Why retargeting ROAS looks better than it is
Retargeting shows ads to people who already visited your store, so the platform can credit the ads with purchases that would have happened anyway. A regional holdout on your own audience measures how many.
Three platforms, three answers, one budget to defend. Your GA4 export already holds the answer, and the first finding is free and never leaves your browser. Get your first finding free
By Joris van Huët, Founder & CEOUpdated 5 min read
Run the numbers for your store: the free break-even ROAS calculator.
Retargeting shows ads to people who already visited your store, a group selected by its interest in buying, so the platform can credit the ads with purchases that would have happened anyway. In a peer-reviewed randomised experiment on one retailer's display retargeting campaign, the ads lifted purchases by 10.5% (Johnson, Lewis and Nubbemeyer, Journal of Marketing Research, 2017). For illustration: a lift of 10.5% means 110.5 purchases where 100 would have happened anyway, so about 9 in every 10 purchases by the people the ads reached were not caused by them.
Why does retargeting report such a high ROAS?
Because of who sees the ads. A retargeting audience is usually people who already visited your store, a selected group that may have bought anyway, and the platform credits the ad when one of them buys inside the attribution window. That is selection, not cause. Lewis and Rao (Quarterly Journal of Economics, 2015) ran 25 large field experiments with major U.S. retailers and brokerages and concluded that selection bias, due to the targeted nature of advertising, is a crippling concern for widely employed observational methods.
Observational estimates also miss by a wide margin when the data is rich. A 2023 follow-up by Gordon, Moakler and Zettelmeyer (Marketing Science) analysed 663 large-scale experiments at Facebook, with access to over 5,000 user-level features, richer data than most advertisers or their measurement partners can access. They compared the median lift in the experiments with the median lift two observational methods estimated for the same campaigns:
| Funnel stage | Experiments | Double/debiased machine learning | Stratified propensity score matching |
|---|---|---|---|
| Upper | 29% | 83% | 173% |
| Middle | 18% | 58% | 176% |
| Lower | 5% | 24% | 64% |
The authors concluded that, despite the experiments and the user-level data, they were unable to reliably estimate an ad campaign's causal effect. Those are Facebook campaigns in general, not retargeting alone. By construction, retargeting selects on prior intent, which is the selection problem they describe.
What did experiments find for retargeting itself?
The ghost-ads paper identified the control-group counterparts of exposed shoppers in a randomised experiment, then applied the method to an online retailer's display retargeting campaign. The ads lifted website visits by 17.2% and purchases by 10.5%. So retargeting can work, but the lift was a small part of the purchases made by the people the ads reached. That is one campaign at one retailer, on display ads. No public measurement of a typical retargeting lift for Shopify brands turned up in the sources read for this page. A retargeting line is the easiest kind for a platform to grade generously.
What does that do to your retargeting ROAS?
Platform ROAS divides the revenue the platform attributes to the ads by spend. Incremental ROAS divides the revenue the ads caused by the same spend. For illustration: a retargeting line spends 1,000 and the platform attributes 4,000 of revenue, a ROAS of 4.0, and a holdout shows that 90% of those purchases would have happened anyway, so the ads caused 400 of revenue and incremental ROAS is 400 / 1,000 = 0.4. For illustration: with a 40% contribution margin, break-even ROAS is 1 / 0.4 = 2.5, so the line looks profitable at 4.0 and loses money at 0.4.
How do you run a holdout on your own retargeting?
Hold retargeting out of a random set of regions for whole weeks and keep everything else running:
- Split regions at random into two groups of similar order volume. Google's own geo experiments randomly assign non-overlapping regions to a control or treatment condition, and the same logic works here.
- Write the decision rule first: the incremental ROAS you need to keep the line, which is at least your break-even ROAS. Use the break-even ROAS calculator.
- Switch retargeting off in the control regions and change nothing else for the test weeks.
- Pull Shopify net sales by shipping region for equal numbers of weeks before and during the test.
- Do the sum. For each group divide revenue during by revenue before. The gap between the two ratios, times the retargeted regions' revenue before, is the revenue retargeting added. Divide it by the actual retargeting spend in those regions during the test weeks.
Pass: incremental ROAS at or above break-even. Fail: below it. Inconclusive: the gap sits inside normal week-to-week variation. In Lewis and Rao's experiments the median confidence interval on return on investment was over 100 percentage points wide, and if retargeting is a small share of your spend, a region-level test can miss it. Read that as not measured, cap the line, and say so. The retargeting holdout design covers how to tell in advance whether your spend is big enough to test.
Sources, 30 September 2026: Ghost Ads: Improving the Economics of Measuring Online Ad Effectiveness (Johnson, Lewis and Nubbemeyer, Journal of Marketing Research, 2017; Crossref record of the publisher's abstract); The Unfavorable Economics of Measuring the Returns to Advertising (Lewis and Rao, Quarterly Journal of Economics, 2015, via EconPapers); Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement (Gordon, Moakler and Zettelmeyer, Marketing Science, 2023, via IDEAS); Measuring Ad Effectiveness Using Geo Experiments (Google Research, vendor-published).
Related answers
Frequently asked questions
Is retargeting worth it?
Sometimes. A randomised experiment on one retailer's retargeting campaign measured a lift in purchases, but most purchases by the people it reached would have happened anyway. Run a holdout on your own audience before you scale it.What is a good incremental ROAS for retargeting?
No public benchmark turned up for this. Compare your holdout's incremental ROAS with your own break-even ROAS, which is one divided by your contribution margin.Why is my retargeting ROAS so much higher than prospecting?
Because retargeting reaches people who already visited your store, so more of them buy with or without the ad, and the platform credits the ad for those purchases. Only a holdout separates the two.
Go deeper: Causal attribution, explained.
Sixty-second versions of these ideas: Causality Engine on YouTube Shorts.
Have an audience of ecommerce operators? Earn 20% recurring commission by recommending the €99 read.
Keep reading
Terms in this article
- Attribution WindowAttribution Window is the defined period after a user interacts with a marketing touchpoint, during which a conversion can be credited to that ad. It sets the timeframe for assigning conversion credit.
- Confidence IntervalConfidence Interval is a statistical range of values that likely contains the true value of a metric. In marketing analytics, it quantifies uncertainty around estimates, indicating the precision of an outcome or causal effect.
- IncrementalityIncrementality is the part of an outcome, such as orders or revenue, that a marketing activity caused: the extra results compared with what would have happened without it. It is a causal effect, not the number of conversions credited to a channel.
- 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.
- Propensity ScoreA propensity score is the probability a unit receives a specific treatment given observed characteristics. It reduces selection bias in observational studies, enabling causal inference when randomized experiments are not possible.
- Propensity Score MatchingPropensity Score Matching is a statistical method that estimates the causal effect of a treatment from observational data. It matches individuals with similar likelihoods of receiving treatment to isolate its impact.
- Selection BiasSelection Bias occurs when data points selected for analysis do not represent the target population. This leads to distorted findings about marketing campaign impact.