Predictive programmatic re-targeting to improve website conversion rates

Abstract

Abstract In the era of programmatic advertising, the advertisers have huge amount of first party data to leverage on enabling them to do highly granular re-targeting. Programmatic re-targeting is the ability to use data to show an ad to a user who has demonstrated an interest in your product offerings before. Re-targeting ads are a powerful conversion optimization tool and are typically known to outperform conventional targeting in terms of performance. As per 99 Firms, 41% of marketing allocation in 2018 to paid display spend was on re targeting and for most of the websites, only 2% of web-traffic converts on the first time visit. In this paper, a conversion is referred as a purchase made and a converting user is one who made the purchase on the website. The question that arises is - “should we be re-targeting all the users who have landed on the site?”. In ad campaigns which has low budgets or in campaigns where the conversion rate is really low even though a huge volume of users visit the site, it may not make complete sense to simply re-target all those users, instead we would want to re-target those who are clearly showing an intent to make a purchase either through their on-site browsing behaviors or their past conversion patterns. Through this paper we present the use of first party privacy preserving data to do predictive programmatic re-targeting of users who are going to make a conversion in the next few days given their past site-browsing and conversion behavior using a structured data science and advanced ML based framework. Additionally, this project allows to tie the model results to real time programmatic activation by the creation of user segments depending on whether the user is going to make a conversion for the first time, or is converting again. The final outputs are these user segments, which are going to be used by in house ad-traders who would be able to bid deferentially for a specified period of time against each of the segments on a demand side platform. We have successfully tested this model on 2 advertising clients and were able to capture 80-85% of the actual converts happening over the next few days of them landing.