- 83% of active buyers want incremental sales measurement from retail media partners, and 76% want ROAS. [1]
- 64% of retailers and networks say they offer incremental or sales lift measurement, and 60% offer ROAS. [1]
- Those percentages come from 60 buy-side and 32 network respondents in one country, in June 2026. [1]
- The word behind them covers four methods: experiments, model-based counterfactuals, econometric models, hybrid proxies. [3]
- Only the first establishes a counterfactual by construction. The survey question did not ask which one.
- The IAB/MRC guidelines are written as recommendations, and admit most networks do not follow the viewability one. [4]
- 73% of buyers name inconsistent metrics across networks as the top measurement challenge, while 86% run three or more partners. [1]
The retail media measurement gap has become one of those facts that circulates without its source attached. Buyers want incrementality, networks cannot supply it, and somewhere in the retelling the numbers acquire the authority of a census. They are not a census. They are one survey, and reading it properly is more useful than quoting it.
Here is the survey in its own words. Incremental sales measurement and ROAS are the most sought-after metrics from retail media partners, cited by 83% and 76% of active buyers respectively, while retailer provision remains uneven, with 64% of retailers and commerce media networks offering incremental or sales lift measurement and 60% offering ROAS. [1]
And here is the base. The report behind it is built on a June 2026 survey of 92 advertising decision makers, 60 of them senior buy-side representatives and 32 of them from retailers or commerce media networks. [1][5] IAB Australia also states that the data comes from experienced retail and commerce media practitioners rather than the broader Australian advertising market. [1]
So the most-quoted number in retail media measurement is 64% of 32 companies, about twenty of them, in one country, answering yes to a question about a word. That is not a criticism of the research, which is careful and says all of this on its own page. It is a criticism of how the number gets used, and it matters because the word itself is the problem.
Four methods, one label
In November 2025 the IAB and IAB Europe published guidelines for incremental measurement in commerce media. They name four methodologies, and say when each applies: experiments, model-based counterfactuals, econometric models, and hybrid proxies, grounded in three principles of credible counterfactuals, control of bias, and separation of signal from noise. [3] The earlier IAB and Media Rating Council retail media guidelines devote a chapter to the same question, listing randomized controlled trials, synthetic controls, matched-market tests, and machine learning models. [4]
Read those lists next to the survey question and the gap changes shape. A network running a geo-split holdout and a network running a regression on last year’s sales can both answer yes to “do you offer incremental or sales lift measurement.” One of them built a counterfactual. The other estimated one. Both are legitimate techniques with published methodologies, and they carry completely different weight in a budget argument.
| Method | Where the counterfactual comes from | What it survives |
|---|---|---|
| Experiment [3] | A group held out by design, randomized or geographic | A finance review, because the comparison was built before the spend |
| Model-based counterfactual [3] | A synthetic control assembled from comparable units [4] | Scrutiny of how the comparison set was chosen |
| Econometric model [3] | Variation in historical spend, estimated | A question about what else changed in the same period |
| Hybrid proxy [3] | Partly observed, partly assumed | Only the question it was built to answer, and only if you know which |
“While MMM infers impact, closed-loop reporting and incremental sales lift testing prove it causally.”
IAB, April 2026That line comes from an IAB article arguing that marketing mix modeling undervalues retail media, written by its own vice president for commerce and retail media. [2] The second half of it is uncontroversial: incremental sales lift testing is a causal method. The first half quietly promotes something else.
Closed-loop reporting is the retailer linking an exposure to a purchase in its own transaction data. That is a deterministic record of what happened, and it is genuinely better than a modeled guess at what happened. It is still not a counterfactual. It does not tell you what the shopper would have bought anyway, which is the entire question incrementality exists to answer, and it is the same conflation that makes a well-attributed channel look like a driver when it is a passenger.
The same article gives the honest version two paragraphs later, recommending that brands treat incrementality testing, meaning control and exposed designs, randomized experiments, and geographic lift tests, as the primary foundation rather than an optional enhancement. [2] That is the right advice. It sits in the same piece as a sentence that lets a network answer the question without following it.
The standard is written in the conditional
It is worth knowing what the rulebook actually obliges anyone to do, because the answer is nothing. The IAB/MRC Retail Media Measurement Guidelines, published in January 2024, are phrased as recommendations from end to end: retail media measurement “is recommended to” use robust methodologies, adhere to privacy regulations, employ consistent data capture. [4]
The clearest admission is on viewability. The document states that the Media Rating Council requires viewable impressions for attribution of outcomes to ad exposures, and that it recommends retail media measurement be compliant with those requirements and ultimately accredited by that body. [4][6] Then it adds: “We acknowledge that the majority of retail media organizations may not reflect or consider viewability in attribution today, and we expect all retail media and publisher organizations will need to move in this direction.” [4]
That was written in January 2024. The 2026 survey finds inconsistent metrics and definitions across networks named as the top measurement challenge by 73% of active buyers, and 59% reporting difficulty measuring ROAS or ROI. [1] Two years, a published standard, and the complaint has not moved, which is what a voluntary standard looks like from the outside.
What to ask instead
The useful response is not to distrust retail media. Spend is rising for good reasons: 63% of active buyers increased spend on retailers’ own websites and apps and 65% on retailer-powered advertising outside them, and satisfaction improved to 66% rating the partner experience as good, up from 44% the year before. [1]
The useful response is to stop accepting a word as an answer. Four questions do most of the work, and they fit in a partner review.
Which of the four methods produced this number?
A good answer names one: an experiment, a model-based counterfactual, an econometric model, or a hybrid proxy. [3] A network that cannot name the method is reporting a result it did not produce, usually because a vendor did.
What was the control group, and how was it formed?
Randomized holdout, matched market, synthetic control, or geography. [4] If there was no group that did not see the ads, the number is an estimate of incrementality rather than a measurement of it, and it should be labeled that way in your deck.
What lookback window, and what counts as a conversion?
The window and the sale definition decide the result before any method runs, which is precisely why inconsistent definitions across networks is the complaint buyers raise most. [1] Two networks with honest processes and different windows will hand you two numbers you cannot add together.
Can I reproduce it from data I can also see?
Not always possible, and a no is acceptable if it is explained. What is not acceptable is a result that exists only inside the seller’s reporting with no path to verification, because that is the structure of every measurement failure this industry has had.
Retailers themselves are candid about where they are. In the same survey, 22% described their retail media offering as advanced and 47% described it as still developing, while 84% said they plan to enhance measurement in the next twelve months. [1] Networks are not pretending to be finished. The pretending happens downstream, in the deck where four different methods arrive as one column.
The gap in the headline is real, and it is probably wider than a survey of experienced practitioners suggests, since the people who answer a retail media questionnaire are the ones paying attention. But the number is not the finding. The finding is that an industry with 86% of its buyers running three or more partners [1] has agreed on a word and not on what it means, and no standard written in the conditional is going to settle that. Your contract can.
Sources
- IAB Australia · Retail media’s reset year: bigger budgets but harder questionsJuly 7, 2026. The 83, 76, 64 and 60 percent figures, the challenge and satisfaction numbers, and the survey base of 92 respondents.
- IAB · Is your legacy measurement sabotaging your growth in the retail media era?April 7, 2026, by Collin Colburn and Priyash Shahane. The closed-loop sentence, and the recommendation to treat incrementality testing as the primary foundation.
- IAB and IAB Europe · Guidelines for incremental measurement in commerce mediaNovember 3, 2025. The four methodologies and the three principles behind them.
- IAB and MRC · Retail media measurement guidelinesJanuary 2024. Chapter 4 on incrementality methods, and the viewability passage acknowledging that most organizations do not follow it.
- IAB Australia · Retail and commerce media state of the nation 2026The report itself, behind the media release above.
- Media Rating Council · Media Rating CouncilThe accreditation body the retail media guidelines point at, and whose minimum standards they reference.
Frequently asked questions
What is the retail media measurement gap, exactly?
It is the distance between what buyers ask for and what networks say they provide. In IAB Australia’s 2026 survey, incremental sales measurement and ROAS were the most sought-after metrics from retail media partners, cited by 83% and 76% of active buyers, while 64% of retailers and commerce media networks offered incremental or sales lift measurement and 60% offered ROAS.
How big is the sample behind those numbers?
Small, and IAB Australia says so plainly. The report is based on a June 2026 survey of 92 advertising decision makers involved with retail and commerce media, including 60 senior buy-side representatives and 32 representatives from retailers or commerce media networks. The 64% figure is 64% of those 32. The report also notes it draws on experienced practitioners rather than the broader Australian advertising market.
What are the four methods the standard names?
The IAB and IAB Europe Guidelines for Incremental Measurement in Commerce Media, released November 3, 2025, set out experiments, model-based counterfactuals, econometric models, and hybrid proxies, and say when to use each. They rest on three principles: credible counterfactuals, control of bias, and separation of signal from noise.
Is closed-loop reporting the same as incrementality?
No. Closed-loop reporting links an ad exposure to a purchase inside the retailer’s own data, which is a deterministic record of what happened. Incrementality asks what would have happened without the exposure, which needs a group that was not exposed. A network can have excellent closed-loop reporting and still never have run a counterfactual.
Are the IAB and MRC guidelines binding on networks?
No. The IAB/MRC Retail Media Measurement Guidelines, published in January 2024, are written as recommendations throughout. On viewability the document states that the MRC requires viewable impressions for attribution of outcomes, that it recommends retail media measurement be compliant and ultimately accredited, and acknowledges that the majority of retail media organizations may not reflect or consider viewability in attribution today.
So what should I ask a network before signing?
Which of the four methods produced the number, what the control group was, what the lookback window is, and whether the result can be reproduced from data you can also see. A network that answers all four is doing measurement. A network that answers with a ROAS figure and a case study is doing reporting.
Is any of this a reason to spend less on retail media?
Not by itself. The same survey found 63% of active buyers increased spend on retailers’ own sites and apps and 65% increased spend on retailer-powered advertising elsewhere, and advertiser experience is improving, with 66% rating it good against 44% the previous year. The argument is about what you can prove with the money, not whether to spend it.
Why does the number of partners make this worse?
Because each one grades its own homework in its own words. IAB Australia found 86% of buyers now work with three or more retail or commerce media partners and a third work with more than five, while inconsistent metrics and definitions across networks were the top measurement challenge for 73% of active buyers. Five incomparable numbers is not five times the evidence.
Mikołaj Salecki
Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.
