One of the greatest challenges of A/B testing is how to optimize for the non-average user. This means that although in a certain campaign, most users prefer the winning variation (that's why it's winning), we are still left with a group of users that perform better when served one of the losing variations. This audience/ group can often be substantial in size, and in value.
The question is, how to determine which groups like which variations, in a manner that is scalable.
Our Predictive Targeting engine not only solves this but also does so with zero effort on your end.
The Predictive Targeting engine uses the audiences available to determine which perform better when served a different variation to the winning one.
The engine runs in the background, and once an opportunity is found, a notification is sent by email as well as highlighted in the A/B test report. The opportunity is presented with a description of the audiences to be targeted, the variation to serve to each, and the expected increased uplift.
Example of a Predictive Analytics Opportunity. Although “Most Popular” variation is winning, by serving “Recently Viewed” variation to the “Referral Traffic” audience, there is a predicted uplift of 12%
Acting on a Personalization Opportunity
Let’s look at this specific example. We have an A/B test with three variations. (1) Recently Viewed, (2) Most Popular, and (3) Control Group. The winning variation is Most Popular with a 10.7% uplift in Purchases.
Our predictive engine shows that Referral Traffic audience prefers a losing variation (Recently Viewed) and therefore, by serving it to this audience there is an uplift of 12% that can be achieved.
When clicking on “Apply Personalization”:
- Once “Apply Personalization” is clicked, the current test version will be ended, and as soon as the new settings are saved, a new test will begin
- Two experiences will be left, one for each of the audiences referred to. In this scenario;
- Referral traffic will be served with the “Recently Viewed” variation
- All other traffic will be served with the “Most Popular” traffic
The engine analyzes all your audiences to find the ones that have a significant inclination towards a non-winning variation and by personalizing the experience, will render a significant uplift.
The more audiences available, the more chance of the predictive targeting engine finding an opportunity.
Please sign in to leave a comment.