Experiments

Design prerequisites

Use an enabled Search campaign, a fresh account sync and completed audit within the last day, confirmed business facts, and a saved account-manager brief with measurement review. Resolve audit blockers first.

Declare the hypothesis, exact change, future dates in the advertiser timezone, conversion lag, combined-spend review threshold, minimum conversions, and minimum useful absolute conversion gain. The current design supports a 50/50 split, conversions as the primary metric, 28–90 days, and at least 30 conversions per arm as a screening floor. The floor is not a statistical power calculation.

Approval and execution are different

Approving a design freezes it; no Google change is executed by design approval.

The native treatment changes positive location targeting from presence-or-interest to presence only. Other treatment designs are tracked externally. The native workflow has separate prepared, reviewed, approved, dispatched, and verified stages; every stage is subject to current account access, paid-access checks, and the global write restriction.

Live native execution is not generally released. Do not interpret available design controls or test fixtures as proof that a real experiment was launched.

Read and assess results

For an external design, link the numeric Google experiment ID. Entryway checks account, Search experiment type, dates, control campaign, two arms, and a 50/50 split before saving the comparison. Native designs may only read their own verified scheduled experiment.

Refresh results, or enable collection every six hours for a linked experiment. Each experiment has its own saved schedule, retry errors and pause control. Collection reads evidence without changing serving or budgets. Entryway saves the provider observation and uses Google-attributed conversions, combined cost, absolute-change estimate, margin of error, and p-value. Missing values stay unknown.

It waits until the scheduled end plus conversion lag. Insufficient data and inconclusive results remain valid outcomes. A candidate needs the predeclared practical gain and statistical checks; treatment CPA is also compared with the approved brief. Confirm experiment integrity before recording the lesson.

Limits that matter

The spend value is a review threshold, not an automatic spending cap. Entryway does not automatically stop an experiment when it is exceeded. Review serving and budgets in Google Ads.

The app does not automatically promote a winner, guarantee qualified leads or causal business lift, run continuous sequential testing, or train a shared model from customer data. Lessons are saved account-scoped evidence for subsequent analysis.