Testing and iteration
Test one meaningful page change
Turn a shopper uncertainty into a controlled experiment with one unchanged control and one verified challenger.
Install Lexsis MCP and skills in your AI client
Run the installer, then follow the authentication prompts to connect Lexsis MCP and add the workflow skills to your supported AI client.
curl -fsSL https://mcp.trylexsis.com/install.sh -o /tmp/lexsis-install.sh && sh /tmp/lexsis-install.shWhat this workflow does
This workflow begins with an existing Lexsis page.
The agent inspects the page and its performance, identifies one meaningful uncertainty, creates a focused test plan, and builds a challenger without changing the control.
It does not start live traffic automatically.
What you need
- Existing landing-page URL
- Primary success metric
- Audience and traffic source
- Testing window or traffic expectations
- Elements that must remain unchanged
- Optional initial hypothesis
Starting prompt
Replace the bracketed details, then give the complete prompt to your connected AI agent.
Create a controlled A/B test for this Lexsis page: [URL]
Primary outcome: [METRIC]
Audience and traffic source: [DETAILS]
My initial hypothesis: [OPTIONAL]
Preserve unchanged: [ELEMENTS]
Inspect the current page, mobile and desktop experience, analytics and source. Identify the most important shopper uncertainty and recommend one focused change to test.
Default to one control and one challenger. Show me the hypothesis and test plan before duplicating or editing anything.
After I approve the plan and duplication credit, build and verify the challenger. Do not create or activate the experiment until I approve the final variant.How the workflow works
Follow the sequence in order. Approval and verification remain separate from page creation and live publication.
- 01
Inspect the control
The agent reads the current page, source, version, analytics and responsive experience.
- 02
Identify one uncertainty
Possible test areas include:
- Hero message
- Offer framing
- CTA
- Proof placement
- Product explanation
- Section order
- Mobile purchase controls
- 03
Write the hypothesis
The hypothesis identifies:
- Observed evidence
- Shopper uncertainty
- Controlled change
- Expected behaviour
- Primary metric
- Guardrail metrics
- 04
Approve the test plan
You approve the hypothesis, control, challenger and measurement approach before duplication.
- 05
Confirm duplication cost
The agent checks available Lexsis credits and asks for approval before creating the challenger.
- 06
Build the challenger
The control remains unchanged.
The agent duplicates the page and applies only the approved controlled change.
- 07
Verify the challenger
The challenger is compiled and checked for:
- Source integrity
- Responsive layout
- Copy accuracy
- Product behaviour
- Cart behaviour
- Variant behaviour
- Unexpected differences from the control
- 08
Approve the variant
You review the challenger preview before the experiment is created.
- 09
Create a draft experiment
Lexsis creates the experiment configuration in a draft state.
Publishing pages or starting traffic requires separate explicit approval.
- 10
Evaluate results
Once sufficient traffic has been collected, the agent can read experiment performance and explain whether the result supports, rejects or fails to resolve the hypothesis.
Scaling a winner remains a separate approved operation.
How Lexsis MCP is used
Each operation stays tied to the selected workspace, store, page and version for this workflow.
- Find the editable control page
- Read the page and source
- Read analytics and experiment history
- Check available credits
- Duplicate the control
- Compile and update the challenger
- Run integrity checks
- Create the draft experiment
- Read experiment results
- Scale a winner only after approval
What you receive
The result includes the working artifacts and verification evidence needed for the next decision.
- Test hypothesis
- Control definition
- Challenger specification
- Measurement plan
- Guardrail metrics
- Verified challenger page
- Preview URLs
- Draft experiment
- Experiment ID and state
- Remaining activation steps