Mobile ad fraud: Empirical patterns in publisher and advertising campaign data

Abstract

Ad fraud has serious consequences for brands. It also contaminates academic research if scholars neglect a significant level of ad fraud in their data. However, only limited theoretical work has addressed this topic, and empirical research is scarce. In this article, we take a first step to document empirical patterns of mobile ad fraud using two datasets. The datasets are commonly available to buyers of advertising services, and the types of ad fraud studied are significant in the advertising market. We identify some app and campaign characteristics correlated with ad fraud, and uncover methods used by fraudsters to conceal the fraud. They often make the ad fraud proportional to the daily traffic but lowering the ratio of ad fraud on high-traffic days. However, when traffic is unstable, they change strategy to use ad fraud to smooth out the traffic. Meanwhile, in advertising campaigns, the fraudsters allocate most part of fraud during the middle of campaign period, an attempt to reduce the risk of being detected. These findings not only help practitioners and academic researchers determine the extent of ad fraud in the data but also provide stylized facts for future research on theoretical modeling of ad fraud.