Ad platforms such as Google and Meta have dominated the digital advertising market for years. However, brands have long been concerned about the advertising efficacy on ad platforms, especially given the skyrocketing advertising costs in recent years. In this study, we examine whether the advertising effect is underestimated when not accounting for the advertising spillover from ad platforms. In a large-scale randomized experiment on ByteDance in collaboration with an automobile brand, we show that exposed users are eight times more likely to convert outside than within ByteDance, which leads to a dramatic underestimation of the advertising effect if ignoring advertising spillover. In addition, advertising spillover affects firms’ targeting strategy. While the commonly used demographic variables by the automobile brand are effective for target marketing with only platform data, they are not when considering the conversions outside ByteDance. Instead, a behavioral variable proposed herein (prior brand home page visits) effectively moderates the advertising effect but has no effect when including only platform data in the analysis. Finally, when considering advertising spillover, the average cost per conversion, which brands commonly use to evaluate the cost of campaigns, shrinks by 5 or 25 times, depending on the methods used to calculate it.
