Multi-Objective Personalization of the Length and Skippability of Video Advertisements

In this paper, we study two features of digital video ads on content-streaming platforms: length and skippability. Working with vdo.ai, we conduct a field experiment and randomly assign users to the Skippable/Long and Non-Skippable/Short versions of the same ad. We find that compared to the Non-Skippable/Short ad, the Skippable/Long ad version in our study increases ad consumption but decreases video consumption. This substitution pattern between ad and video consumption leads to a challenge for platforms seeking to maximize both outcomes. To address this challenge, we develop algorithms for multi-objective personalization that use individual-level substitution patterns to optimize ad and video consumption. The results show that multi-objective personalized policies can significantly improve both ad and video consumption outcomes over single-objective policies. In particular, we show that compared to a single-objective policy optimized for video consumption, there exists a multi-objective policy on the Pareto frontier that increases ad consumption by 61% at the expense of only a 4% decrease in video consumption. Similarly, compared to the single-objective policy optimized for ad consumption, there is a multi-objective policy that increases video consumption by 47% while decreasing ad consumption by just 13%. We conclude by discussing the practical implications for platform decision-making in real time