Google Ads custom experiments compare a trial campaign with its original campaign while sharing traffic and budget under the configured split. Eligibility and setup vary by campaign type, and only one running experiment can be associated with a campaign at a time. Good experiment design begins before the interface is opened.
VISUAL LESSON
What you will learn
- 01Write a testable campaign hypothesis.
- 02Build and QA control and experiment cells.
- 03Apply or stop the result without losing evidence.
ILLUSTRATIVE WORKED EXAMPLE
Plan an illustrative 50/50 experiment
PRACTICAL INTERFACE MAP
Move from hypothesis to apply-or-stop decision
Define population, one major variable, primary outcome, guardrails, minimum duration, sample rule, and decision thresholds.
Choose an eligible original campaign, create the experiment, edit only the treatment, set traffic, budget, dates, and tracking, then test both paths.
Review the same date and attribution window, verify downstream quality, preserve the result, and use the supported apply or end path.
STEP-BY-STEP LESSON
Hypothesis → controlled split → stable run → apply or stop
THE LEAD ATLAS METHOD
Lead Atlas Data can prepare a campaign-specific business-contact list by category, market, and location while the paid team tests one advertising change against a stable control.See how custom list research works ↗Write the experiment card
State the population, current control, one major treatment change, mechanism, primary metric, qualified business outcome, guardrails, minimum runtime, sample or information rule, decision threshold, owner, and prohibited mid-test changes.
Choose a question the eligible campaign can answer, such as a bid-strategy, keyword, audience, creative, landing-page, or other supported change. Avoid bundling several changes that cannot be interpreted separately.
Confirm eligibility and baseline
Check the current supported campaign types, existing experiments, shared budgets, automated changes, conversion actions, values, attribution, audience settings, policy state, billing, seasonality, and traffic. Save the original campaign settings and recent baseline.
Google documents custom experiments for specific campaign types and says only one running experiment can be associated with a campaign at a time. Follow the controls available in the live account.
Build and QA the treatment
Create the custom experiment from the intended original, name it clearly, edit only the planned treatment variable, and set start, end, traffic split, budget behavior, success metrics, and any supported sync option.
Compare every setting in control and treatment, then test URLs, tracking templates, forms, calls, conversion events, values, audiences, assets, extensions, negatives, schedules, location, device, and policy. Log unintended differences before launch.
Run without moving the goalposts
Annotate the launch and monitor eligibility, serving, spend split, traffic, errors, policy, conversion delay, primary outcome, downstream quality, and guardrails. Fix only pre-defined critical failures and document any intervention.
Evaluate the same dates, attribution window, maturity, and metric definition for both cells. Do not stop at the first favorable day, switch primary metrics, or call a low-volume fluctuation a conclusive winner.
Apply, iterate, or stop
At the decision point, review effect size, uncertainty, qualified outcomes, value, margin, operational capacity, and guardrails. Apply the treatment through the supported workflow only if it meets the pre-agreed rule; otherwise keep the control or design a new test.
Deliverable: experiment card, eligibility check, baseline snapshot, control-treatment diff, QA record, traffic and budget settings, intervention log, mature results, quality and guardrail review, apply-or-stop decision, archive, rollback plan, and owner.
THE TAKEAWAY
Test one important hypothesis with stable measurement and pre-agreed decision rules, then apply, iterate, or stop based on business outcomes—not the most flattering metric.OFFICIAL REFERENCES