Best for
- Teams running campaign experiments or Performance Max experiments
- Advertisers making decisions from small or incomplete tests
- Agencies that need standardized experiment readouts
Audit native experiment setup, control and treatment arms, traffic split, conversion actions, runtime, metric movement, and confidence before deciding whether to adopt, reject, extend, or redesign a test.
A defensible Google Ads experiment readout, adoption decision, learning register, and prioritized next-test roadmap.
Quick view
Check the audience, core tools, and access before you start the detailed steps.
Best for
Core tools
2 tools used across the workflow.
Access
Preview the open steps, then unlock the remaining implementation details and prompts.
Why this works
A campaign experiment is valuable only when the hypothesis, control, treatment, traffic split, conversion actions, and decision rule are documented before interpreting results. Google Ads exposes the experiment arm, dates, split, performance metrics, and confidence indicators, but a positive directional result may still be operationally weak or confounded. Claude can reconcile setup quality with outcome evidence and preserve inconclusive findings as learning rather than forcing a winner. The next-test plan compounds knowledge instead of repeating disconnected optimizations.
Expected results
Records or configurations reviewed
100% of the approved in-scope population
The run reconciles every eligible record or configuration item to the signed source manifest rather than relying on an informal sample.
Evidence validation
Stratified QA before action
Every major finding class and high-impact segment is checked against source records before operational changes are approved.
Decision output
One owner-ready action register
Findings are converted into deduplicated actions with evidence, confidence, owners, approvers, deadlines, and rollback requirements.
Operational reuse
Versioned recurring runbook and Claude Skill
The same inputs, rules, prompts, schemas, validation gates, and metrics can be rerun while preserving a visible change history.
Step-by-step workflow
The first 2 steps are open. Pro unlocks the remaining steps, copy-paste prompts, pro tips, tool-by-tool setup guidance, and implementation details.
30-60 min
30-60 min
Document hypothesis, primary metric, guardrails, changed variable, control, treatment, split, planned runtime, minimum detectable effect, and stop conditions. Work from Google Ads using experiment ID, base campaign, trial campaign, experiment split as the minimum evidence set. Complete this work in Google Ads or the controlled working file; no Claude prompt is needed for this step. Save the finished artifact in the experiment decision and next-test brief with the run date, owner, evidence reference, confidence, and approval status. Treat budget caps as a separate exception class and do not count it as failure unless the policy says so.
Recover the pre-test decision record completed as a dated section of the experiment decision and next-test brief, with experiment ID, incremental conversions, evidence links, owner, and approval status for graduate the winner.
Do not let experiment ID stand in for base campaign; that shortcut creates false positives in campaign experiment readout. Document the result in the same run folder so the next cycle can compare like with like. Apply it specifically during “Recover the pre-test decision record.”
30-60 min
30-60 min
Check experiment status, dates, campaign eligibility, traffic split, conversion actions, budget, bidding, geography, audiences, and change history. Mark invalid comparisons before reading results. Capture trial campaign, experiment split, start date, end date in a dated working table before interpreting the result. Complete this work in Google Ads or the controlled working file; no Claude prompt is needed for this step. Save the finished artifact in the experiment decision and next-test brief with the run date, owner, evidence reference, confidence, and approval status. Quality-check the result against arms ran for the planned window, then route any contradiction to the named data owner.
Verify experiment integrity completed as a dated section of the experiment decision and next-test brief, with base campaign, cost-per-conversion delta, evidence links, owner, and approval status for extend the experiment.
Keep auction shocks visible as its own class because merging it into the main failure rate will distort the decision. Document the result in the same run folder so the next cycle can compare like with like. Apply it specifically during “Verify experiment integrity.”
Pro workflow preview
Previewing 2 of 12 steps
Get the remaining 10 steps, copy-paste prompts, pro tips, tool-by-tool setup guidance, and ongoing workflow and prompt releases.
$9/month
Related workflows
Continue with workflows that share a similar GTM motion, category, or tool stack.
A Performance Max asset-group scorecard, evidence-backed asset backlog, and approved restructuring or test plan.
An approval-ready budget and structure change plan with scenario assumptions, guardrails, and monitoring checkpoints.
Create a campaign postmortem, hypothesis log, and prioritized next-test roadmap for paid media.