Closed-Loop Attribution in Retail Media: How Walmart, Amazon & Instacart Measure True Lift (2026)

By Najfee HyderLast updated: May 2026

Series context — This is a spoke article in Osmos's Attribution & Measurement pillar. Parent hub: Closed-Loop Attribution in Retail Media: The 2026 Measurement Playbook →. Sibling spoke: Closed-Loop Attribution: The Key to Unlocking Higher ROAS →. Read this if you specifically need vendor-by-vendor depth on Walmart Connect, Amazon AMC, and Instacart Carrot Insights. For the broader framework, read the hub first.

Walmart Connect, Amazon Ads, and Instacart Ads all advertise "closed-loop attribution" — but they solve the problem in fundamentally different ways. Walmart Connect runs a Multi-Touch Attribution (MTA) model unified across onsite, offsite, and in-club touchpoints, with Walmart Luminate (rebranding to Scintilla) sitting behind it as the clean room (Walmart Corporate, 2025; Chain Store Age, 2025). Amazon combines Amazon Marketing Cloud (AMC) — a SQL-based clean room whose ad-traffic lookback jumped from 13 to 25 months at unBoxed 2025 — with Amazon Attribution, a tag-based measurement product for non-Amazon channels (Amazon Ads, 2025; Amazon Ads). Instacart has used linear multi-touch attribution as its default since August 2022, layers basket-level reporting on top, and in January 2026 launched Data Hub — its first clean-room equivalent (Instacart Ads Help Center; Instacart IR, 2026). The three architectures look similar on a pitch deck and diverge sharply in the weeds — which is why attribution in retail media has a structurally different problem shape from traditional digital.

This article is for operators choosing between — or reconciling reporting across — Walmart Connect, Amazon Ads, and Instacart Ads. It covers each network's default attribution model, clean-room architecture, API surface, offsite and in-store coverage, lookback windows, and access tier, then closes with the operator problem: how do you normalize the three when every campaign touches more than one?

The Vendor Comparison Matrix

The short answer to "how do these three compare?" is this matrix. The rest of the article explains each row in context.

Definition — Closed-loop attribution. Closed-loop attribution is the ability to tie an ad exposure directly to a downstream purchase using the retailer's own transactional data, so ROAS is measured on actual sales rather than inferred from modeled conversions.

Definition — Data clean room. A data clean room is a secure, privacy-safe environment where advertisers can join their own first-party data against a retailer's pseudonymized signals for analytics or audience creation — without either party seeing the other's raw customer records.

Definition — Multi-touch attribution (MTA). MTA is an attribution model that distributes conversion credit across multiple ad touchpoints a shopper encountered on the path to purchase, rather than assigning 100% to the last click or last view.

Definition — Lookback window. The lookback window is how far back in time the platform will still count an ad touchpoint toward a conversion. A 14-day click lookback means a purchase counts as ad-driven if the shopper clicked an ad within the past 14 days.

With those primitives in place, here's each network in depth.

Walmart Connect: MTA + Luminate + In-Store MMM

Walmart Connect's measurement stack has three layers. The top layer is the Multi-Touch Attribution (MTA) model, which the company describes as "unified measurement across all channels, from search to display, offsite and in-club. Powered by AI, MTA helps brands understand how every touchpoint contributes to conversion" (Walmart Corporate, 2025). In practice this is Walmart's answer to the old complaint that Sponsored Search last-click views and offsite display performance lived in separate reports.

The default sponsored-search windows

For Sponsored Search specifically, Walmart Connect applies a 3-day click attribution window by default, with a 14-day option (Empower Media, 2025). The 3-day default is tuned for CPG consideration cycles — grocery and household items shoppers decide on quickly. For longer-consideration categories (electronics, apparel), the 14-day window is closer to what Amazon uses and what most advertisers benchmark against.

Luminate (rebranding to Scintilla)

One layer down is Walmart Luminate, now rebranded as Scintilla. Luminate is Walmart's shopper-insights platform — basket-level transactional data (not receipt scans), digital and in-store shopper behavior, loyalty-vs-switching signals, and customer-segment analytics. The 2025 change that matters for measurement is the Luminate Insights Activation — a self-serve module that pipes Luminate signals directly into the Walmart Connect campaign platform for Charter suppliers (Retail TouchPoints, 2025). That's how basket data closes the loop on campaign performance, not just describes shoppers in the abstract. Walmart reports the Luminate client base grew 173% versus 2023, with SMB representation up more than 100% YTD (Chain Store Age, 2025) — the audience is no longer only CPG giants.

In-store attribution

The bottom layer is Walmart's in-store attribution, which is structurally different from onsite or offsite digital attribution and worth understanding on its own. Walmart Connect exposes three distinct metrics (Intentwise, 2025):

• In-store advertised sales — a shopper clicks an ad for a specific product and later buys that same product in-store.
• In-store other sales — a shopper clicks an ad but buys a different product from the same brand or category in-store (the halo).
• In-store attributed sales — the combined total of the two above, which is how brands should report "true" in-store contribution.

Unlike onsite ROAS, these metrics are derived from Walmart's loyalty signal and transactional matching — not a last-click cookie — which is what allows them to work in the physical store at all. Read alongside Walmart's offsite DSP numbers, the result is a genuinely omnichannel picture, which is why Walmart positions MTA as the glue.

The API surface

Programmatic access is via the Walmart Connect Sponsored Search APIs, including the Snapshot Reports endpoints (Walmart Developer). This is where agencies and retail-media ops teams pull daily reporting into their own warehouses to reconcile Walmart data against Amazon and Instacart. Luminate Insights Activation is separate — it's not (yet) a SQL-on-Luminate experience, more an insights-as-feature activation inside Walmart Connect. This is the biggest architectural difference vs AMC, which is SQL-first by design.

Amazon: AMC Clean Room + Amazon Attribution

Amazon splits the measurement problem into two products. AMC is the clean room for analyzing Amazon-attributable behavior in depth. Amazon Attribution is the tag-based product that measures non-Amazon campaigns that drive to Amazon. Confusingly, both ladder up to "closed-loop attribution" in Amazon's marketing, but they do different jobs. Before going deep on AMC, it's worth disambiguating the full set of products that Amazon advertisers routinely confuse with one another.

Amazon's Five Attribution Tools, Disambiguated

The single most consequential recent change came on January 1, 2026, when Amazon replaced its long-standing 14-day "blind" view-through window with a "shopping-signal enhanced last-touch attribution model" for Sponsored Brands, Sponsored Display (on a vCPM basis), and Amazon DSP campaigns in the Amazon Store. In Amazon's own words, the new model "focuses on moments of brand discovery along the shopper path, giving credit to ad views that reach shoppers during early discovery moments such as exploratory browsing or general category searches, and applies a shorter attribution window to reflect shopper behavior in the store" (Amazon Ads, January 2026). The new window is a shorter, ML-determined window — Amazon has not disclosed the exact length. Click-based attribution is unchanged, and Amazon preserved the legacy 14-day view data as a "Purchases (all views)" metric for historical comparison. Advertisers running view-based display and video should expect what one agency analysis called "double-digit declines in attributed DSP revenue" — a measurement methodology change, not a sales regression, because "if the system determines a shopper was already likely to purchase, the ad view may no longer receive credit" (Code3, March 2026; PPC Land, January 2026).

That change lands across several distinct Amazon products that buyers routinely treat as one. The high-confusion query space — "amazon attribution api," "attribution amazon," "amazon retail media measurement" — comes from the fact that Amazon ships five overlapping things under the same conceptual umbrella. The table below disambiguates them.

The five tools sit on a spectrum from automated-default to SQL-native analyst. The Ads console (tool 1) is what every Sponsored Products advertiser sees with no setup — Amazon's default last-click surface, always on. Amazon DSP (tool 2) sits adjacent but in a separate reporting interface and billing relationship, handling programmatic formats and requiring DSP access that Sponsored-Ads-only advertisers do not have. Amazon Attribution (tool 3) is opt-in and tag-based — advertisers running Google Ads or Meta campaigns that drive to Amazon must actively implement it to close the loop on those external channels. AMC (tool 4) is the clean-room SQL layer above all three; since September 2025 it is accessible to all Sponsored Ads advertisers (Skai, 2025). Amazon Retail Media (tool 5) is a category name, not a product.

The most common buyer confusion is treating "Amazon Attribution" (tool 3, the off-Amazon cross-channel tag) and "Amazon Retail Media" (tool 5, the umbrella) as synonyms, then conflating both with AMC (tool 4, the clean room). They are three distinct things at three distinct layers. Amazon Attribution closes the loop on external channels driving to Amazon; AMC closes the loop on Amazon-internal, multi-touchpoint user-path analysis. Note too that "Amazon Attribution" is not itself an API — programmatic access to Amazon Ads reporting runs through the Amazon Ads API and Amazon Marketing Stream, while AMC has its own query endpoints. For how Amazon's five tools compare to Walmart Connect and Instacart at the vendor layer, the cross-vendor attribution comparison across all major retail media networks covers the full framework.

Amazon Marketing Cloud (AMC)

AMC is, in Amazon's own words, "a secure, privacy-safe, and cloud-based clean room solution in which advertisers can easily perform analytics and build audiences across pseudonymized signals, including Amazon Ads signals as well as their own inputs" (Amazon Ads). Three things matter for attribution specifically:

1. SQL is the interface. AMC is SQL-native. Amazon publishes an Instructional Query Library — more than 60 pre-written SQL queries covering path-to-conversion, reach-and-frequency, new-to-brand, custom attribution, and audience construction (Amazon Ads, 2025). A typical "Path to Conversion by Campaign Groups" query returns metrics including path occurrences, reach, impressions, total cost, user purchase rate, and new-to-brand rate by campaign combination (Adbrew, 2025). Advertisers can run instructional queries as-is, customize them (many teams remove the default consecutive-duplicate-ad dedup logic to see the fuller path), or write fully custom SQL.

2. Lookback windows just got much longer. At unBoxed 2025 on November 11, Amazon extended the AMC ad-traffic lookback from 13 months to 25 months (Amazon Ads, 2025; PPC Land, 2025). It also unlocked 5-year historical access for video viewership, sales signals, and Amazon Store purchase signals. The expansion works through existing AMC endpoints — no code changes required. US and Canada launched the same day. The UK, Germany, Italy, France, Austria, Turkey, and Japan roll out in Q1 2026. For advertisers building lift models, reach-frequency curves, or cohort analyses, this is the biggest measurement change in retail media in 2025.

3. Amazon Attribution handles offsite. AMC analyzes Amazon-side behavior. Amazon Attribution is the measurement tag you append to non-Amazon ad click URLs — Google Ads, Meta, display, email, influencer links — so purchases on Amazon driven by those external clicks are reported back to the advertiser (Amazon Ads). The attribution window is 14 days click. Metrics reported include detail-page views, add-to-carts, purchases, and sales. This is Amazon's closed-loop answer for the offsite problem — because, as industry observers have noted, "Amazon and Walmart don't allow Meta or Google to place conversion tags on their sites, making it very difficult to capture the sales that a campaign drove" (Incremental, 2025) without a vendor-provided solution.

AMC example: a lightweight path-to-conversion query

A simplified version of the path-to-conversion use case — for illustration, not drop-in production code — returns a distribution of ad-path combinations and downstream purchase metrics:

SELECT path, COUNT(DISTINCT user_id) AS reach, SUM(impressions) AS impressions, SUM(total_cost) AS spend, SUM(purchases) AS purchases, (SUM(purchases) / COUNT(DISTINCT user_id)) AS user_purchase_rate FROM path_to_conversion_by_campaign GROUP BY path ORDER BY purchases DESC

In practice the production instructional query handles consecutive-duplicate dedup, NTB flagging, and time-ordering (Adbrew, 2025). The point is that advertisers can see which combinations of DSP + Sponsored Display + Sponsored Products touchpoints actually drove conversions, rather than accepting Amazon Ads console's default last-click attribution.

AMC access

Access has broadened. AMC was historically gated to DSP advertisers and large brands with data teams. Through 2025, Amazon expanded AMC availability to all Sponsored Ads advertisers, removing the biggest practical barrier (Skai, 2025; eMarketer, 2025). Paired with the new AI-powered Ads Agent that lets users ask AMC questions in natural language instead of writing SQL, the effective access tier for AMC is now much wider than it was 12 months ago. For methodology on layering hold-out tests on top of these reported paths, see how closed-loop attribution complements incrementality testing for true ROAS measurement — AMC is excellent for observed-path analytics, but true lift requires a designed experiment.

Instacart: Carrot Insights + Data Hub + Ads API

Instacart's measurement story is often underestimated because it lacked a clean-room product until January 2026. But Instacart has quietly had one structural advantage over Amazon and Walmart since 2022: multi-touch attribution is the default, not an advanced option.

The default model: linear multi-touch, since 2022

"On 8/1, Instacart launched a multi-touch attribution model and an improved last touch attribution model. Multi-touch attribution distributes purchase credit across all relevant advertising touchpoints in a consumer's journey, which is the method now used by default" (Instacart Ads Help Center). Both Ads Manager and the Ads API use linear MTA as the default reporting view, with last-touch available as a comparison. That's architecturally similar to where Walmart Connect's MTA landed in 2025.

Closed-loop, basket-level, by design

Because Instacart is the end-to-end transaction (ad + purchase + delivery), closed-loop attribution is native — there is no "did the ad on Instacart drive a purchase at a different retailer" version of the problem. Instacart's measurement surface is basket-level: metrics like Basket Penetration (percentage of all Instacart baskets containing a brand's product) and Basket Share (percentage of unique deliveries containing a brand's product within a selected category) are standard reporting columns. The Consumer Insights Portal, launched in September 2025, gives marketing teams self-serve access to SKU-level performance, search behavior, substitution patterns, and promotion impact based on actual transactions (Instacart IR, 2025). This is Instacart's equivalent of the Walmart Luminate dashboard experience — an insights layer, not a clean room.

Data Hub — Instacart's clean room, debuted CES 2026

The missing piece for a long time was clean-room parity with AMC. Instacart closed that gap at CES 2026 with Data Hub. In the company's own framing: "Data Hub gives CPG brands and their agency partners secure, privacy-safe access to Instacart's rich first-party grocery data. Data Hub has been in pilot with a select group of agencies and CPGs and will expand to additional partners throughout 2026" (Instacart IR, 2026). This puts Instacart on the same shelf as Amazon Marketing Cloud and Walmart Luminate in terms of category — advertisers can run joint analyses against their own first-party data without seeing Instacart shopper PII, and vice versa.

The practical caveat is that Data Hub is still pilot-phase for most advertisers. In Q2 2026, AMC's SQL-native self-serve experience and its newly expanded 25-month lookback are substantially ahead of what Instacart Data Hub offers to the median advertiser. That gap will narrow through 2026 as Instacart expands access.

The Carrot Ads API

Instacart's programmatic surface is the Carrot Ads API, documented at docs.instacart.com/ads. The Ads API exposes sponsored products, display, and shoppable formats, with reporting endpoints that return the same linear-MTA conversion data advertisers see in Ads Manager. Because Instacart began white-labeling Carrot Ads for off-platform retailers in 2025 (AdExchanger, 2025), the Carrot Ads API is now the access point for measurement across both Instacart-proper and partner-retailer Instacart-powered networks.

Limitations vs Amazon — honestly

Advertisers evaluating Instacart against Amazon on measurement should weigh four gaps. First, AMC's SQL self-serve is more mature than Data Hub's pilot. Second, Amazon's new 5-year video and sales history has no direct Instacart equivalent. Third, Amazon Attribution provides a dedicated off-Amazon tagging product; Instacart's off-platform expansion is via Carrot Ads-powered networks, which is different in structure. Fourth, Amazon's MRC accreditation is long-established across many formats; Instacart's expanded in November 2025 to cover impressions, clicks, CTR, and viewable impressions on Sponsored Product, Display, Shoppable Display, and Shoppable Video (Instacart IR, 2025). The gap is closing. In the specific dimension of closed-loop attribution on the Instacart shopper itself, Instacart's native multi-touch default is an advantage — buyers just need to factor in where else those dollars are being spent.

Where the Three Networks Diverge (and Where They Converge)

Looking across the matrix and the deep dives, the three networks converge on intent — every one of them offers a closed-loop attribution product — and diverge on four practical dimensions.

1. Default attribution model

Walmart launched unified MTA. Instacart has had linear MTA since 2022. Amazon defaults to last-click in the Ads console but offers custom MTA inside AMC via SQL. For an advertiser running the same campaign across all three, the numbers will not line up unless each is forced to the same model. Amazon's default undercounts upper-funnel contribution relative to Walmart's and Instacart's defaults.

2. Clean-room architecture

AMC is SQL-first, mature, and broadly accessible (Sponsored Ads advertisers can now use it). Walmart Luminate / Scintilla is insights-first, with activation into the Walmart Connect ads platform via Luminate Insights Activation. Instacart Data Hub is the newest of the three and still pilot-phase. The practical implication: if your workflow depends on running custom SQL against retailer-side purchase data, Amazon is the only network where that is first-class today for most advertisers.

3. Lookback windows

AMC's extension to 25 months (and 5 years for video/sales/Store signals) is the longest retail-media lookback now available. Amazon Attribution on offsite is 14 days. Walmart Connect Sponsored Search defaults to 3 days with a 14-day option. Instacart's dashboard windows vary by format and are exposed via the Ads API. Year-over-year trend analysis and long-tail cohort work are simply more feasible on Amazon's stack today than on the other two.

4. Offsite and in-store coverage

Amazon has the most developed offsite product (Amazon Attribution tags on external channels). Walmart has the most developed in-store product (the three in-store sales metrics plus Luminate basket data and in-club / Sam's Club coverage). Instacart is the category specialist in grocery delivery — so offsite and in-store are both less relevant in its native frame, although the Carrot Ads off-platform expansion changes that on the retailer side.

For context on why this matters — the measurement gap that retail media still has not fully closed on offsite and omnichannel — see the analyst perspective from Incremental. The hub article, Closed-Loop Attribution in Retail Media: The 2026 Measurement Playbook, covers the full framework across all retail-media networks.

AI/ML in Retail Media Attribution: 2026 Capabilities

Across 2025 and 2026, both Amazon and Walmart moved attribution away from fixed time windows toward machine-learning-driven, signal-based credit allocation — the most consequential methodology shift in retail media measurement in years. The change shows up in three places: how platforms select and run attribution models, how they forecast lift before campaigns conclude, and how AI augments (but does not replace) incrementality testing.

ML-driven attribution model selection

Amazon's January 1, 2026 view-attribution change is the clearest live example. Rather than crediting every ad view inside a fixed 14-day window, the new "shopping-signal enhanced last-touch" model uses ML to evaluate "placement context, user behavior patterns, and shopping journey stage," concentrating credit on genuine brand-discovery moments (PPC Land, January 2026). Amazon's multi-touch attribution product — in beta (launched at unBoxed 2024, rolling out through 2025–2026) — goes further, using "insights from experiments, machine learning, and historical shopping signals to gauge the influence of every ad engagement on conversion outcomes" across Sponsored Products, Brands, Display, Sponsored TV, and DSP (Amazon Ads, 2024). Walmart Connect's MTA is described publicly only as "Powered by AI" (Walmart Corporate, 2025); the underlying model and credit-allocation mechanics are not disclosed, so the transparency gap between the two platforms is real. For retailers running their own media networks, StratEdge for lift modeling and ROAS forecasting built on ML-normalized attribution signals provides the normalized feed that makes bid strategy and revenue forecasting cross-comparable across these divergent platform models.

The most detailed ML-attribution disclosure to date

Amazon's August 2025 research paper is the most technically detailed retail-media attribution disclosure any major platform has published. Its MTA uses a "Causal Calibration Model" — training on hundreds of thousands of RCT campaigns, building an ensemble of attribution approaches (last-touch, rule-based, and a causal ML model), then disaggregating campaign-level predictions down to individual ad touchpoints. The paper is blunt about why the hybrid is necessary: "ML models trained purely on observational data are easy to scale and can yield precise predictions, but the models might produce biased estimates of ad effects. RCTs yield unbiased ad effects but can be noisy" (arXiv:2508.08209, Amazon Ads Research, August 2025). Observational-only attribution, the same paper notes, can be catastrophically wrong — a cited 2023 Meta study showed "errors in the estimated ad effects ranging from 488% to 948%." The model "captures signals from a variety of sources, including traffic events, bid logs, retail information, ASIN catalogs, and more." The takeaway: RCT results serve as the ground-truth training signal for the ML calibration layer, achieving experimental accuracy at machine-learning scale.

Amazon has also removed the SQL barrier for AMC through generative AI. Building on a natural-language SQL generator first shown at CES 2025 (Amazon Ads, January 2025), the Ads Agent for AMC (announced at unBoxed 2025) translates plain-English questions into AMC SQL — including attribution queries — reducing query development time "from hours to minutes" with "deep knowledge of AMC datasets, syntax, and privacy rules" (Amazon Ads, November 2025). It is available across more than 30 countries per the product page listing.

Predictive lift modeling and AI-augmented incrementality

Industry momentum toward AI-augmented incrementality is clear, but adoption barriers remain high. As of the eMarketer/TransUnion July 2025 survey, 52.0% of US brand and agency marketers use incrementality testing, 27.6% prioritize expanding it, and 36.2% plan investment within twelve months (eMarketer, November 2025). Roughly half of US marketers have adopted AI and machine learning for automated reporting, and 60.9% rank "generative insight summaries" as their top AI enhancement for next-generation marketing mix modeling (eMarketer/Rakuten, October 2025) (eMarketer, April 2026). On the tooling side, AI-driven systems "can automate the creation of test-and-control experiments... [and] recommend optimal control groups" (Dataslayer, 2025), and Skai's attribution forecasting feature offers predicted incrementality metrics based on historical testing data, enabling proactive budget allocation before full results are available (Skai, 2025). Google has even lowered incrementality experiment costs from roughly $100,000 to $5,000 by adopting Bayesian statistical models, opening testing to mid-market brands (eMarketer, April 2026).

But the same surveys show 44% of marketers still cite accuracy and reliability as the top incrementality barrier and 43% struggle to apply it across multiple ad types — ML is augmenting experimental design, not replacing it. Amazon's own arXiv paper makes the point explicitly: RCTs remain the ground truth that calibrates the ML layer. AI tools automate test design and forecast lift, but closed-loop attribution as the foundation for true ROAS measurement remains the feed layer that makes any of it trustworthy.

Normalization Across Networks — The Operator Problem

Every dimension above is legitimate vendor-by-vendor comparison. But the operator problem — the problem faced by a brand spending on all three, or by a retailer running its own media network that needs to benchmark against them — is not "which vendor attribution model is best?" It's "how do I reconcile reports that use different models, different lookback windows, different clean-room architectures, and different API schemas, so a CFO or a planner can compare ROAS apples-to-apples?"

This is where a retail media operating system fits. Osmosphere's retail media operating system is designed to sit above the individual retailer APIs — ingesting reporting from Walmart Connect, Amazon Ads, and Instacart Ads (and from retailers' own networks) into a single operational layer that normalizes metric definitions across the three. ControlHub's unified campaign operations layer handles the day-to-day — pacing, budget routing, inventory, scheduling — so a campaign manager is not flipping between three consoles to understand pacing or overlap. StratEdge for lift modeling and ROAS forecasting runs on top of the normalized feed, which is the only place bid strategy and revenue forecasting make sense, because only the normalized feed is cross-comparable.

Osmos is deliberately not itself a clean room, not a DSP, and not a competitor to AMC, Luminate, or Data Hub. It's the layer that takes what those products produce and makes it operational for a single campaign team or a retailer running a media network.

Frequently Asked Questions

How do Walmart Connect, Amazon AMC, and Instacart each define "closed-loop attribution"?

All three tie ad exposures to actual first-party purchase records, but the mechanism differs. Walmart Connect uses Multi-Touch Attribution across onsite, offsite, and in-club touchpoints, tied to Walmart's transactional data via Luminate/Scintilla. Amazon uses AMC for on-Amazon attribution (SQL queries against pseudonymized Amazon Ads signals) and Amazon Attribution tags for off-Amazon campaigns that drive to Amazon detail pages. Instacart uses linear multi-touch attribution by default, with basket-level closed-loop reporting against Instacart's first-party transactional data.

How much does AMC cost? Is it free?

Amazon has expanded AMC access considerably. Through 2025, AMC became available to all Sponsored Ads advertisers — previously gating was tighter and effectively limited to DSP advertisers and larger brands with dedicated data teams (Skai, 2025; eMarketer, 2025). Amazon does not publish a standalone AMC price list; the practical answer is that the measurement-use access barrier has effectively been removed for most advertisers who already run Sponsored Ads.

What lookback windows do Amazon, Walmart, and Instacart use?

Walmart Connect Sponsored Search defaults to a 3-day click attribution window with a 14-day option (Empower, 2025). Amazon Attribution uses a 14-day click window for offsite measurement (Amazon Ads). AMC's ad-traffic lookback is now 25 months, with 5-year historical access for video viewership, sales, and Amazon Store purchase signals (Amazon Ads, 2025). Instacart's windows vary by format and are exposed via the Carrot Ads API.

What is Walmart Luminate and how do I access it?

Walmart Luminate (rebranding to Scintilla in 2025) is Walmart's shopper-insights platform — basket-level transactional data, digital and in-store shopper behavior, and segment analytics. Access is tiered: Charter suppliers get access to the full platform including Luminate Insights Activation, which pipes Luminate signals into the Walmart Connect campaign platform for self-serve activation (Retail TouchPoints, 2025). Walmart does not publish public pricing.

What is Instacart Data Hub and when is it available?

Instacart Data Hub is Instacart's clean-room solution, launched at CES 2026. It gives CPG brands and agency partners secure, privacy-safe access to Instacart's first-party grocery data so they can run joint analyses against their own first-party data. As of early 2026 it's in pilot with a select group of agencies and CPGs, and Instacart has announced broader partner expansion throughout 2026 (Instacart IR, 2026).

What is the Amazon Attribution API, and is it the same as Amazon Attribution?

They are not the same thing — and the confusion is common. Amazon Attribution (the measurement product) is a tag-based tool: advertisers append click-tracking tags to URLs in Google Ads, Meta, or email campaigns, and it reports the purchases on Amazon driven by those external clicks. It is not itself an API. Programmatic access to Amazon Ads reporting — including Amazon Attribution data — runs through the Amazon Ads API v3 and Amazon Marketing Stream, and AMC has its own separate query endpoints. So there are three distinct access layers: (1) Amazon Attribution, the off-Amazon measurement product (tag-based); (2) the Amazon Ads API, the programmatic reporting access layer for Amazon Ads products; and (3) the AMC API, the clean-room query access layer. Searching for an "Amazon Attribution API" usually means you want programmatic access to Amazon Attribution data — which is delivered via the Amazon Ads API, not a standalone product by that name.

How do I normalize reporting across Walmart, Amazon, and Instacart?

Four steps. First, fix the attribution model — force all three to the same methodology (most commonly linear MTA) so percentages are comparable. Second, fix the lookback — choose one window (14 days is a common middle ground) and apply it across all three. Third, fix the metric definitions — impressions, clicks, conversions, ROAS, and view-through are defined slightly differently across vendors; normalize to a single schema. Fourth, ingest via a platform like Osmosphere that handles steps one through three automatically, rather than rebuilding the reconciliation logic in a spreadsheet every month.

Can I run AMC SQL queries against Walmart or Instacart data?

No. AMC SQL queries only run against Amazon Ads and advertiser first-party data inside AMC. Walmart's equivalent is Luminate Insights Activation (insights-first, not SQL-first). Instacart's clean-room equivalent is Data Hub, now in pilot, which will support its own query surface. Running a single SQL query across all three retailer-side datasets requires either a neutral clean-room vendor or an operator layer that pulls normalized outputs from each.

Which network is best for offsite retail media attribution?

Amazon's Amazon Attribution product is the most mature closed-loop solution for offsite channels today — it measures paid search, social, display, video, and email campaigns that drive to Amazon. Walmart's MTA incorporates offsite within the Walmart Connect DSP ecosystem. Instacart's offsite story is newest, partly driven by Carrot Ads going off-platform to partner retailers. For advertisers whose offsite spend is larger than their onsite spend, Amazon Attribution is the current best-in-class; for fully omnichannel programs, MTA's unified measurement across channels is the more complete frame, but fewer advertisers have the Luminate access to operate it at depth.

Are AI-powered attribution models replacing traditional holdout experiments in retail media?

No — ML attribution models are augmenting experimental design, not replacing it. Amazon's own research (arXiv:2508.08209, August 2025) makes this explicit: its Causal Calibration Model trains on hundreds of thousands of RCT campaigns precisely because observational ML alone produces severe bias (the cited 2023 Meta study showed errors of 488%–948% in ad effect estimates). ML solves the scale problem; RCTs solve the bias problem. The retail media industry reflects this tension: 52.0% of US brand and agency marketers now use incrementality testing (eMarketer/TransUnion, July 2025), but 44% still cite accuracy and reliability as the top barrier — meaning even with ML tools, designed holdout experiments remain the gold standard for unbiased lift measurement.

Sources

1. Amazon Ads — Unlock insights with AMC's expanded ad traffic lookback window, November 11, 2025.
2. PPC Land — Amazon extends Marketing Cloud lookback window from 13 to 25 months, November 12, 2025.
3. Amazon Ads — Amazon Marketing Cloud: Discover advertising insights.
4. Amazon Ads — Amazon Attribution: Get started with advertising analytics.
5. Amazon Ads — Instructional Queries now available in Amazon Marketing Cloud, June 2025.
6. Adbrew — Amazon Marketing Cloud Path to Purchase Insights, August 2025.
7. Skai — Amazon Ads unBoxed 2025: Full-Funnel, Relevant, and Measurable, November 2025.
8. Walmart Corporate — Elevating Retail Media to Retail Experiences, April 29, 2025.
9. Retail TouchPoints — Walmart Makes Shopper Insights Available Via its Connect Retail Media Platform, June 2025.
10. Chain Store Age — Walmart expands, renames Luminate data platform, September 2025.
11. Intentwise — Walmart Connect in-store attribution: What to know, July 2025.
12. Empower Media — Making Sense of Sponsored Search on Amazon and Walmart: Part 2 – Attribution, September 2025.
13. Walmart Developer — Introduction to Walmart Connect – Sponsored Search APIs.
14. Instacart Docs — Carrot Ads API documentation.
15. Instacart Ads — Linear attribution reporting.
16. Instacart Investor Relations — Instacart Debuts Data Hub, a Clean Room Offering for Enhanced Media Performance, January 2026.
17. Instacart — Measuring Your Multi-Tactic Ad Strategy on Instacart, October 2025.
18. Instacart Investor Relations — Instacart Launches Consumer Insights Portal, September 2025.
19. AdExchanger — One Giant Leap For Cartkind: Instacart Takes its Ad Tech Off-Platform For The First Time, November 2025.
20. Instacart Investor Relations — Instacart Receives Expanded MRC Accreditation for Carrot Ads, November 2025.
21. eMarketer — FAQ on data clean rooms: How retail media is driving adoption, July 2025.
22. Incremental — The challenge of offsite retail media measurement, August 2025.
23. Amazon Ads — View Attribution Updates for Amazon Store Ads, January 1, 2026.
24. Code3 — Amazon Quietly Tightened Attribution, and It's Changing How DSP Performance Is Measured, March 2026.
25. PPC Land — Amazon tightens view attribution as ROAS reporting splits, January 2026.
26. Intentwise — Explained: How does Amazon attribute ad sales?, April 2023 (foundational click-window specification, confirmed unchanged as of January 2026).
27. Amazon Ads — Accelerate Amazon Marketing Cloud workflows with Ads Agent (beta), November 11, 2025.
28. Amazon Ads — Amazon brings gen AI to Amazon Marketing Cloud, January 2025 (CES 2025 announcement).
29. Amazon Ads Research — Multi-Touch Attribution via Causal Calibration (arXiv:2508.08209), August 11, 2025.
30. Amazon Ads — Uncover ad value across the funnel with Multi-touch attribution (beta), November 2024.
31. eMarketer — Incrementality takes center stage, but there's still work to be done, November 7, 2025.
32. eMarketer — FAQ on incrementality: How to prove your ads actually work in 2026, April 3, 2026.
33. Dataslayer — Incrementality: Top Retail Media KPI for 2025, November 2025.
34. Skai — Retail Media Incrementality Testing: Measuring True Campaign Impact, January 2025.