Campaign experiment
A campaign experiment is a Google Ads test that splits traffic between a campaign and a changed version of it over the same dates to show what the change does.
How it works
On the Experiments page you pick a base campaign and a change, such as the bidding strategy, target ROAS or Performance Max against Standard Shopping. Google runs the original (control) and the changed version (treatment) at the same time and splits traffic between them. Both versions face the same season, competitors and stock, which a before-and-after comparison cannot promise.
Google advises at least 4 weeks to cover weekly cycles and conversion delays. A campaign runs one experiment at a time. For each metric, Google gives the gap a p-value and a confidence interval, which show its statistical significance.
Our guides cover experiments for bid strategies and a typical store test: PMax against Standard Shopping.
Example
Example store, not client data.
The tableware shop tests a new bidding target with a 50/50 split, and each version spends 20,000. Control brings 150 orders worth 90,000: ROAS 4.5. Treatment brings 140 orders worth 92,000: ROAS 4.6. Its ROAS is 2% higher, but orders are 7% fewer.
How GetProfit reads it
The portal changes nothing in the account itself, so its change log compares each change before and after. A confident verdict comes at 14 days. “Worked” means ROAS rose more than 20% and conversions fell no more than 20%. “Harmed” means ROAS fell more than 20% or conversions by more than half.
Each verdict notes that season and assortment may have played a part. An experiment rules them out with its control.
Not to be confused with
- A/B test — two versions of a website page shown to different visitors. A campaign experiment splits ad traffic between two versions of a campaign.
- Incrementality — sales that would not happen without ads at all. Most experiments compare two versions of advertising, not ads against none.
Right and wrong readings
- Wrong: “Treatment has the higher ROAS, so the change worked.” Right: in the example, orders fell 7%. Until Google marks the gap as significant, it can be noise.
- Wrong: “Treatment led after the first week, so we can stop.” Right: with automated bidding, Google advises ignoring the first 1–2 weeks.
Sources
- Experiments overview — Google Ads API: control and treatment, traffic split, experiment types. Checked 2 October 2026.
- Report on experiments — Google Ads API: significance, 4 weeks, ramp-up, one per campaign. Checked 2 October 2026.
- Set up Standard Shopping campaign vs. Performance Max experiments — Experiments page, traffic split. Checked 2 October 2026.