My major research interests lie in building theoretical models about the use of technology (e.g., artificial intelligence) in online platforms to study its impact on customers and firms. In my research, I focus on topics such as content moderation, influencer marketing, gig economy, etc., and try to find a novel aspect to illustrate some unexpected and understudied impact of adopting technology. 

Publications/Accepted papers:

Marketing Science, 2022

Abstract: This paper develops a theoretical model to study the economic incentives for a social media platform to moderate user-generated content. We show that a self-interested platform can use content moderation as an effective marketing tool to expand its installed user base, to increase the utility of its users, and to achieve its positioning as a moderate or extreme content platform. For the purpose of maximizing its own profit, a platform will balance pruning some extreme content, thus losing some users, with gaining new users because of a more moderate content on the platform. This balancing act will play out differently depending on whether users will have to pay to join (subscription vs advertising revenue models) and on whether the technology for content moderation is perfect.

We show that when conducting content moderation optimally, a platform under advertising is more likely to moderate its content than one under subscription, but does it less aggressively compared to the latter when it does. This is because a platform under advertising is more concerned about expanding its user base, while a platform under subscription is also concerned with users' willingness-to-pay. We also show a platform's optimal content moderation strategy depends on its technical sophistication. Because of imperfect technology, a platform may optimally throw away the moderate content more than the extreme content. Therefore, one cannot judge how extreme a platform is by just looking at its content moderation strategy. Furthermore, we show that a platform under advertising does not necessarily benefit from a better technology for content moderation, but one under subscription does, as the latter can always internalize the benefits of a better technology. This means that platforms under different revenue models can have different incentives to improve their content moderation technology. Finally, we draw managerial and policy implications from our insights.

Abstract: Recent years have witnessed significant advancements in matching technologies to improve the matching between workers and employers requesting job tasks on a gig-economy platform. While the conventional wisdom suggests that technologies with higher matching quality benefits the platform by assigning better-matched jobs to workers, we discover a possible unintended revenue-decreasing effect. Our stylized game-theoretic model suggests that while a technology’s matching enhancement effect can increase the platform’s revenue, the jobs assigned by the better matching technology can also unintentionally reveal more information about the uncertain labor demand to workers, especially when the demand is low, and thus unfavorably change workers’ participation decisions, resulting in a revenue loss for the platform. We extend our model to the cases where (1) the share of revenue between workers and platform is endogenous, (2) the matching quality can be improved continuously, (3) the opportunity cost of workers is affected by competition between platforms, and (4) workers compete for job tasks. We find consistent results with additional insights including the optimal matching quality the platform should pursue. Furthermore, we examine two approaches to mitigate the potential negative effect of employing an advanced matching technology for the platform and find that under certain conditions, the platform can be better off by revealing the labor demand or competition information directly to workers. Our results shed light on both the intended positive and unintended negative effects of improvements in matching quality and highlight the importance of thoughtful development, management, and application of matching technologies in the gig economy.

Consumer Preferences and Firm Technology Choice (with Pinar Yildirim and John Zhang)

International Journal of Research in Marketing, 2023

Abstract: Advances in technology change the way consumers search and shop for products. Emerging is the trend of home-shopping devices such as Amazon's Alexa and Google Home, which allow consumers to search or order products. We investigate how consumer brand and technology preferences may interact with the functionalities of technology-enabled shopping (TES) devices to determine the channel structure and market competition.

In specific, we break the functionalities of the TES devices into two: (1) the shopping support functionality (SSF), and (2) the ordering convenience functionality (OCF). Via a series of experiments, we document that stronger brand preferences are negatively correlated with the willingness to use a TES device that offers SSF. However, there is no association with brand preferences and desire to use a TES device when it offers OCF. 

We build an analytical model integrating the findings from these experiments, and then derive the equilibrium channel and pricing strategies for two competing retailers. Our findings show that the functionality of TES devices results in vastly different distribution and pricing strategies in retail markets. In particular, consumers' heterogeneous valuation of the SSF results in a monopolistic adoption of TES devices by the retailers in equilibrium, and generates Pareto improvements for both. In contrast, when the TES devices offer OCF, in equilibrium, retailers adopt TES channels competitively, resulting in a prisoner's outcome. 

In the extensions,  studying a third-party technology developer's decision to invest in OCF and SSF technologies, we show that the contrast between the channel strategies under the OCF and the SSF also impact the incentives to develop TES. We show that in some cases, in an effort to mitigate downstream retailer competition, the provider may prefer not to offer the best possible OCF technology to consumers. These findings shed light on the future adoption and the functionalities of shopping technologies offered by retailers.

Abstract: This paper explores the impact of using advanced technology such as artificial intelligence (AI) to match marketers with social media influencers. We develop a theoretical model to examine how matching accuracy affects the competition between influencers and the profitability of a social media platform. Our findings show that improving matching accuracy may not always benefit the platform, especially for platforms with intermediate follower density. Two opposing effects of technology improvement affect the prices of influencer marketing campaigns: advanced technology such as AI enhances the matching between influencers and marketers, but also intensifies competition between different types of influencers. The overall effect on prices can be negative for some influencers due to the asymmetric nature of such matching technology: the matching outcome for influencers with a narrower audience ("niche" influencers) is more sensitive to matching accuracy than that for those with a broader audience ("general" influencers). As a result, more niche influencers begin to participate in marketing campaigns when matching accuracy improves, which reduces the prices offered by sufficiently general influencers and may lead to a decline in platform revenue. Additionally, we found that adjusting commission rates in response to technology improvements could help mitigate the negative impact, although it may not eliminate it entirely. Our findings offer valuable insights for social media platforms seeking to remain competitive in the influencer marketing landscape.

Other working papers:

Talking without Speaking: Paid Trolls on Social Media and Court Decision (with Tony Cui)

Abstract: With the massive growth of social media and other informational platforms that businesses and individuals may use to share their opinions, a society's perspectives of certain contentious issues may be influenced significantly. At the same time, it has been noted that various parties have used paid trolls (i.e., fake comments made by bot accounts) to sway public opinions concerning some hotly debated issues or legal proceedings. This article aims to investigate how public opinions and paid trolls may affect the outcome of legal disputes between opposing parties. Our research indicates that "practice what you preach, unite knowledge and practice" may lose out to "words speak louder than actions" when businesses are involved in lawsuits with social media engagement. Specifically, firms purchase paid trolls to influence public opinions, and they do so more significantly when the truth does not match the prior expectation of the public. We also discover that social media users' sophistication may be a pitfall for the involved parties, including the competing firms and the judge who is adjudicating the case. Additionally, the court may be either constrained or aided to identify the truth by having a tendency to render decisions that align with prevailing public opinions, depending on the stake of the lawsuit.

GenAI Assistance in a Professional Service Market: The Perish of Second Opinion (with Jane Gu and Rachel Chen)

Abstract: We investigate the impact of generative AI (GenAI) adoption in a market of professional services such as medical treatment and financial consulting where the product is a professional opinion. Providers of professional services  often differentiate in their opinions. This allows consumers to benefit from seeking a second opinion, particularly when they are constrained from obtaining advice from a high-expertise provider. AI assistance improves the advice quality of a low-expertise provider, but has minimal impact on the advice quality of a high-expertise provider. Additionally, when two low-expertise providers both adopt AI assistance, their opinions become less diversified, reducing consumers' incentive to seek a second opinion. Considering a vertically differentiated market of professional services where a high-expertise provider and two low-expertise providers serve consumers with heterogeneous quality preferences, our study reveals important strategic implications of this opinion homogenizing effect of AI adoption. First, we show that after service providers adopt AI assistance, the high-expertise provider may enhance profit, despite quality improvement occurring to its low-expertise rivals but not to itself. In contrast, the low-expertise providers always suffer a severe drop in margins, although their total payoffs may increase under a sufficiently large benefit in corporate image associated with AI adoption. Two interesting regions arise: a "lose-lose" region, where AI adoption harms all service providers, and a "win-lose" region, where the high-expertise provider is better off but neither of the low-expertise providers. Second, we show that after providers' AI adoption, low-type consumers always enjoy a surplus gain but high-type consumers may suffer a surplus loss. Total consumer surplus declines if low-expertise providers are highly differentiated in opinions before AI adoption. Lastly, when the two low-expertise providers make endogenous decisions regarding whether to adopt AI assistance, equilibrium may arise where neither of them adopt AI assistance or only one adopts.

Data and Algorithm: Designing Marketplace Analytics for Platform Sellers (with Fei Long)

Abstract: The rise of e-commerce and the abundance of data have spurred AI-powered marketplace analytics, such as competitive intelligence and automated pricing, enabling sellers to make informed, data-driven decisions. Third-party providers (e.g., Jungle Scout) compete with platforms themselves (such as Amazon's brand analytics) in offering marketplace analytics. Yet platforms have different attitudes toward third-party analytics providers and adopt various strategies in sharing data with them, ranging from restrictive to permissive (e.g., permitting data scraping) with some even actively facilitating (e.g., API sharing).  In this paper, we ask why and when an e-commerce platform may benefit from an open data-access policy to accommodate the competing analytics service provided by third-party providers, despite the platform's inherent advantages in data access, and the capability to design its own analytics services for better control over sellers' actions. We also study how platforms should design their analytics services, modeled as an algorithm to predict market competition levels.  We find that platforms may use an over-optimistic algorithm (by downplaying competition) in their own analytics to increase the total revenue on the platform. This may lead to sellers' reluctance to adopt the platform's analytics. When market competition is moderate, this can (but not always) result in a lose-lose situation, prompting the platform to allow data access for third-party analytics providers (e.g., permitting self-scraping). As competition intensifies, the platform may even actively share more data (e.g., via APIs). However, in highly competitive markets, it benefits both the platform to mislead sellers into believing the market is good, and the sellers, to be deluded into this belief. Overall, platforms only gain from an open data-access strategy in markets with moderately strong or weak competition. Finally, our analysis implies that privacy legislation like GDPR, aimed at curtailing platforms' data-sharing practices, may inadvertently hurt consumer surplus.

On the Role of Trailer as a Marketing Research Tool: The Economic Value of the Comments it Generates (with Josh Eliashberg)

Abstract: Trailers are commonly employed in the pre-launch campaigns of new products as an advertising tool to generate awareness and interest among the potential audience. In this paper, we argue that such trailers, whose costs range and are rising, should also be considered as a marketing research tool having additional economic value. The incremental value is driven by the audience comments data that the trailer of a new product generates. The data, once analyzed, can signal the post-release electronic word-of-mouth (eWOM) that will prevail and can affect the new product success, but is uncertain to the product manager a priori. That is, trailers, if released early, have the potential to assist in making more informed pre-launch marketing decisions (specifically, advertising spending), and thus have additional economic values in guiding risk-mitigation strategies. This incremental value should thus be considered in the cost/benefit analysis of the trailer production. This paper quantifies such value by formulating various economic value of information (information value) metrics based on a Bayesian decision-theoretic framework, which are applicable to many product categories. We investigate the metrics both analytically and empirically, employing a dataset of 363 movies released in 2014-2018. The information value is analyzed with respect to the two key dimensions of the uncertain post-release eWOM: volume and valence. We carry out various sensitivity analyses, comparing the magnitudes of information value under different scenarios. Based on our dataset, we demonstrate that the volume of comments generated by early trailers is more critical than its valence and consequently has a higher incremental economic information value, which may be worth $240,000. We also find that the trailer comments of different genres have different economic values, with action movies having the highest.