Skip to content

Lift Test

A lift test is an experiment that measures lift, the extra conversions or sales a campaign causes, by comparing a group that could see it with a control group that could not. Lift is reported as a difference or as a ratio to the control group's result.

By , Founder & CEOUpdated 6 min read

Run the numbers for your store: the free holdout test planner.

What is Lift Test?

Lift is the gap in an outcome between the group that could see a campaign and the control group. Google's Conversion Lift reports it two ways: incremental conversions, also called absolute lift, which is treatment conversions minus control conversions, and relative lift, which is incremental conversions divided by control conversions. Meta's help center describes a lift test as an experiment that compares people who were given the chance to see the advertising with people who were not, and says people stay in their assigned group for the whole test.

Meta's lift tests count everyone who was eligible, not only the people who saw an ad. Meta calls this an intent-to-treat design, with a randomly assigned test group that includes people who never see the ad and a control group that is not eligible, and it scales the holdout to the size of the test group before comparing. Google says its tool compares all conversions between the experiment and control groups over the whole study and, unlike its standard reports, does not apply conversion tracking settings or attribution rules. Both help centers therefore treat lift as a different quantity from attributed conversions.

There are several kinds. Meta lists Conversion Lift tests, for sales, leads and engagement objectives, and Brand Lift tests, for awareness objectives. Google offers Conversion Lift based on users and based on geography; the geography version compares regions that show the ads with regions that do not.

Lift is a ratio, so it needs care. Google warns that relative lift is easy to misread across studies, because a very small control baseline makes large relative lifts easy to get. The word also has a separate meaning in data mining, where lift scores a targeting model by dividing the response in a target group by the average response. That is not an experiment on a campaign.

For illustration, suppose the test group has 45,000 people who place 990 orders and the control group has 5,000 people who place 100 orders. Scaling the control group to the size of the test group gives 100 x 45,000 / 5,000 = 900 orders. Absolute lift is 990 - 900 = 90 orders, and relative lift is 90 / 900 = 10 percent.

Why Lift Test matters for ecommerce

Ad platforms and GA4 attribute orders to ads and clicks; a lift test is the experimental way to find out how many of those orders a campaign added. Both Meta and Google say lift results are computed differently from attributed conversions, so the two should not be compared as if they measured the same thing. Lift tests have costs and limits. Withheld audiences are not shown the ad, which Google calls an opportunity cost. Platform tools set eligibility rules, and a small store may not qualify or may get a result too noisy to use: a Quarterly Journal of Economics paper by Lewis and Rao reports that measuring advertising returns is difficult because individual sales are very volatile relative to the cost of the advertising. A low-certainty result does not settle that the ads fail: Google says it means the study could not detect lift with high confidence.

How to use Lift Test

  1. State the question and one outcome before launch, such as orders, and choose a window long enough to cover the lag between seeing the campaign and ordering.
  2. Choose the design: a platform lift tool if your account qualifies, a geo split by region, or a holdout on a list you control.
  3. Check the feasibility or power estimate if the tool shows one, and choose a holdout share that balances data against lost exposure. A larger holdout gives more data; a smaller one needs a longer study.
  4. Launch and leave the campaign alone. Do not change creative or audiences mid-study, and note any promotions or outages in the window.
  5. Read absolute lift first, then relative lift, then the interval or certainty. Convert to cost per incremental conversion (spend divided by absolute lift) and incremental ROAS (incremental value divided by spend).
  6. Check the result. Pass: the control stayed unexposed, the study reached its planned size, and the interval supports a budget decision. Fail: low certainty, a thin control baseline producing a large relative lift, or a campaign changed mid-test. Then rerun it longer or larger rather than concluding the ads do not work.

Formula

Absolute lift = conversions in test group - scaled conversions in control group

Relative lift = absolute lift / scaled conversions in control group

Cost per incremental conversion = ad spend / absolute lift

Incremental ROAS = incremental conversion value / ad spend

Where:
- conversions in test group: conversions among everyone eligible to see the campaign
- scaled conversions in control group: conversions in the withheld group multiplied by test group size / control group size
- ad spend: spend on the campaign during the study
- incremental conversion value: revenue in the test group minus scaled revenue in the control group

Lift is an estimate, so read it with an interval or a certainty level.

Common mistakes

  1. Comparing relative lift across studies without looking at the control baseline. Google says very large relative lifts occur easily when the control group has few conversions, so a large percentage can sit on a small absolute gain.
  2. Reading a low-certainty or no-lift result as evidence that the ads do not work. Google says such a result does not necessarily mean the ads were ineffective: the study may not have collected enough data to detect lift.
  3. Judging the test before the conversions arrive. Google recommends waiting until the end of the study for the most accurate lift results and setting the duration to capture the average conversion lag.
  4. Testing only some of the campaigns in a channel and reading the result as the channel's value. Google says holdback users might still see unmeasured campaigns, which reduces the chance of detecting lift.
  5. Changing creative or audiences during the study. Google says this might impact the performance of the study and makes it hard to learn what is working.

Frequently asked questions

  • Is a lift test the same as an incrementality test?
    In the platform documentation read for this entry, they describe the same kind of experiment. Google calls Conversion Lift an incrementality tool, and Meta says running a lift test is a way to understand the incremental effect of its advertising. Neither page draws a line between the two terms.
  • What does a low-certainty or no-lift result mean?
    It means the study could not detect lift with high confidence, not that the ads failed. Google says this happens when too little data was collected or the lift is small, that a no-lift label does not necessarily mean the ads were ineffective, and that it recommends setting up a new test.
  • Why do lift results differ from the conversions in my ad reports?
    They are computed differently. Meta says lift tests count all conversions in the test and holdout groups, while Ads Manager shows conversions attributed to specific ads in specific attribution windows. Google says Conversion Lift does not apply the usual conversion tracking settings or attribution rules.

Further reading

Your platforms guess.
We run the math.

Upload a GA4 export and see what each channel caused, next to last-click, in 1–2 minutes. The read is yours to keep.

Get your first finding free

Free, in your browser: your file is not uploaded. The full read is €99, refundable within 30 days. Prices exclude VAT.
Or book a 30-min call.