TikTok is the most popular short video platform in the world with over 500M daily active users. We’ll show three sets of early results using a unique dataset with detailed information on influencer-created advertising videos, user engagement with the video (e.g., play, like, comment, and share), and product page visits and sales on Douyin (the Chinese version of TikTok). By exploiting the differential timing of video posting (treatment) for each treated product, we construct a synthetic control group with similar products that are not yet treated to estimate the causal effect of influencer advertising on product page visits and sales, and calculate influencer ROI. We use methods in computer vision to extract feature embeddings from the videos and show that influencer fixed effect explains more variation in sales than video content. Somewhat surprisingly, user engagement with the video is not predictive of sales, which suggests that it might not be a good idea for brands to choose influencers based on engagement if they want to generate short-term sales.
Jeremy Yang is a doctoral candidate in the marketing group at Sloan School of Management, Massachusetts Institute of Technology. He is broadly interested in the intersection of machine learning and causal inference, social networks, and behavioral economics.