How Do CPG Brands Measure Retail Media ROI and Incrementality Across Amazon, Walmart, and Instacart?

By the end of this guide you will know why retailer-reported ROAS overstates performance, how to run an incrementality test across Amazon, Walmart, and Instacart, which measurement tools each platform gives you, and how to defend a retail media budget to a CFO with numbers that hold up.

Why Retail Media ROI Is So Hard to Measure Across Retailers

Retail media is measurable in a way trade promotion never was, but the measurement comes with a catch: every retailer grades its own homework. Amazon reports on Amazon Marketing Cloud, Walmart reports through its own attribution, and Instacart reports inside its own dashboards. Each one uses a different attribution window and a different definition of a conversion, and each one has an incentive to show its channel in the best light.

The result is a fragmented view. You end up with three ROAS numbers that do not add up to one truth, no shared baseline, and no clean way to say which network actually drove incremental sales. Stitching those walled gardens into a single picture is the core problem CPG measurement teams are trying to solve.

The other trap is siloed, last-touch attribution. When each platform claims the last click before a purchase, upper-funnel and cross-retailer effects get systematically undercounted. A 2026 analysis by Incremental found that siloed last-touch attribution understated the return on upper-funnel retail media by a factor of three to five, as reported by PPC Land.

This matters more every year because the budget at stake keeps climbing. US retail media ad spend will more than double between 2023 and 2027, reaching $109.40 billion, and it is growing more than 20 percent a year (according to eMarketer). When a line item grows that fast, the pressure to prove it works grows with it.

Reported ROAS vs Incremental ROAS: The Number That Actually Matters

Reported ROAS is revenue attributed to your ads divided by ad spend. It is easy to pull and easy to over-trust. The problem is that it counts every attributed sale equally, including the shopper who searched your brand name, saw your sponsored listing, and bought the product they were already walking to the shelf for.

Incremental ROAS strips that out. iROAS answers a narrower, harder question: of the sales credited to this campaign, how many would not have happened without it? That is the only version of return a CFO should fund against, because it is the only one that reflects new money.

The gap between the two is large and predictable. Incremental ROAS typically runs below last-click ROAS, and the gap is widest on branded search, where you are mostly paying to reach demand you already own. A campaign can post a 6x reported ROAS and a 2.5x incremental ROAS at the same time, and only one of those numbers tells you whether to keep spending. For the broader case that a healthy-looking ratio can still hide the truth, see our breakdown of the number that tells you if paid actually pays back.

How to Run a Retail Media Incrementality Test

You cannot read incrementality off a dashboard. You have to create a comparison the platform does not give you, by withholding your ads from part of the market and measuring what happens. The cleanest designs for retail media:

• Geo holdout (matched markets): Split comparable regions into test and control. Run your retail media in the test markets, hold it out of the control markets, and keep every other variable flat. The sales difference, adjusted for baseline, is your incremental lift.

• Audience holdout: Where a platform's clean room supports it, suppress ads from a randomized slice of the audience and compare their purchase rate to the exposed group. This is the most rigorous design because assignment is random.

• Before-and-after with a control: When a true holdout is not possible, compare the campaign period against a matched prior period, using a non-advertised region or SKU as the control to net out seasonality.

A few rules keep the test honest. Match your test and control on baseline velocity, not just population, or the comparison is rigged from the start. Keep spend and promotions flat everywhere else during the window, since a coupon drop or a distribution change will swamp your ad effect. Run it long enough to clear normal purchase cycles, which for a pantry staple is weeks, not days. And decide your success threshold before you look at the data. For the wider set of methods behind this, our marketing attribution guide covers how holdouts, MMM, and attribution fit together.

Retail Media Measurement Methods Compared

Different methods answer different questions at different levels of rigor. Most CPG programs use more than one: a platform dashboard for daily optimization, an incrementality test to calibrate the truth, and a model to allocate across the whole mix.

Unifying Measurement Across Amazon, Walmart, and Instacart

The practical way to compare walled gardens is through their clean rooms, the privacy-safe environments where you can analyze a retailer's data without the raw shopper records ever leaving the platform. Each of the big three now offers one, and mid-market brands can finally reach them without a full in-house data-science team.

• Amazon Marketing Cloud (AMC): Amazon's clean room, offered at no media cost to eligible advertisers, returns aggregated analytics across Sponsored Ads and DSP so you can model reach, frequency, and new-to-brand outcomes. It is the deepest of the three for custom analysis. Our Amazon advertising guide covers the ad products AMC measures.

• Walmart's Scintilla: Walmart rebranded its Luminate insights platform as Scintilla, and in 2026 opened a media data feed that streams roughly 500 retail and operational data elements, from omnichannel sales to velocity and inventory, to approved partners. It is the strongest link between media and the physical shelf.

• Instacart's Data Hub: Instacart's clean room brings grocery basket and purchase data into the same kind of privacy-safe analysis, useful for measuring basket impact and the in-store halo Instacart reports on its ads.

The unifying move is to define one scorecard, incremental sales per dollar, and force every network to report up to it. Let each platform optimize in its own terms underneath, but judge the mix on a shared, incrementality-based number. For how the networks themselves stack up, see our comparison of the best retail media networks for CPG brands and our guide to running Walmart Connect, Target Roundel, and Amazon Ads.

How to Defend a Retail Media Budget to a CFO

A CFO does not fund clicks or a platform's self-graded ROAS. She funds incremental revenue and contribution margin. That means walking in with the iROAS, not the reported ROAS, and being honest that the two differ. A team that volunteers the gap earns more trust than one that quotes the flattering number and waits to get caught.

Frame the ask in the finance team's language. Show the incremental sales a test produced, the contribution margin after trade and cost of goods, and the payback window. Then show what you would cut if the budget were flat, so the trade-off is explicit. Incrementality testing is now mainstream, used by 52.0 percent of US brand and agency marketers (EMARKETER and TransUnion, July 2025), which means a CFO increasingly expects this proof rather than treating it as extra credit.

How We Approach Retail Media Measurement at jetfuel.agency

We are a remote-first performance marketing agency, started in Orange County, and we run paid media and retail media strategy for DTC, CPG, and food and beverage brands. We treat measurement as the first job, not the last, because you cannot scale a channel you cannot trust.

In practice that means we start every retail media program by separating branded defense from category offense, so reported ROAS stops flattering itself before we make any budget calls. We calibrate the real number with matched-market and geo-holdout tests rather than accepting a platform's attributed ROAS at face value, and we roll Amazon, Walmart, and Instacart up to one incrementality-based scorecard instead of three dashboards. We tie that back to what is happening on paid social and the DTC site, since a Meta launch that lifts grocery search will show up as retail media performance the platform is happy to take credit for.

Frequently Asked Questions About Measuring Retail Media

The Bottom Line

Retail media is only as good as the number you judge it by. Report ROAS to optimize day to day, but measure incremental ROAS to decide what to fund, calibrate it with matched-market and holdout tests, and roll Amazon, Walmart, and Instacart up to one incrementality-based scorecard. Do that, and retail media stops being three dashboards you half-trust and becomes a channel you can defend to the CFO and scale with confidence.