How do I test if my ads are incremental?
Hold your ads back from a random slice of people or regions and compare total sales with everyone else. Use Meta's or Google's lift study if your account qualifies, or a regional holdout if not. If both groups sell about the same, the ads were collecting sales, not causing them.
By Joris van Huët, Founder & CEOUpdated 7 min read
Usually by holding the ads back from a random slice of people or regions and comparing total sales with everyone else. Use a lift study inside Meta or Google Ads if your account qualifies, or run a regional holdout yourself. If both groups sell about the same, the ads were collecting sales, not causing them.
A click report can't settle this. It shows who touched a sale, not whether the sale needed the touch. Only a comparison with people who saw no ads does that, and you can build one three ways.
| Test | Who makes the groups | What it needs | Weak spot |
|---|---|---|---|
| Platform lift study | Meta or Google, at random, person by person | Enough spend and conversions to qualify | Purchases it can't match to someone in its groups |
| Regional holdout | You, by switching ads off in some regions | Many regions with steady sales | Small effects, because regional sales are noisy |
| Switch-off over time | Nobody: you compare weeks | A calendar | Everything else that changed that week gets mixed in |
The platform route has an entry ticket. As a guide, Meta asks for a campaign that started in the past year with a spend of $5,000 USD or more. That campaign also needs at least 500 conversions under a 1-day click, 7-day click or 1-day view setting. In Google Ads, the study based on users is available for Video, Discovery and Demand Gen campaigns. Other types, such as Search and Shopping, go through an account representative, and Google says Conversion Lift isn't available for all accounts. A regional holdout you run yourself needs no ticket, though Google notes geo studies tend to need a bigger budget than user-based ones.
What one store's data shows
One store's anonymised GA4 export, 1 January 2024 to 21 August 2026. It holds shares of revenue only: no ad spend, no order counts.
| What the export shows | Share of revenue | Source cell |
|---|---|---|
| Journeys with 1 touch (0.5 days to buy) | 79.5% | Journeys sheet, 1 touch row |
| Journeys with 2 to 3 touches (12.5 days to buy) | 12.2% | Journeys sheet, 2 to 3 touches row |
| Journeys with 4 to 9 touches (16.9 days to buy) | 5.4% | Journeys sheet, 4 to 9 touches row |
| Direct, in last click, first click and touched views | 57.7% | Channels sheet, Direct row |
Start with the 79.5% in that store's Journeys sheet: journeys with 1 touch, at 0.5 days to buy. With one touch on the path, every attribution model hands that touch the whole sale. First click, last click and data-driven agree, because they have one candidate. Agreement here is not evidence.
That is where a test earns its keep. Whatever ad sits on those quick, one-visit sales gets full credit under every rule, and no report can say whether the buyer needed it. If your own one-touch purchases arrive through brand search or retargeting, those are your first candidates.
The slower journeys set the clock. In the Journeys sheet, journeys with 2 to 3 touches took 12.5 days to buy, and 4 to 9 touches took 16.9 days. A test that ends after one week scores the quick buyers and misses the considered ones. Google has found up to a 17% drop in Absolute Lift when studies with a long conversion lag run for less than 14 days.
Direct holds 57.7% of revenue in all three views of that store's Channels sheet. Click reports can't say what sent those buyers. A lift study doesn't need to know, because Meta's lift tests count all conversions in the test and holdout groups. An ad that works quietly shows up as a gap between the groups, even when GA4 files the sale under Direct.
What the export can't do is run the test. It holds no spend and no order counts, so it can't tell you whether you clear Meta's entry ticket. Nor can it say how long your test must run to see a small effect. It can point the test. It can't replace it.
Why does switching the ads off for a week mislead?
It is the cheapest test there is, and it leaks in four places.
- No control group. You compare this week with last week, and everything else moved too: payday, weather, a promotion, a rival's sale. A dip can't be pinned on the ads, and neither can a rise.
- Buyers straddle the pause. People who saw your ads before it keep buying during it. The sales the pause loses turn up later, after you have stopped looking.
- The restart costs you. Meta counts pausing an ad set for 7 days or longer as a significant edit. Once unpaused, the ad set re-enters learning, so your comeback week measures the restart, not the ads.
- A model is not a test. Meta's incremental attribution setting uses machine learning models that predict whether a conversion is caused by an ad. Handy for steering delivery. It is still a prediction, not a holdout.
Which ads should you test first?
Test where a wrong answer costs the most: the biggest line with the least evidence behind it.
Google's own geo guide frames the question well. Does a campaign drive incremental sales, or just capture organic conversions? Capture is likeliest where buyers were already on their way: brand search, retargeting, and any ad on the last click of a one-touch path.
Test the whole channel before a single campaign. Meta suggests starting with an account level test that looks at the overall effect of your advertising. Google adds that measuring only a subset of campaigns reduces your chances of detecting lift.
And test one channel at a time. Pause Meta and trim Google in the same weeks, and any change in sales belongs to both. Which means it belongs to neither.
What to do this week
- See which channel owns your one-touch sales. In GA4, click Advertising, then Key event attribution paths under the Key events dropdown. Filter Path length to equal to 1 touchpoint. Pass: a paid channel owns a large share of those purchases, so it is your first test. Fail: nearly all are Direct, so test the paid channel you suspect feeds Direct.
- Find a quiet window in Shopify. Go to Analytics > Reports, filter the Category by Sales and open Total sales over time. Group it by week over the past year. Pass: a run of steady weeks, longer than your buyers take to decide, with no sale or launch. Fail: every month has a spike, so compare groups during the test, never weeks.
- Ask Google which campaigns it will test. In Google Ads, open the Lift studies tab under Campaigns > Experiments and select the plus button. Choose Conversion Lift, then Based on users. Pass: you can select your main campaigns for the study. Fail: you can't, or they are Search or Shopping campaigns that need your account representative, so plan a regional holdout.
Check the homework. Your GA4 Attribution paths export already holds the evidence. Causality Engine reads that one file and shows what each channel caused next to what last-click gave it, in 1 to 2 minutes, for €99 once (excluding VAT), refundable within 30 days. Check the homework
Sources, 1 October 2026: About Conversion Lift (Meta Business Help Center); Comparing lift types (Google Ads Help); Set up Conversion Lift based on users (Google Ads Help); Differences between Conversion Lift test results and other reporting tools (Meta Business Help Center); Significant edits and learning phase (Meta Business Help Center); About incremental attribution (Meta Business Help Center); Implement campaigns for geo experiments (Google Ads Help); Best practices to get started with Experiments (Meta Business Help Center); Key events attribution paths report (Google Analytics Help); Sales reports (Shopify Help Center).
Related answers
Frequently asked questions
Can I test incrementality by pausing my ads for a week?
You can, but read the result with care. A one-week pause has no control group, so a payday, a promotion or the weather can move sales more than the ads did. Pause in some regions only, run longer than your buyers take to decide, and judge it on total store sales.Should I test one channel at a time?
Yes, usually. Change one thing per test so the gap has one cause. If two channels change in the same weeks, a shift in sales can't be pinned on either. Run the tests one after the other, or give each channel its own set of regions.Is Meta's incremental attribution the same as a lift test?
No. Meta says incremental attribution uses machine learning models that predict whether a conversion is caused by an ad. A Conversion Lift test withholds ads from a random control group and counts what each group buys. Use the model to steer delivery and the test to check it.
Go deeper: Causal attribution, explained.
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Keep reading
Terms in this article
- AttributionAttribution identifies user actions that contribute to a desired outcome and assigns value to each. It reveals which marketing touchpoints drive conversions.
- Attribution ModelAn Attribution Model defines how credit for conversions is assigned to marketing touchpoints. It dictates how marketing channels receive credit for sales.
- Control GroupControl Group is a segment of an audience intentionally not exposed to a marketing campaign, used to measure the campaign's true causal impact.
- Google AnalyticsGoogle Analytics is a web analytics service that tracks and reports website traffic.
- Holdout TestA holdout test is an experiment where a portion of the audience does not see a campaign. This measures the campaign's true incremental impact.
- IncrementalityIncrementality measures the true causal impact of a marketing campaign. It quantifies the additional conversions or revenue directly from that activity.
- Incrementality TestingIncrementality Testing measures the additional impact of a marketing campaign. It compares exposed and control groups to determine causal effect.
- Machine LearningMachine Learning involves computer algorithms that improve automatically through experience and data. It applies to tasks like customer segmentation and churn prediction.