If I ran a retail media network
Every network in the category reports a sales number that no independent party has ever checked. The first one to put that number through an audit owns the only unoccupied position left.
You know the quarterly review. Six retail media networks on six slides, each with a return-on-ad-spend figure next to it, and somebody in the room asks whether 4.1 at one retailer is better than 3.7 at another. The honest answer is that nobody in the room can tell, including the networks, because the two numbers were calculated by different parties using different windows against different definitions of a sale. The meeting moves on anyway, because the alternative is admitting that the largest growing line in the budget is allocated on figures that cannot be compared.
I spent last week reading what has actually been independently checked in this category. The answer turned out to be narrower than I expected, and it points at a move no network has taken.
Of the six largest US retail media networks, the number with an independently accredited sales metric. Four hold MRC accreditation; every one of those covers impressions, clicks and viewability. Attributed sales, ROAS and incrementality are accredited nowhere.
The audit stops before the number that matters
Amazon's accreditation covers Sponsored Products clicks and DSP impressions, clicks and viewability. Walmart Connect's covers Sponsored Search impressions, clicks and CTR, granted September 2025. Criteo's covers onsite display rendered impressions and clicks, from March 2024. Instacart's covers impressions, clicks, CTR and viewable impressions, continued in June 2026. Roundel and Albertsons Media Collective are not in the MRC's digital directory at all.
Every one of those letters certifies that an ad was served. None of them touches the sales figure the budget was approved against. An advertiser reading an accreditation badge on a retail media deck is reading a guarantee about delivery and inferring one about outcomes, and the gap between those two things is where the entire category sits.
One network has moved. Kroger submitted attributed sales, incremental sales, aROAS and iROAS for MRC accreditation, and has been in pre-audit since June 2026. Nothing has been granted. It is the first outcome metric in the council's retail media pipeline, and it came from a network with roughly a tenth of Amazon's share rather than from either of the two companies taking most of the money.
Albertsons Media Collective publishes the problem
March 2026The category's answer to untrustworthy ROAS has been incrementality, and nearly every network now reports a version of it. In March 2026 Albertsons Media Collective published research finding that iROAS can vary 6.5x, and flips the result in 83% of campaigns, depending on filtering, matching and revenue-calculation choices. Those choices belong to the network. A control group, a set of matching variables and a revenue definition are three places to put a thumb, and none of them is visible to the buyer.
A company that sells incrementality reporting published evidence that its own headline metric moves 6.5x on decisions the buyer cannot see — which is why it carries more weight than any agency complaint.
Read that against what incrementality is sold as. It is positioned as the rigorous alternative to attribution, the number that separates demand a brand created from demand it merely caught. Unaudited, it is the opposite: a figure with more discretion inside it than ROAS, because it has more construction inside it. Three private methodology choices sit between the raw data and the headline, and all three belong to the party being graded.
Audited outcomes, and the data to check them
So here is what I would do with a challenger network, and it is one move rather than three. Submit the outcome metrics to independent audit, report incrementality as the default rather than as a premium tier, and let the advertiser pull log-level data out and re-derive the lift in their own environment. Those three only work together. Audited methodology without portable data is a certificate the buyer still cannot check. Portable data without an audit is a pile of logs and an argument. Incrementality without either is what the category already has.
The objection arrives immediately, and it is real. Audited numbers are smaller numbers. A network that reports honestly competes against six that do not, and the first quarters are worse. I went looking for evidence on how that trade actually lands and found none — no retailer in this set discloses retail media revenue separately enough to show what a transparency move did to it. That cost is unmeasured, and anyone telling you the move is safe is guessing.
What is measurable is the demand. Advertisers name performance measurement, modeling recommendations and business plan adherence as their top three budget-allocation factors, all at 38%. Performance already moves the money. It is simply that no verified performance exists in this category for it to move toward.
The part that is not about honesty
Transparency arguments lose to performance arguments, and they deserve to. A buyer holding a number that looks good does not want a lecture about where it came from. If this were only an ethics pitch I would not run it.
But a network cannot improve against a measure it writes. Optimization needs an objective the optimizer did not author. A network tuning toward a figure whose methodology it controls, and which moves 6.5x on those controls, cannot distinguish getting better from counting differently — and across enough cycles the system learns the counting. That is a capability ceiling, and it applies to networks acting in perfectly good faith.
Which turns the sales conversation around. Against a closed-loop competitor reporting a higher return, the line is not our numbers beat theirs. It is that theirs cannot be checked, ours can, and only one of us has a mechanism for getting better at this. That argument survives losing the number comparison, because the number is the thing in dispute.
The networks doing the most are not winning
Honesty about the state of play: the measurement leaders in this category are the challengers, and the challengers are losing. Amazon took 79.7% of US retail media spend in 2025 and Walmart 8.0%. Of roughly $10.53bn in net-new US spend forecast for 2026, $9.42bn — over 89% — goes to those two, with every other tracked network expected to hold flat or shrink through 2027. Neither has an accredited sales metric between them.
That is the strongest case against everything above, and the data currently wins it. What the data cannot say is whether measurement would move share, because nobody has shipped the thing being argued for. Instacart has accredited delivery metrics and an open clean room, and no accredited outcome. Albertsons has the best-designed incrementality in the set and no accreditation at all. Kroger has the outcome metrics in audit and stands alone there. Each has a piece. A null result against a treatment nobody has administered is a null on the market, not on the idea.
- You are not the default buy. A network already receiving budget by reach has revenue to lose from an honest number and nothing to win from one. This is a challenger's move.
- Your methodology survives being read. An audit publishes the control-group construction and the matching variables. If those choices were made to flatter the figure, the audit is where it is found out.
- You can carry two to four weak quarters. The reported number goes down before the share goes up, and the second half is unproven.
- Your buyer is on the media side, not only the trade side. Sixty-two percent of retail media dollars are still sourced from trade and shopper budgets, and those arrive with obligations a measurement argument does not reach.
- CEO The only differentiated position in a category where 89% of new money goes to two competitors is the one none of the seven has taken, and audits take quarters. Being right early looks identical to being wrong for about a year. Commit to the audit in this planning cycle or concede the position to whoever does.
- CFO Audited outcome metrics mean a smaller reported number against the same media revenue, and no retailer discloses enough for anyone to price that drop. The media line is the highest-margin thing on a grocery P&L and this puts a hand on it. Size the revenue at risk before signing, and do not accept a plan that leaves it unquantified.
- CRO Advertisers already allocate on performance at 38%, tied with business plan adherence, and none of that performance is verified. The sales team has to sell a lower number against six competitors quoting higher ones. Arm them with the comparability argument, not the magnitude one.
- CPO Log-level export and audited methodology are infrastructure, not features, and the 6.5x variance shows how much of the current number lives in undocumented choices. Shipping the export surfaces every inconsistency the reporting layer has been absorbing. Build the export first — it forces the methodology to become real before the auditor asks.
- Legal Portable log-level data is shopper purchase data leaving the retailer's control, under loyalty terms written before anyone proposed it. A clean-room boundary is the difference between a differentiator and a disclosure event. Settle the data-egress boundary before the commercial team sells it.
- Risk An audit publishes the methodology, which means a finding that the previous reporting was generous becomes a documented fact about historical campaigns. There is no version of this where the pre-audit numbers go unexamined. Run the internal re-derivation first and know the answer before an auditor does.
Reconciled. The CFO and Risk seats pull hardest against the CEO's, and they are arguing about the same thing from opposite ends — the honest number is lower, and the gap becomes public. The single course that satisfies five of six is to re-derive the last four quarters internally before committing to anything external: it gives the CFO the size of the revenue at risk, gives Risk the answer before an auditor finds it, gives the CPO the methodology discipline the export needs, and leaves the CEO free to commit or walk with the number in hand rather than without it. Only Legal's question sits outside that work, and it is a separate week.
Re-derive one quarter of your own iROAS under three different defensible methodologies, before you decide anything.
Take a single completed quarter of campaigns. Have your analytics team compute incremental lift three ways — your current control-group construction, a stricter matched design, and one that excludes the halo and same-brand sales your current definition includes. The spread between those three is the number a buyer would discover in an audit, and it is the number the whole decision turns on. If it is narrow, the audit is cheap and you should book it. If it is 6.5x, you have learned what your reported figures have been carrying, and you have learned it privately.
- What it costs
- Two analysts for roughly three weeks against data you already hold, plus the willingness to look at an answer you may not want.
- How you'll know
- The spread across the three methodologies. Under about 1.5x and your numbers are robust enough to audit this cycle; above that, the work is fixing the methodology before anyone outside sees it.
The evidence behind this piece is published in full.
Seven networks scored on four measurement dimensions, with the registered predictions, the one that was written badly, and what the evidence cannot say.
Read the study See the Decision Briefs → Or commission a custom brief →