Predict-Then-Prove: An Explainable AI-based Ad Analyzer for Pre-Flight Winning Creative Identification

This paper discusses an AI ad analyzer that provides clear explanations of its reasoning for why it thinks a specific piece of advertising (the “creative”) will perform better than others. It also gives recommendations on how to continue optimizing your creatives as they are running in real time. The most pressing issue facing the advertising industry today is the inability for advertisers to accurately and reliably predict how well their ads will perform in advance of spend. Most current solutions focus on two areas - optimizing ad delivery or reporting on results after-the-fact rather than using data to accurately and consistently predict how well a creative will perform before running a campaign and providing insights into the key attributes of the creative that are most likely to impact its success. The Analyzer uses both visual and textual information from the advertisements and employs prediction modelling to predict important campaign metrics (such as the click-through rate (CTR), conversion rate (CVR), and return on ad spend (ROAS)). The Analyzer includes visual explanation tools (such as SHAP), continuous evaluation to optimize campaigns and all operating to protect consumer privacy. Previous academic and industry studies have demonstrated that using this methodology leads to improved accuracy of predictions, enhanced operational efficiencies, and improved ROI on ad spend. The methodology and results also address some of the most important areas of concern related to advertiser biases and brand safety. This research study fills a major gap in the current body of work related to ad analysis.