RMN Measurement, Compared.
The cross-network measurement matrix with the column no vendor publishes honestly: accreditation status.
Retail media network measurement is how each RMN defines, attributes, and verifies the sales its ads claim. This maintained matrix compares six networks — Amazon, Walmart Connect, Criteo, Instacart, Roundel, and Albertsons Media Collective — across attribution model, windows, sales-metric definition, clean rooms, incrementality, and MRC accreditation, with a source for every cell.
Every network grades its own homework. The only honest comparison labels who publishes their methodology, who doesn't, and what the independent auditor has actually verified.
Fast read
- What it is
- A maintained, per-cell-sourced matrix comparing how six retail media networks define, attribute, and verify the sales their ads claim.
- Networks
- Amazon Ads, Walmart Connect, Criteo (Commerce Max), Instacart Ads, Roundel (Target), Albertsons Media Collective.
- Columns
- Attribution model, attribution windows, sales-metric definition, clean room, incrementality offering, MRC accreditation status.
- The headline
- Every MRC accreditation in retail media covers delivery metrics — impressions, clicks, viewability. No network has an accredited attributed-sales metric.
- Honesty policy
- Every sourced cell links its source. Where a vendor publishes nothing, the cell says "not publicly documented" — it is not guessed.
- Watch-out
- ROAS is not comparable across networks. Models, windows, and halo scope differ enough that identical campaigns report different numbers.
The measurement matrix.
Six networks, six dimensions. Sourced cells link their source inline. Cells marked not publicly documented mean exactly that — the vendor publishes nothing primary; claims that circulate only in vendor or agency content are flagged vendor-asserted or secondary-sourced. Last validated 2026-09-08.
| Network | Attribution model | Windows | Sales-metric definition | Clean room | Incrementality | MRC accreditation |
|---|---|---|---|---|---|---|
| Amazon Ads | Last-touch, click-based. From Jan 1, 2026, view-through moved to an ML "shopping-signal enhanced last-touch" model — views earn credit only in "discovery moments." Amazon | 14-day click, standard reporting. Amazon Post-2026 view window is ML-determined and shorter — the number is not publicly documented. SP 7-day (sellers) vs 14-day (vendors) click split is secondary-sourced only. Feedvisor AMC allows custom lookbacks. AMC | "Total Sales" = promoted (tracked) ASINs + brand halo. Amazon Halo scope differs by ad product — SP counts same-seller products; SB/SD count any same-brand product from any seller — secondary-sourced; Amazon's help page documents the Total construction (DSP-scoped), not the per-ad-product split. Feedvisor No single public formula. | Amazon Marketing Cloud (AMC) — the category-defining clean room; built on AWS Clean Rooms. Amazon | AMC holdout / lift analyses; Brand Lift surveys with a published methodology page. Amazon | Accredited: SP clicks (total) + SIVT detection; Amazon DSP rendered impressions, clicks, viewability (per-supply-source caveats). MRC letter MRC No attributed-sales metric accredited. |
| Walmart Connect | Post-click last-touch in advertiser reports; view-through exists. Walmart The circulating "unified MTA across onsite / offsite / in-club" claim has no public methodology document — vendor-asserted. In-store matching via the same card used online and in-store, plus Walmart+ accounts: Intentwise | Walmart's public help: default 14-day post-click, selectable 3- and 30-day. Walmart Secondary sources widely claim a 3-day default for US Sponsored Search — conflicts with the public doc; verify in console. | "Total Attributed Sales Revenue," including in-store attributed sales and same-brand "Other Sales" halo. Walmart Flywheel Per-product halo scope: not fully publicly documented. | Scintilla (formerly Walmart Luminate; rename announced Oct 2024, effective 2025); Insights Activation feeds Scintilla data into Walmart Connect campaigns. Walmart | Incremental sales measurement announced for Sponsored Search. Walmart Methodology detail: not publicly documented. | Accredited Sept 26, 2025 (announced Oct 16, 2025): Sponsored Search first-party impressions (display + video), clicks, CTR incl. GIVT filtration across Sponsored Products, Sponsored Brands and Sponsored Video placements. Walmart MRC letter MRC Onsite Display impressions + viewability (incl. SIVT via HUMAN) in MRC audit since Nov 2025 — under review, not accredited. MRC Attributed sales not in scope. |
| Criteo (Commerce Max / Yield) | Single-touch with published priority rules — clicks beat impressions, tighter product match beats looser, then the interaction closest to purchase — within configurable windows; three match scopes per account type (brand accounts: same product / same category / same brand; seller accounts: same product / same category / same seller). Still the best-documented public attribution logic of the six. Criteo | Configurable post-click 7/14/30-day; post-view none/1/7/14/30-day, capped at the click window. Criteo Offsite default: 30-day post-click, 30-day post-view. Criteo The onsite Sponsored Products default (30-day click / 1-day view at the July 2026 pass) now sits behind a login-gated help page — no longer publicly documented. Retailers can override — "Criteo-powered" RMNs differ by retailer. | Attributed sales per configurable match scope (same SKU vs category vs brand) — the configurability is itself a comparability trap. Criteo | No proprietary clean room; closed-loop offering via LiveRamp partnership (launch partner: Boots Media Group). PMW | Incrementality Testing (user-split or geo-split holdouts) plus "Real-time Incrementality" — an always-on directional iROAS metric, vendor-defined, proprietary. Criteo | Accredited Mar 2024 (first retail-media tech accreditation): display rendered impressions + clicks, Onsite Sponsored Products + Onsite Display, incl. GIVT. Offsite and video not in scope. Criteo MRC letter MRC No sales metrics. |
| Instacart Ads | Linear multi-touch by default (since Aug 1, 2022; refined Apr 1, 2025 so that on a path mixing clicks and views only the clicks share the even split) — the outlier among the six. Last-touch view available for comparison. Instacart | 14-day default, matched to weekly-basket shopping behavior. Instacart Perpetua | "Direct sales" = the campaign's products; same-brand halo reported separately as Halo ROAS — a cleaner split than Amazon's blended Total Sales. Instacart | Data Hub clean room, built on Snowflake (documented since Aug 2024; public without login since Jan 2026): join first-party data with Instacart purchase signals; Overlap, LTV, New-to-Brand, Purchase Frequency, Path-to-Purchase (Mar 2026) and Spend-Based Segmentation (Jul 2026) templates. Public docs. Instacart | Sales lift testing — "gold-standard experimentation" in Instacart's words; the test design and the "15%+ average incremental sales lift" figure cited at the July 2026 pass are no longer on the page — methodology not publicly documented. Instacart | Accredited Mar 2024 (Marketplace), expanded Nov 2025 to Carrot Ads partners, continued by MRC letter of Jun 1, 2026: impressions, clicks, CTR, viewable impressions across Sponsored Product, Display, Shoppable Display, Shoppable Video. PR Instacart MRC letter No sales metrics. |
| Roundel (Target) | Closed-loop matching of ad exposure to online and in-store purchases via Target first-party guest data covering online and in-store purchases; Roundel describes "multi-touch attribution." Roundel Per-touch credit rules: not publicly documented. | Not publicly documented on Roundel's site. The 30-day click / 14-day view defaults reported by a secondary source at the July 2026 pass are unverified — that source is now offline; confirm in your agreement. | Attributed sales spanning online and in-store via guest identity matching. Roundel Granular definition (halo scope, channel weighting): not publicly documented. | Secure data matching against Target first-party data. Roundel A self-serve clean-room product: not publicly documented. | Matched Market Model (geo holdout, reported launched 2025); incrementality positioned as the "North Star" metric — secondary-sourced; original source offline. | Not listed in the MRC digital accreditation directory as of September 2026. MRC Roundel is a named contributor to the IAB/MRC retail media guidelines. IAB/MRC |
| Albertsons Media Collective | Match-based attribution via loyalty data; public materials emphasize incrementality over attribution credit. Albertsons Per-touch credit rules: not publicly documented. | Not publicly documented in public docs. Secondary research reports 14-day as standard, with windows varying to 28 days — secondary-sourced. Ovative/Kellogg | iROAS-forward reporting. Its own March 2026 research with Ovative Group and Kellogg professors: iROAS can vary 6.5x — and flip results in 83% of campaigns — on filtering, matching, and revenue-calculation choices alone. MediaPost | LiveRamp clean-room collaboration (early adopter); multi-party clean room with Omnicom built on Snowflake + VideoAmp. LiveRamp P2PI | In-store matched-market incrementality (Jan 2026, ~60 store-matching variables) and onsite incrementality (Apr 2026) — store-level test vs control. Albertsons Albertsons | Not listed in the MRC digital accreditation directory as of September 2026. MRC Drafted an industry standardization framework naming third-party verification a priority. whitepaper |
Maintenance note
This table is maintained, not archived. Each cell is re-resolved against its linked source on every validation pass; the LAST_VALIDATED stamp above and in the footer is the date of the most recent pass. Vendor help-center URLs move — one Instacart help URL returned 404 at the July 2026 pass and was replaced with the company documentation page. At the September 2026 pass three cited URLs had died — Criteo's MRC help page, Instacart's display-analytics help page, and the Roundel secondary — one went login-gated (Criteo's campaign attribution settings), and two Amazon blog posts now redirect to product pages; each was re-pointed to a primary document or its cell was downgraded. If a linked source no longer supports its cell, the cell changes.
How to read this table.
The matrix looks comparable because the columns line up. The definitions underneath do not. Six traps account for most cross-network misreads.
- 01
"Total Sales" is not one number
Amazon blends promoted ASIN + brand halo into one figure (and halo scope differs by ad product). Instacart splits Direct vs Halo ROAS. Walmart folds in in-store "Other Sales." Same column header, three different quantities.
- 02
View-through defaults do silent work
Criteo's onsite default added 1-day post-view credit when last publicly documented; its offsite default is 30-day post-view. Amazon's view window is now ML-determined and undisclosed. Roundel's reported 14-day view default is unverified. Always ask what share of attributed sales is view-through.
- 03
Configurable windows break comparability
Criteo windows are retailer-overridable (7/14/30, click and view), so two Criteo-powered RMNs can report the same campaign differently. Walmart lets advertisers select 3, 14, or 30 days.
- 04
Model heterogeneity
Instacart distributes credit linearly across touchpoints; everyone else is last-touch. Identical media plans produce structurally different ROAS. Cross-network ROAS league tables are definitionally broken.
- 05
Accreditation covers delivery, never outcomes
Every MRC accreditation in this matrix covers impressions, clicks, CTR, or viewability — scoped to specific products and channels. Zero attributed-sales metrics are accredited. Treat any "MRC-accredited" claim as scoped until you have read the scope.
- 06
iROAS is methodology-sensitive
Albertsons Media Collective's own research (vendor-sourced): iROAS varies up to 6.5x, and results flip in 83% of campaigns, purely on test/control filtering, matching, and revenue-calculation choices.
The accreditation column, spelled out.
In retail media the media owner, the measurement provider, and the owner of the sales data are usually the same company. MRC accreditation is the only independent audit in this matrix — which makes its scope the most load-bearing fine print in the category. Four of six networks hold accreditations. All four cover delivery: rendered impressions, clicks, CTR, viewability, invalid-traffic filtration. None covers attributed sales, ROAS, or incrementality — per the MRC digital accreditation directory as of September 2026. The pipeline has started to move, though: Kroger has submitted attributed sales, incremental sales, aROAS and iROAS for MRC pre-audit (in process as of June 2026), and Kevel's pre-audit scope includes conversions and incrementality — under review, none accredited.
What this means in practice
When a network says "MRC-accredited," the audited claim is that an ad rendered, was viewable, or was clicked — measured to MRC standards within a stated scope. The sales number attached to that ad remains self-reported, under a methodology the network chose and, in several cells above, does not publish. The IAB/MRC Retail Media Measurement Guidelines (January 2024) exist precisely to push definitions, disclosure, and audit toward the outcome metrics — adoption is the open question this page tracks.
Questions to ask your RMN.
Seven definitional questions grounded in the IAB/MRC guidelines, then three grounded in the IAB incrementality guidelines and their three principles. Ask them before the annual commit, not after the QBR.
- Which attribution model produces my reported sales — last-touch, linear, ML-weighted? Is the methodology publicly documented, or only asserted?
- What are the exact click and view windows on my account — and can the retailer or platform change them without notice?
- What is inside the "sales" number: promoted SKU only, brand halo, same-seller vs 3P, in-store? Show the formula per ad product.
- What share of attributed sales is view-through vs click-through?
- Where do your metric definitions deviate from the IAB/MRC Retail Media Measurement Guidelines?
- Which metrics are MRC-accredited, at what scope (ad product, channel, device) — and is attributed sales among them? (As of September 2026, it never is.)
- Can I export event-level or clean-room data to reproduce your number independently?
- 08
Credible counterfactual
Is lift measured against a true holdout or control — user-split, geo-split, matched market — or against a model proxy? Which of the four IAB methodology families is this?
- 09
Control of bias
How are test and control matched, which filters are applied to transactions and audiences, and who chose them — you or the party being graded?
- 10
Signal vs noise
What is the statistical confidence, and would the result survive a different revenue-calculation choice? (Albertsons’ own research says it often does not.)
Comparability is a contract term, not a dashboard feature.
No network will normalize its definitions for you — halo-inclusive sales and configurable windows report bigger numbers. The leverage sits in procurement: pin the model, the windows, the sales-metric formula, and the view-through share in the agreement; require export or clean-room access to reproduce the number; and price any metric the vendor cannot document publicly as exactly what it is — unverified.
Frequently asked questions.
Which retail media networks have MRC accreditation?
As of September 2026: Amazon (Sponsored Products clicks plus SIVT detection; Amazon DSP rendered impressions, clicks, and viewability), Criteo (March 2024 — onsite display rendered impressions and clicks on Commerce Max / Commerce Yield), Walmart Connect (September 2025 — Sponsored Search impressions, clicks, and CTR), and Instacart (March 2024 for the Marketplace, expanded November 2025 to Carrot Ads — impressions, clicks, CTR, viewable impressions). Roundel and Albertsons Media Collective are not listed in the MRC digital accreditation directory.
Is any RMN attributed-sales or ROAS metric MRC-accredited?
No. Every retail media MRC accreditation to date covers delivery metrics — impressions, clicks, CTR, viewability — scoped to specific ad products and channels. No network has an accredited attributed-sales, ROAS, or incrementality metric as of September 2026; the first outcome metrics in MRC's pipeline are Kroger's attributed sales, incremental sales, aROAS and iROAS, in pre-audit since June 2026. That is the single most important scope limit to keep in mind when a network cites its accreditation.
Can I compare ROAS across Amazon, Walmart, and Instacart?
Not directly. Amazon reports last-touch Total Sales including brand halo; Walmart reports Total Attributed Sales Revenue including in-store halo with selectable 3/14/30-day windows; Instacart distributes credit linearly across touchpoints and splits halo into a separate metric. Different models, windows, and sales scopes mean identical campaigns produce structurally different ROAS. Normalize definitions first, or compare via independent incrementality tests.
What is the difference between attributed sales and incremental sales?
Attributed sales assign credit to ad exposures under a chosen rule (last-touch, linear, ML-weighted) — including purchases that would have happened anyway. Incremental sales estimate the causal lift versus a counterfactual: a holdout, geo split, or matched market. The IAB Guidelines for Incremental Measurement in Commerce Media (November 2025) define four methodology families and rest on three principles: credible counterfactuals, control of bias, and separation of signal from noise.
Why do some cells in this matrix say "not publicly documented"?
Because that is the accurate answer. Several networks do not publish attribution windows, view-weighting methodology, or halo-scope formulas. Where a claim circulates only in vendor or agency content without a primary document, this matrix labels it vendor-asserted or secondary-sourced instead of presenting it as fact. Treat those cells as questions for your RMN contract negotiation.
Negotiating measurement terms with an RMN?
Bring the checklist. Normalize the definitions, pin the windows in the contract, and separate accredited delivery metrics from self-graded sales claims before committing budget.