Causal impact of digital display ads on advertiser performance
(A2022-108183)

AUTHORS

Koen Pauwels, Amazon Ads; Manuele Caddeo, Amazon Ads; German Schnaidt, Amazon Ads

ABSTRACT

Brands are searching for innovative ways to reach customers online. Sponsored Display (SD) by Amazon Ads is a new way to do so, and allows targeting by category, product and audience. However, advertisers are uncertain how much SD improves their performance over different time horizons. This paper studies more than 40,000 brands with two different methods: a diffusion-regression state-space time-series analysis that predicts response counterfactuals during a 20-weeks period post SD adoption of audience targeting, and a newly developed Two-stage Gaussian Process algorithm that generates probabilistically-equivalent twins for causal inference in a shorter time frame of 1 month post SD adoption of category and product targeting. The performance variables include impressions, page views, sales, new-to-brand consumers and Return on Advertising Spend. The results are consistent and quantify how much adding SD to the ad mix increases performance.