Can AI help marketers improve video ads, increasing their effectiveness with specific audiences? Adobe Research scientists asked this big question—and have designed a novel technology to answer it.
Advertisers know that the 60-second spot of the broadcast TV era has given way to new expectations across social media, digital TV, and brand websites. An experimental deep learning framework developed by Adobe Research’s Vishy Swaminathan, Saayan Mitra, and Somdeb Sarkhel could help marketers achieve improved results for video ads across all these platforms. It was showcased at Adobe Summit, Adobe’s annual digital marketing event, on March 28, 2018.
This technology taps into several Adobe Research projects—from metadata generation to computing effectiveness—all working together to help create successful video ad campaigns by drawing on multiple deep neural networks.
The Adobe Summit Sneak, led by Swaminathan, shows this tool in action. Swaminathan begins with a TV spot that needed to be rolled out across social media platforms. The video is first enriched with ad-relevant metadata (also known as tags) using deep learning. The generated metadata and extracted features are then fed to another deep learning network trained on past performance of similar content. That step creates an effectiveness or “watchability” score for the ad on various platforms such as Instagram or Facebook.
After that, recommendations to improve the ad’s score—such as an auto-generated video summary of much shorter duration—are sent to advertiser’s creative team for editing, giving the customer creative control and final approval.
The process promises to generating high-performing video ad campaigns across multiple platforms.