A CFO asks a simple question in a planning meeting: Did the offsite retail media buy create sales, or did it just get credit for sales that were coming anyway? The answer is yes, it can create incremental sales, but the proof is much weaker when the network controls the audience, attribution and measurement. I reached that answer by comparing current industry measurement guidance with published offsite campaign evidence and methodology research.
Key Takeaways
- Offsite retail media can be incremental, especially when it reaches buyers the brand could not otherwise reach.
- A strong ROAS report can still overstate the sales caused by the media.
- Closed-loop measurement is useful, but matching an ad exposure to a purchase is not the same as proving causality.
- A holdout or geo test is the cleanest practical way to test whether the campaign changed sales.
What Offsite Retail Media Actually Is
Offsite retail media uses a retailer’s shopper audience to buy advertising outside the retailer’s own website or app. That can include display, social, video, connected TV, audio and other third-party inventory. The retailer brings the audience and purchase data, while another platform supplies the place where the ad appears.
That is the key difference between onsite and offsite. Onsite media can put a sponsored product beside a shopper’s search on the retailer’s own property. Offsite can use the same first party data to reach that shopper somewhere else, often before the shopper is actively searching.
The trade-off is measurement. Onsite sits close to the transaction. Offsite has to rebuild the connection between an exposure on one platform and a purchase somewhere else. A 2025 analysis of 573 offsite display campaigns found that changes in ROAS methodology alone could move reported ROAS by an average of 63%.
That does not mean offsite is ineffective. It means the number attached to it needs more scrutiny.
Why The Closed Loop Argument Weakens Offsite
The Purchase Happens Somewhere Else
Closed-loop measurement sounds simple: identify the shopper who saw the ad, find the later purchase and connect the two. The retailer’s first party data makes that possible in ways many other media channels cannot match.
But the loop gets harder once the ad leaves the retailer’s property. The exposure may happen on a different device, platform or household identifier. The purchase may happen later in a store or through an account the measurement system cannot fully connect.
Current industry guidance treats controlled experiments as the strongest way to establish causal lift. It identifies randomized tests and matched-market approaches as strong methods for validating ROI while noting that platform-level proxies can provide only directional evidence.
Match Rate Changes: What You Can See
Match rate is simply the share of records that can be connected across the measurement chain. If the retailer can identify only part of the exposed audience in its transaction data, the reported result is based on an incomplete view.
That matters because the missing shoppers are not necessarily identical to the shoppers who can be matched. In the 573-campaign analysis, lower match rates increased the risk of inaccurate ROAS when untraceable sales were extrapolated. The analysis illustrated how a 90% retailer-to-identity match combined with a 60% identity-to-media match would produce a 54% combined match rate.
The practical question is not just “What was the ROAS?” It is “How much exposed audience could you connect to a purchase?”
What The Available Evidence Actually Shows
The evidence falls into four groups, and they do not deserve equal weight.
Network Case Studies
Network and media-platform case studies show that campaigns were followed by sales, lift or new buyers. They are useful for understanding the measurement of a network and outcomes of reports. They are not enough to establish a general claim that offsite retail media is incremental. The seller has a commercial interest in the campaign, and the methodology can vary by network.
For example, one published offsite audio campaign analysis across 43 campaigns reported a $10.30 average attributable ROAS and a 14% sales uplift. The source itself says the figures should not be treated as performance benchmarks.
That is evidence of reported performance. It is not a universal incrementality benchmark.
Agency And Measurement Analyses
The stronger evidence often comes from work that exposes how the number is produced.
The 573-campaign offsite analysis found that product attribution, untraceable sales, household versus customer attribution and impression definitions could materially change reported ROAS. Its average 63% shift shows why a single network number cannot be compared casually with another network number.
This is where retail media incrementality becomes a different question from attribution. Attribution asks which purchases can be linked to media. Incrementality asks what changed because the media ran.
Independent Tests
Independent measurement is more useful when the advertiser and media seller do not control the entire measurement process.
One current example pairs a retail network with a third-party measurement provider for household-level lift measurement. That approach compares exposed households with a control group rather than simply counting purchases after exposure. But the test still needs to be judged on its audience, control construction, measurement window and methodology.
Published Holdouts
This is the evidence buyers should want more of. A real holdout leaves a comparable group unexposed and compares its sales with the exposed group. The difference provides the basis for an incremental result.
The industry has published methods for doing this. Current guidance identifies experiments as the strongest option for proving causal lift while noting that identity fragmentation and ecosystem silos make cross-network comparisons difficult.
The uncomfortable finding is that public evidence isolating offsite incrementality remains much thinner than the volume of network case studies and ROAS claims. That does not prove offsite lacks incrementality. It means the category has less public causal evidence than its sales reporting can make it appear to have.
Where It Probably Is Incremental
Offsite is most convincing when it reaches people who were not already on the path to purchase.
New Category Buyers
A brand launching a new product can use retailer audiences to find category shoppers who have not bought that brand before. The value is higher if the campaign reaches people the brand could not efficiently find through its existing media.
New-to-brand results can help here, but they are still not the same as a causal test. A new buyer may have purchased without the ad.
Prospecting And Launches
Offsite can also make more sense for launches and prospecting because the campaign has a job beyond closing an existing intent signal.
A 2025 Google and Rakuten study involving 44 advertisers found only 4% overlap between shoppers who purchased through onsite retail media and those who purchased through offsite retail media. That suggests the two activities can reach different buyer groups, although the result by itself does not prove that the offsite sales were incremental.
The conditions matter: the audience needs to be genuinely incremental, the product needs to be available and the test needs a credible baseline.
Where It Probably Is Not
The weakest case is retargeting a shopper who was already highly likely to buy.
Existing Intent
A shopper who has searched for the product, visited the retailer or already added the item to a basket may need very little persuasion. An offsite ad can appear before the purchase and receive credit without being the reason the purchase happened.
That is why sponsored products and offsite media should not be judged by the same causal standard. Sponsored products sit closer to an active purchase decision. Offsite placements can influence demand earlier, but that also makes the counterfactual harder to observe.
Cheaply Reachable Audiences
If the same audience can be reached through another channel at lower cost, the retailer’s first-party audience advantage needs to show up in the test result. Otherwise, the brand may be paying a premium for data it could already access elsewhere.
The same applies when the attribution window is broad enough to capture purchases that were likely to happen anyway. A longer window can produce more attributed sales without producing more incremental sales.
How To Test It Yourself
Do not start with the report. Start with the control group.
- Choose One Campaign. Pick a campaign where the result could change your budget decision.
- Create A Holdout. Use a randomized audience holdout when the platform can support it. If that is not practical, use matched geographies or stores with similar historical sales.
- Set The Rules Before Launch. Agree on the test period, sales outcome, minimum useful lift and statistical threshold before seeing the result.
- Ask For The Match Rate. Get the share of exposed users or households that can be connected to purchase data. Also ask how untraceable sales are handled.
- Measure Total Sales. Do not rely only on attributed sales. Compare the treatment group with the control group and measure incremental sales and incremental ROAS.
- Check For Leakage. Make sure the holdout group is not receiving the same campaign through another device, platform, audience or retailer program.
- Write The Result Into The Contract. If the network refuses a holdout or will not disclose how the control group works, treat that limitation as part of the buying decision.
The test does not have to be perfect. It has to be designed before the result and strong enough to answer the question you are actually paying to answer.
The Answer, Stated Plainly
So, is offsite retail media incremental? Yes, sometimes. But a network’s attributed ROAS is not proof of it.
The strongest case is offsite media that reaches genuinely new or hard-to-reach buyers and passes a credible holdout or geo test. The weakest case is retargeting high-intent shoppers with a long attribution window and no control group.
The evidence changes my view only when the measurement moves from “people exposed later bought” to “people exposed bought more than a comparable group that was not exposed.”
That is the standard buyers should put into the next contract: a credible incrementality test, with the control design and match rate disclosed before the campaign starts.
