Personalize AI Visual Editor variations with contextual bandits

Contextual bandits can now serve variations you build in the AI Visual Editor. You design page changes without code, and the bandit decides which variation each user gets based on their context. Contextual bandits also gain a new Health tab.

What changes

Until now, a contextual bandit served its variations through a contextual bandit rule on a linked feature flag. Each variation was a flag value that your code had to read. With this release you can:

• Build variations in the AI Visual Editor and let the bandit pick one for each user context.
• Review the bandit on a new Health tab.

On a marketing page, that means you can try headline, banner, or call-to-action variations without waiting on a deploy. The bandit still learns separate weights for contexts such as country or device.

Setup notes

• Visual contextual bandits need the JavaScript or React SDK 2.0.0 or later.
• The SDK Connection that serves your site needs Include Visual Experiments enabled.
• The setup guide recommends adding the optional leaf_id, bandit_version, and variation_weights columns to your assignment query for health checks like SRM.

Getting started

Contextual bandits remain a beta feature on Enterprise plans. You create a contextual bandit under Experiments → Contextual Bandits. The contextual bandit setup guide walks through the tracking callback and the assignment query. Create the page variations you want to test in the AI Visual Editor.