- Optimizely X Web Personalization
THIS ARTICLE WILL HELP YOU:
- Track how well personalized experiences perform, compared to the original
- Set or change the holdback
Every Personalization campaign has a “holdback,” which is like the control of an A/B test. Optimizely's Stats Engine uses the holdback to help you track how well your personalized experiences perform, compared to the generic experience.
By default, Optimizely Personalization shows 95% of visitors in a campaign a personalized experience, and holds back 5% who will see the original.
Visitors are bucketed into the holdback by chance.
Holdbacks are persistent. Once a visitor is bucketed into the holdback, they'll always see the holdback experience for that campaign.
Make the holdback larger to get data faster, or shrink it if you’re very confident. We recommend 5%, so you can always monitor how well your personalization is doing.
Optimizely integrates testing and personalization, so you can be confident that the experiences you deploy generate lift. In Personalization, consider directing the majority of your traffic to the personalized experience (versus the holdback) to maximize lift. If you haven't thoroughly A/B tested your experiences yet, consider setting a higher holdback to get significance more quickly.
Set the holdback
When you create a campaign, set the holdback in the final step.
A larger holdback will help you gather data about statistical significance faster, so you know the impact of your campaign. A smaller holdback allows you to show personalized experiences to more of your visitors.
Change the holdback
Optimizely sets one holdback per campaign. So, no matter how many experiences or variations you add to a campaign, the holdback will stay the same.
Here's how to change the holdback for an existing campaign.
Navigate to Campaign dashboard and select the campaign. Under Manage Campaign dashboard, click Settings.
Click to edit the holdback and hit Save.
To reset the holdback, select Revert.
Changing the holdback for a running campaign is not currently supported by Stats Engine. If you choose to do this, you may see statistical reports that are invalid.