Sam's Club's Rest-of-Market Report: A Three-Question Framework for Retail Media Incrementality

Somewhere right now, a head of growth is walking into a quarterly media review with a slide that says retail media delivered an 8x ROAS. The CFO will nod. The CMO will nod. Budget will get redirected. And nobody in the room will actually know whether that 8x is a real number or an accounting fiction.

Sam's Club MAP shipped something last week that should make that meeting uncomfortable. Its new Rest-of-Market analysis — ROM, in the deck — compares members who saw a given brand's ad against members who didn't, using deterministic member IDs rather than the usual probabilistic soup, and reports back an incremental ROAS. Not a blended one. Not a modeled one. The number that tells you what the ad actually caused.

For most retail media networks, that kind of disclosure is terrifying, which is why we haven't been seeing it.

The long con of last-touch ROAS

Almost every retail media dashboard you've ever stared at tells the same story. A user clicked your sponsored listing, a user bought the product, therefore the ad drove the sale. Fine story, clean math, seventy-plus billion dollars of annual ad spend justified.

The trouble is that most people clicking a sponsored listing were already walking toward the shelf. They typed "Tide" into the search bar because they buy Tide. They've got it in their cart every three weeks. Your sponsored impression caught them on the way to a conversion that was going to happen regardless, and then collected the credit, and then sent you a report congratulating you on an 8x ROAS you did not, in any meaningful sense, earn.

Everyone on the sell side has known this for a decade. It survives because every party in the transaction gets paid by the fiction. Platforms sell more inventory when blended ROAS looks juicy. Brand teams defend headcount when their channels post big numbers. The one person with an incentive to dig is the finance lead — and most finance leads, reasonably, don't know enough media math to push back when the dashboard looks good. That quiet truce is what last-touch ROAS is built on, and it's starting to break.

What Sam's Club did, in one subtraction

Incrementality is a subtraction:

iROAS = (revenue with ads − revenue without ads) / ad spend

To do the subtraction you need a control group — actual shoppers who didn't see the ad but otherwise look like the ones who did. Retail media's historical trouble is that constructing that group honestly is hard. Device graphs go stale between the ad and the aisle. Cookie-based panels can't follow a shopper into a physical store. "We modeled a counterfactual" has been the polite industry shorthand for "we made one up."

Sam's Club's panel gets around most of that by accident of its business model. Every Sam's Club shopper is logged in, because the login is the membership and the membership is the store. The identity layer is deterministic by construction. ROM extends that spine outward to partner retailers through SKU-level deterministic matches, so the test group (saw your ad) and the control group (same profile, didn't) are both real members with real receipts. Not audiences. Not look-alikes. Real humans, both sides of the subtraction.

None of this makes ROM a perfect study. Nothing is. But it passes the three-part test an incrementality report has to pass before anyone should take it seriously: is there a real control group, is the join deterministic, and is the reported iROAS broken out from the gross. Three yeses, one honest number.

The questions that separate measurement from marketing

Bring the same three questions to your next RMN review. If the account team has to reach for a sales deck to answer them, the deck is the answer.

Start with the control group. "We estimate baseline using predictive algorithms" is not a control group; it's a wish list dressed up as methodology. A real control is a set of shoppers who demonstrably did not see the ad and whose subsequent purchases you can observe directly. If the vendor can't describe how that set is identified, there isn't one.

Then ask about identity. Logged-in member IDs and deterministic household matches hold up over time; probabilistic device graphs lose resolution at every attribution-window extension, and retail is a long-window game — consideration, list-making, a trip to the store, a purchase that may happen this week or in three. If the vendor's incrementality study runs off a DSP-era probabilistic graph, the iROAS at the bottom is a guess in a lab coat.

Finally, ask what's left after you subtract the baseline. The gross ROAS is what the platform puts on its case studies. The net is what your CFO needs to hear, because it's what determines whether the spend is incremental revenue or a subsidy. A 5x gross often lands somewhere between 1.2x and 1.5x net. That range is often still worth buying, but it's a different budgeting conversation.

The reallocation that's coming

Sam's Club won't stay alone on the disclosure curve. Kroger Precision, Target Roundel, and Walmart Connect all have CFO customers asking the same iROAS questions right now, and nobody wants to be the network whose measurement story collapses the first time someone reads past the headline. Expect comparable studies from at least two of them in the next six months. Expect a few others to quietly delay.

The budget reallocation that follows won't be dramatic. It'll look like senior marketers moving a few points of spend every quarter, away from networks that can't produce an incrementality number and toward the ones that can. Across a year, that's several billion dollars of retail media budget tilting toward the platforms with honest math. The vendors who ship disclosures ahead of the curve get the flow. The vendors hiding behind blended ROAS eventually discover that the floor they thought they were standing on was rented.

Here's the practical move, and you can start it this week. Pick the retail media contract whose blended ROAS you'd most like to believe, and ask the account team for an incrementality study that meets all three of the above: real control group, deterministic join, iROAS net of baseline. If they stall, if the request gets punted between analytics and account management, if the numbers that come back sidestep the three questions — that stalling is the measurement. You didn't need to run the study; you already learned what you needed to learn.

Retail media mostly works. Some of it works extraordinarily well. But which parts, for which brands, at which spend levels — that's the question last-touch ROAS has never been able to answer, and the question every network is finally going to have to.