high likelihood of finding a suitable product. Second, we observe that an increase in entertainment bandwidth initially boosts consumer traffic to shoppertainment sessions, but this trend may reverse once the bandwidth becomes excessively high, indicating an inverted U-shaped relationship. Conversely, the streamer’s expected profit per attending consumer first decreases and then increases with higher entertainment bandwidth, forming a U-shaped relationship. Third, our results suggest that a commission-maximizing streamer facing bandwidth constraints should not exclusively focus on providing product information. Instead, it is advantageous to allocate significant bandwidth to entertainment, even with limited entertaining capability, and potentially beneficial to dedicate all bandwidth to entertainment if the capability is substantial. Lastly, we demonstrate that the streamer’s quality threshold moderates the impact of entertainment bandwidth on these dynamics. This paper was accepted by David Simchi-Levi, information systems. Supplemental Material: The online appendix is available at https://doi.org/10.1287/mnsc.2023.01724 .

This commentary reflects on insights from a panel discussion at the 2024 Annual MIS Academic Leadership Conference, where six senior MIS scholars discussed the impact of Generative AI on scholarly research and peer review. The discussion underscored the importance of responsible use, transparency, and ethical standards, as well as the irreplaceable role of human judgment in maintaining research integrity. This commentary explores the potential of Generative AI as a collaborative tool across various stages of the research lifecycle, highlighting the "human-in-the-loop" approach to harness AI's capabilities while preserving essential human insight. This commentary synthesizes the senior scholars’ perspectives on the responsible integration of Generative AI, emphasizing opportunities to enhance research efficiency and foster interdisciplinary collaboration, while advocating for policies that ensure AI supports—rather than substitutes—human intellectual contributions in academic research.

This paper proposes a smart ad display system to provide personalized delivery of video ads. The proposed system records consumers’ facial expression and eye gaze stream data as they watch an ad and analyzes data at the frame level. The recognized facial expression and detected eye gaze are matched to the corresponding frame of the video ad, thereby linking facial expressions to specific visual objects appearing in the ad. By tracking a consumer’s facial expressions in response to various visual objects in real time, the system learns the consumer’s individual preferences toward different ads, searches the ad pool, and selects and subsequently displays a new ad that is most likely to elicit positive attitudinal and behavioral responses. We demonstrate the feasibility and effectiveness of the proposed system with two empirical studies. The results show that by tracking a consumer’s facial responses to only one ad or even part of an ad, our proposed system is able to make reasonably accurate inferences about a consumer’s ad preferences, with or without using information about other consumers. These inferences are used to make personalized recommendations that help enhance consumers’ ad viewing experiences and elicit favorable responses.

Interpretable deep learning is increasingly central to high-stakes business applications, as stakeholders cannot act on predictions they cannot verify. Yet the two dominant paradigms, intrinsic methods (e.g., attention values) and post-hoc methods (e.g., Shapley values), routinely produce inconsistent and even contradictory explanations for the same model on the same data. We trace this tension to a structural problem we term \emph{interpretation misalignment}, and prove with four impossibility results for attention values and three inferiority results for Shapley values that neither method in isolation can yield reliable interpretations. To resolve this, we propose an \emph{Interpretation Alignment Framework} that exploits a mutually beneficial relationship in deep learning: more accurate intrinsic interpretations enable a better-calibrated prediction model, which in turn yields more accurate post-hoc interpretations. Unlike prior approaches that rely on externally specified targets, our framework derives the alignment target endogenously and iteratively co-optimizes prediction and interpretation tasks. Theoretically, our framework is guaranteed to converge and mitigate the issues in existing interpretation methods. Empirically, we apply the framework to a face perception task across seven business-to-consumer contexts, where it significantly outperforms state-of-the-art baselines on both prediction accuracy and interpretation quality; an illustrative cost-benefit analysis translates the gain into \$160{,}000 of additional net benefit, and a downstream fairness-aware classification model built on these interpretations achieves significant improvements on accuracy and fairness simultaneously. Finally, generalizability analyses on two additional image-classification tasks and an NLP task confirm that the alignment principle holds across modalities, establishing it as a flexible, interpretable deep learning framework in practical business applications.

We examine the productivity implications of external knowledge flows obtained through an internet-mediated discussion forum in which IT professionals help one another solve problems related to the implementation and use of enterprise software. We extend elements of the absorptive capacity (ACAP) framework that have not previously been studied in the information systems (IS) literature to a new context. Consistent with prior results from the IS literature, we first show that IT spillovers—acquired through employees’ participation in this forum—only accrue to firms with prior related investments in enterprise software. We then demonstrate boundary conditions for ACAP based on characteristics of external knowledge affecting the ease of learning. Our results show that IT spillovers are not “free”; the ability to derive the value of IT spillovers through informal channels—such as online communities—critically depends on both prior related IT investments by the recipient firm and the novelty of external knowledge. Less intuitively, when knowledge originates from relatively novel or emergent domains, the role of prior related knowledge in absorbing spillovers becomes more important.