Show me the money: Understanding causality for ad attribution

Often marketing teams lack data-driven performance metrics to assess ad campaign performance. To know whether an ad campaign has an effect, they need to tie marketing activities to specific business outcomes. Attribution modeling is the collection of techniques they use to estimate the relative contribution of different ad campaigns to sales, signups, or brand favorability. Traditionally, these teams use click-through rate or last-touch attribution to measure outcomes. However, these techniques have known issues.

Take for example an ecommerce company that’s using digital advertising to drive online sales. While simultaneously running 10 different pieces of creative through multiple digital ad vendors, they realize that prospective customers go through a complicated journey of ad exposures before making a purchase or disappearing from the system. As expected, they drop the least effective creatives and buy more ads for the top performers. But how do they make this decision?

The ideal approach to making this decision would be to run a randomized controlled trial (RCT). In an RCT, you randomly show ads to one group of people and not another and measure the difference in the outcomes between the two groups. Unfortunately, running an RCT is often expensive, disruptive, or impossible. Drawing on case studies from two of the biggest digital and television advertisers in the country, April Chen shares an approach to attribution modeling that measures ad campaign performance more accurately than industry standards. April details common approaches, highlighting their shortcomings, and walks you through how she uses techniques from nonexperimental causal inference to simulate an RCT: leveraging optimal matching techniques to construct pseudotreatment and pseudocontrol groups in order to approximate an ideal experiment and measure the treatment effects of ad campaigns.