Understanding video advertising effectiveness is essential, given advertisers’ substantial investment in the format and its ubiquitous presence in our daily lives. But understanding video ad feature effectiveness is challenging due to the limited availability of video ad data and the complexity of video features.
We therefore collected video ad and performance data on more than 90,000 video ad campaigns, registering 1 trillion impressions, across more than 20 industry segments, on six popular social media platforms, including Facebook, YouTube, SnapChat, LinkedIn, Twitter and Pinterest. We establish a taxonomy of video ad features, based on the Elaboration Likelihood Model (ELM), using unsupervised clustering to explore how different features cluster across our sample. We then perform a feature importance analysis, using group Lasso and Random Forest models, and employ multilevel linear models of video feature effects on ad performance.
We find, for example, that the presence and early appearance of text in videos reduces view-related performance metrics, while the presence and early appearance of people improves view-related metrics. We also explore the heterogeneity of video ad feature performance effects across platforms, industries and campaign objectives. As predicted by ELM, for example, text reduces ad performance in the luxury industry and increases performance in professional services ads.
To our knowledge, ours is the largest analysis of video ad features and their performance implications to date. We hope to provide a foundation for future causal studies of video ad features and their performance effects.
