Research

My research focuses on digital marketing and AI. I examine how platform design and access to technological resources shape participation, production, and consumer value.

Research Interests

Substantive: Platform Economics; Creator Economy and Governance; AI and Consumer Behavior

Methodological: Causal Inference; Empirical Industrial Organization; Machine Learning

Working Papers

Monetizing the Long Tail through Direct Audience Support: Platform Design and Creator Responses

with Kai Zhu

Job-Market Paper ยท Major Revision at Marketing Science

Digital platforms increasingly allow audiences to support creators directly, yet little is known about how expanding access to this support changes creator behavior. We study Twitch's 2017 Affiliate Program, which extended channel subscriptions and virtual donations to smaller creators. Using a time-shifted difference-in-differences design, we find that the program increased weekly streaming hours by 42%. Relative responses were strongest among smaller creators, broadening the distribution of content production. Per-stream viewership did not detectably decline despite the increase in supply. Monetization access also changed creators' content strategies. Together, these findings identify expanded access to direct audience support as a platform-design lever for cultivating the long tail.

Earlier version on SSRN

When New Users Stop Arriving: Replenishment and Incumbent Activity on Matching Platforms

with Kai Zhu, Qiaoni Shi, and Sara Valentini

Matching platforms rely on continual acquisition not only for entrants' own activity but also to renew incumbents' opportunities. Using daily data for major dating apps across 100 Chinese cities, we study Tantan's 2019 suspension from major Android app stores, which sharply restricted acquisition while largely preserving incumbent access. Bayesian synthetic controls show that registrations fell by 56%, the active contribution of recent cohorts by 64%, and incumbent activity by 16%. Null pre-suspension placebos, the cumulative depletion of recent cohorts, and larger incumbent declines in thicker pre-shock markets support interrupted replenishment over direct access loss, pre-existing trends, or a uniform reputation shock. The closest casual-dating rival gained registrations, whereas the marriage-oriented Jiayuan did not, indicating differentiated substitution. Under the replenishment exclusion restriction, the ratio of the incumbent and recent-cohort reduced forms implies a fresh-to-incumbent elasticity of 0.25: a 10% decline in fresh-user activity reduces incumbent activity by about 2.5%. A cohort-feedback model uses this estimate as a benchmark calibration and shows how acquisition shocks propagate through cohort retention, incumbent participation, and later acquisition. The results show that acquisition sustains the effective matching pool and incumbent participation, not merely the number of registered users.

Work in Progress

The Value of Domain-Specific Data for AI

with Qi Wang and Kai Zhu

Developing a field experiment with Boohee to study the value of domain-specific data in AI-enabled services, comparing specialized AI, general-purpose AI, and no AI support.

The Price of AI: Compute Costs, Experimentation, and Creative Production

with Kai Zhu

Examines how compute costs shape experimentation, creative output, and the profitability of AI-assisted production in short-drama markets.

Copyright Enforcement and Content Supply on Twitch

with Kai Zhu

Investigates how copyright enforcement affects content production and creator participation, including potential chilling effects beyond directly targeted creators.