We develop an algorithm to predict the effect of influencer video ads on product sales. We propose the motion concept, which captures the pixel-level spatio-temporal interaction between content engagement and product placement in a video. We estimate content engagement as a saliency map by fine-tuning a deep 3D convolutional neural network on video-level engagement. We locate product placement by matching images of the product to each frame of the video with an object detection algorithm. Analogous to a fundamental law in Newtonian mechanics, motion (sales) is generated when the object (product) is impressed with a force (engagement) in the space and time (video). We validate the algorithm with an analysis of 40,000 influencer video ads on Douyin, the Chinese version of TikTok and the largest short video platform in the world. We leverage variation in video posting time to identify the causal effect of video ads on product sales. Videos of higher motion are indeed more effective in driving sales. This effect is sizable, robust, and is more pronounced among impulsive, hedonic, and lower-price products. We trace the mechanism to influencers’ incentives to optimize for engagement rather than sales. We discuss how product sellers can use motion to screen video ads pre-launch in a scalable way and also as a novel contractual lever to mitigate the agency problem for greater ROI.
This event is open to faculty, doctoral students, academic researchers, and graduate students.