Create and manage experiments
An experiment compares personalization variants against a control by splitting visitor traffic between them, so you can tell which one actually performs better instead of guessing. Run one before making a change permanent.
Set up an experiment
- From Personalization, create an experiment on the decision or recommender you want to test.
- Keep the current experience as the control, then add one or more variants (for example, a different recommender or point configuration).
- Set the traffic allocation that splits visitors across the control and each variant.
- Choose the success metric the experiment should optimise for (for example, clicks, conversions, or revenue).
Run and act on results
- Start the experiment so it begins serving the control and variants to live traffic.
- Let it run until it reaches enough visitors and time to be meaningful, not just until a variant looks ahead.
- Review the results per variant against your chosen success metric.
- Promote the winning variant to 100% of traffic, or end the experiment if no variant clearly wins.
Discussion
Comments are not configured for local development. They appear once the site owner connects GitHub Discussions (Giscus).