In many contexts, experiments designed around randomized control trials are not possible. Although this strategy has known limitations, many companies will engage in “natural experiments” to estimate causal effects. Here, Scott Cunningham describes how companies like eBay and Facebook have used observational causal inference to understand the impact of their investments in paid search advertising. He explains how eBay examined the heterogeneity of treatment effects to determine that paid advertising would influence users less familiar with their site. He also reviews a study wherein Facebook compared insights gleaned from an RCT to conclusions based on observational causal inference. After comparing the results of both approaches, Facebook researchers found that false assumptions about conditional independence might have led them to overestimate the causal effects of advertising on revenue when analyzing observational data. Cunningham advises readers to only use observational causal inference when they have an impenetrable understanding of the process by which data is generated.