Incrementality testing
Conversion Lift study / lift test / holdout test
Incrementality testing is a controlled experiment that splits an audience into a group that can see the ads and a group held back from them, then compares their conversions. The difference estimates the sales or leads the ads caused, apart from the attribution rules behind the conversions an ad account reports.
A Conversion Lift study splits the audience in two. People in the treatment group can see the ads, and people in the control group are held back. Incremental conversions are treatment conversions minus control conversions. Google says this count ignores the attribution rules behind reported conversions.
In a study based on users, incremental cost per action is total ad spend divided by incremental conversions. Incremental ROAS is incremental conversion value divided by spend. Google warns that relative lift is easy to misread across studies, because a study with a low control count reaches high relative lift more easily. Studies based on geography report incremental conversions, conversion value, ROAS and cost.
Google says lift results fall within a confidence interval. It recommends aiming for 90% certainty of lift, says results between 50% and 90% can be used as directional insights, and leaves the right level to each advertiser's needs and risk tolerance.
In a study based on users, the holdback can be set from 1% to 50%, and it must stay at 30% alongside Brand Lift and Search Lift. A larger holdback gathers more data but hides the ads from more people. A smaller one needs a longer study.
Google says a study should capture the average conversion lag, the time from impression to conversion. It allows studies as short as 7 days, and short purchase cycles can run 7 to 14 days. It recommends at least 14 days, and more for expensive products. It has found up to a 17% drop in absolute lift when studies with a long lag ran under 14 days.
Conversion Lift is not available on every Google Ads account. In a study based on users, conversions must fire unconditionally, and store visits, store sales and offline imports without personally identifiable information are unsupported. That study type cannot be saved below a $5,000 USD budget. Google says a budget above $5,000 USD with 1,000 conversions gives access to directional results.
Studies based on geography, which Google describes as a beta, compare regions that show the ads with regions that do not. Campaigns must target a single country, and offline conversions aggregated to ZIP or city level go through the Google account team. Travel between test and control regions lowers the reported incrementality.
In practice
A home services company arranges a Conversion Lift study based on users for its Search campaigns through its Google account representative. It spends $25,000, and the account reports 1,250 conversions at $20 each. The study estimates 500 incremental conversions at 90% certainty, an incremental cost per action of $50. With $40 of profit per job before ad costs, each extra job lost about $10 on average. The figures are a worked example.
Not the same as
- Attribution model
- Attributed conversions in an ad account are counted under the tracking settings and attribution rules of each conversion action, such as click or view-based attribution windows. Google Conversion Lift ignores those rules and compares all conversions between people who could see the ads and people held back.
Why it matters to you
A lift test estimates the leads and sales the business would lose without the ads. Google suggests testing before major budget decisions, and notes that every study carries an opportunity cost.
What to ask or check
- 01Is Conversion Lift available on your Google Ads account, and would your study based on users reach the $5,000 USD minimum budget?
- 02Is your conversion volume high enough for the study power estimate to reach 90% certainty?
- 03What holdback share will your study use, and will it run at least 14 days and cover your typical time from ad to sale?
- 04Does the result show incremental conversions with a confidence interval or certainty level, or only relative lift?
- 05For a study based on users, do your conversions fire unconditionally, and do your offline lead imports include personally identifiable information?
What people get wrong
That the conversions in an ad account are the conversions the ads caused. Google says Conversion Lift ignores the attribution rules behind those reports and measures the difference in all conversions between people who could see the ads and people held back.
Red flags
- A lift result quoted only as relative lift, with no incremental conversions or incremental cost per action.
- A lift figure quoted as one exact number, with no confidence interval or certainty level.