Skip to main content
After you generate a variation in the PBX builder, click Prepare for launch to enter PBX Configure. This article walks you through each configuration step and explains how to launch the experiment. For help generating a variation, see Create a prompt-based experiment. To open PBX Configure, click the cog icon in the PBX sidebar. To return to the prompting view, click the pencil icon.

Configure targeting

In the targeting step, select who to expose to the experiment. Choose one of the following options:
  • All visitors: Exposes every visitor to the experiment.
  • Specific segment: Lets you choose an existing segment from your project.
  • Build your segment with AI: Lets you describe your audience in natural language so PBX generates the targeting conditions.
The "Who will be exposed to this experiment?" panel with the "Build your segment with AI" option selected.
To build a segment with AI, select Build your segment with AI, click Apply, and describe your intended audience. For example:
  • Mobile users
  • Returning visitors
  • Users from the United States
  • Visitors who viewed a product page
You can reference existing Kameleoon segments by name in your prompt. PBX checks for duplicates against existing assets in the project scope and reuses a matching asset by name rather than creating a new one. A checkbox lets you save the created segment for future reuse across the project. Without saving, the segment is scoped to this experiment only. If PBX cannot confidently interpret part of your prompt, it will ask for clarification before generating the segment.

Configure triggers

In the trigger step, choose when to activate the experiment. Choose one of the following options:
  • When webpage is reached: Activates the experiment when a visitor lands on the URL you specified during experiment creation. PBX pre-populates this URL.
  • When specific trigger occurs: Lets you choose from existing triggers in your project.
  • Build your trigger with AI: Lets you describe the activation condition in natural language.
The "When it needs to be triggered?" panel with "When webpage is reached" selected and the experiment URL displayed.
To build a trigger with AI, select Build your trigger with AI, click Apply, and describe when to activate the experiment. For example:
  • Visitors who spend more than 30 seconds on a product page
  • Users who return to the website
  • Visitors who scroll past the pricing section
PBX generates the trigger configuration based on your description. You can reference existing Kameleoon triggers by name in your prompt. PBX checks for duplicates against existing assets in the project scope and reuses a matching asset by name rather than creating a new one. A checkbox lets you save the created trigger for future reuse across the project. Without saving, the trigger is scoped to this experiment only. If PBX cannot confidently interpret part of your prompt, it will ask for clarification before generating the trigger.

Combine segment and trigger instructions in one prompt

When you provide a prompt that contains both audience and activation instructions, PBX identifies what belongs to the segment and what belongs to the trigger, then splits the prompt across both configurations. PBX handles this bifurcation regardless of which step (segment or trigger) you are currently in. The following scenarios describe how PBX responds to different prompt inputs:
  • Your prompt contains a mix of segment and trigger conditions: PBX walks you through the segment step first, then moves to the trigger step with each pre-filled based on your initial prompt. You do not need to re-enter your instructions.
  • Your prompt contains only conditions that belong to the other category: For example, if you enter trigger conditions while in the segment step, PBX skips ahead directly to the correct step rather than requiring you to navigate there yourself.
  • Your segment is already defined, but a subsequent prompt in the trigger step affects the segment: After you finalize the trigger, PBX returns you to the segment step so you can review and confirm the additional conditions it identified.
For example, if you enter the following prompt:
Target USA visitors who are browsing a page /shop on desktop, and who have not converted the Add to Cart goal.
PBX assigns each condition to the correct configuration step:
  • Segment: USA-based visitors on desktop devices.
  • Trigger: Page URL contains /shop, and the visitor has not converted the Add to Cart goal.
Review both the segment and trigger configurations after PBX processes your prompt to confirm that each condition lands in the expected step.

Configure traffic allocation

In the traffic allocation step, choose how PBX distributes visitors between the control and the variation. Choose one of the following allocation methods:
  • Manual: Assigns a fixed percentage of traffic to each variation.
  • Multi-armed Bandit: Allocates more traffic to better-performing variations automatically over time.
  • Contextual Bandit: Allocates variations based on visitor attributes and real-time performance data, serving the most relevant variation to each visitor.
Use the Exposed traffic control to set the percentage of eligible visitors who participate in the experiment.
The "Set traffic allocation" panel showing Manual, Multi-armed Bandit, and Contextual Bandit tabs, with 100% exposed traffic and a 50/50 Control/Variation 1 split.
After you configure traffic allocation, click Apply Traffic.

Configure goals

Goals determine how Kameleoon measures experiment performance. In the goals step, you can:
  • Search for and select one or more existing goals from your project.
  • Create a new goal manually by clicking + Create a new goal. For detailed instructions, see Create a goal.
  • Build a goal with AI by selecting Build your goal with AI and describing the conversion event you want to track in natural language.
Configure an AI goal.
To build a goal with AI, click Apply, then describe the action you want to measure. For example:
  • Visitors who click the Add to Cart button
  • Visitors who reach the order confirmation page
  • Visitors who submit the newsletter signup form
PBX generates the goal configuration based on your description. Review the generated goal before you apply it. You can use the on-screen toolbar in the PBX preview zone to add visual context to your goal prompt. Click or rectangle-select page elements to identify them, which makes the generated goal output more accurate. Goals are implemented as custom JavaScript. The goal contains the JS code generated and, once created, can be used in this experiment as well as in other experiments if needed. You can reference existing Kameleoon goals by name in your prompt. PBX checks for duplicates against existing assets in the project scope and reuses a matching asset by name rather than creating a new one. If PBX cannot confidently interpret part of your prompt, it will ask for clarification before generating the goal. After you select your goals, click Apply Goals.

Configure integrations

In the integrations step, select the integrations that receive experiment data. The available integrations depend on your project configuration. Toggle on each integration you want to receive experiment data, then continue to the review step.

Review the experiment configuration

PBX displays a full summary of your experiment configuration before launch.
The "Experiment ready!" summary page showing Targeting, Trigger, Traffic, Goals, and Integrations, with Simulate, Schedule, and Launch buttons.
Review each configured item. To make changes, click the relevant section to return to it. When the configuration is correct, choose one of the following options:
  • Simulate: Preview how the experiment behaves before you expose it to real visitors.
  • Schedule: Set a future start date and time for the experiment.
  • Launch: Start the experiment immediately.

Launch the experiment

Click Launch to activate the experiment. PBX begins tracking performance in real time.
The post-launch state showing a green "Results will be available soon" toast and an "Experiment launched" confirmation with a rocket illustration.
After launch, experiment data appears in reporting as visitors are exposed to the variation, and configured integrations begin receiving experiment information. To view results, go to Experiences > A/B Test, open the experiment, and click Results.