Professor Puneet Manchanda, Professor of Marketing at the Ross School of Business, University of Michigan will give us a visit and be our guest speaker at our PhD Seminar in Quantitative Marketing Research.

His most recent research work has focused on consumer and firm behavior in digital markets. He is a thought leader, currently holding the Senior Editor position at Marketing Science, and is a frequent speaker at various academic and industry events.

Finding the Sweet Spot: Ad Scheduling on Streaming Media (joint work with Prashant Rajaram and Eric Schwartz)

In 2018, 55% of US households subscribed to at least one video streaming service. Not surprisingly, ad spending on such services is growing at a rapid pace. In contrast to linear TV, on-demand streaming services allow consumers to consume content in a non-linear manner and/or not on fixed temporal schedules e.g., via binge-watching. There is little research that has investigated the role of advertising in these settings. Our objective is to develop an “optimal” advertising schedule that maximizes ad exposure without compromising the content consumption experience. We use a novel dataset that tracks the viewing behavior of 10,000+ Hulu viewers for over 4 months. Our research approach has three stages. In the first stage, we use findings from the consumer psychology literature as motivation for the development of parsimonious metrics – Bingeability and Ad Tolerance – to capture the interplay between content consumption and ad exposure in non-linear consumption settings. Specifically, Bingeability represents the number of completely viewed unique episodes of a TV show while Ad Tolerance captures the willingness of a viewer to watch pods (block of ads) and to watch content from the TV show after being exposed to pods in a session. In the second stage, we use detailed data from a moving window of one-week of viewing activity along with a rich description of the ad delivery process (captured via pod length, pod frequency and ad diversity) to predict these metrics for a given viewer at a given time. The prediction is carried out on holdout samples using a tree-based machine learning algorithm, Extreme Gradient Boosting. We control for the non-randomness in ad delivery to a focal viewer using instrumental variables based on ad delivery patterns for other viewers. In the third stage, we use the predicted metrics to identify sessions where ads enhance, or at least do not detract from, content consumption. We then use a novel constrained optimization procedure to provide an optimal advertising schedule for the streaming provider that maximizes ad exposure via frequency of pod delivery (holding pod length constant). We conclude by discussing the benefits of our recommendations for the streaming platform and the viewer.

Keywords: Advertising Response, Advertising Scheduling, Advertising Metrics, Streaming Media, Binge-Watching, Machine Learning, Optimization