Retail media advertising is growing at a rate that has systematically outpaced the development of the measurement infrastructure that would allow advertisers to evaluate it rigorously. Amazon’s advertising business generated approximately $31.2 billion in 2021, making it the third-largest advertising platform in the United States. Walmart Connect, Kroger Precision Marketing, Target Roundel, Instacart Ads, Home Depot Orange Apron, and dozens of smaller retail networks are all building programmatic advertising infrastructure on top of their first-party purchase data, pitching ROAS metrics that cannot be compared across platforms in any meaningful way.

The IAB Tech Lab published retail media measurement guidelines for public comment in late 2022, and the publication is an acknowledgment of a problem the industry has been papering over with enthusiasm for a category that genuinely delivers results — just results that are measured differently by everyone selling them. Before retail media doubles in size again by 2024, the industry needs to have a real conversation about what standardized measurement looks like and what it would actually require.

Why Every Retail Network Reports ROAS Differently

The measurement chaos in retail media is not a failure of capability. Amazon has some of the most sophisticated attribution infrastructure in digital advertising. The problem is incentive alignment: every retail media network has strong incentives to report ROAS in the way that makes its performance look best, and weak incentives to adopt standards that would enable side-by-side comparison with competitors.

The fundamental measurement divergences across retail networks involve attribution windows, on-site versus off-site definition, same-SKU versus total basket attribution, and new-versus-returning customer segmentation.

Attribution windows vary from 7 days to 30 days depending on the network. Amazon’s default sponsored products attribution is 7 days for Sponsored Products and 14 days for Sponsored Brands. A competitor network using a 30-day attribution window will appear to generate higher ROAS for the same campaign simply by capturing more conversions in a longer lookback window, not because it is actually more effective.

On-site versus off-site attribution is particularly significant as retail networks expand into off-network programmatic — the Sponsored Display and Amazon DSP products that run on third-party publisher inventory. Including off-site conversions in network-level ROAS reporting inflates the apparent contribution of sponsored search placements on the retail site itself.

Same-SKU attribution — only crediting ads for purchases of the specific advertised product — versus total basket attribution — crediting ads for any purchase in the session — creates dramatically different ROAS outcomes. A shopper who clicks a sponsored cereal ad and then adds pasta, juice, and detergent to their cart has a total basket value that is multiples of the cereal SKU value. Attribution methodology determines whether the ad gets credit for $4.99 or $47.

Amazon’s $31B Business Is the Measurement Baseline That Everyone Will Be Judged Against

Amazon’s advertising business is the dominant reference point for retail media measurement because it is the largest and most mature. Amazon’s annual advertising revenue has grown from roughly $10 billion in 2018 to $31.2 billion in 2021, driven primarily by sponsored search placements that appear in organic-looking positions in Amazon’s product search results.

The measurement advantage Amazon holds is structural: Amazon observes both the ad click and the purchase in the same transaction system, without any of the attribution uncertainty that plagues off-platform advertising. When a user clicks a Sponsored Products placement and purchases on Amazon, the attribution is deterministic — there is no conversion modeling, no match-back methodology, no last-touch assumption. Amazon knows exactly what was purchased, when, and by whom.

This deterministic measurement advantage is what retail media as a category is fundamentally selling: the claim that you can know whether your advertising drove a purchase, rather than inferring it probabilistically. The claim is legitimate for on-site retail media. It becomes murkier as networks extend into off-site programmatic buying, where the same pixel-based attribution gaps that affect other digital advertising apply.

Smaller retail networks that are pitching on the basis of their first-party purchase data are implicitly benchmarking against Amazon’s measurement quality. When Kroger Precision Marketing reports ROAS, buyers are comparing it against their Amazon numbers. The comparison is often misleading because the methodology is different, but the conversation happens anyway because there is no common framework to make the comparison explicit.

What the IAB Tech Lab Guidelines Are Trying to Fix

The IAB Tech Lab’s draft retail media measurement guidelines focus on establishing common definitions for the key metrics: what counts as an impression, what constitutes a click, how attribution windows should be disclosed, and how on-site versus off-site media should be distinguished in reporting.

The guidelines do not attempt to mandate a single attribution model — that is probably not achievable across networks with as much competitive sensitivity as Amazon and Walmart. What they propose is disclosure standardization: a common reporting template that requires networks to report the attribution methodology applied, the attribution window used, and the scope of media included (on-site only, or including off-site programmatic) alongside any ROAS or sales lift figures.

This disclosure-based approach is more modest than what advertisers ideally want, which is a single comparable ROAS figure across all retail media networks. But disclosure of methodology is a meaningful step toward comparability, because it enables buyers to normalize reported ROAS figures for window length and attribution scope, even if the normalization is manual rather than automated.

The IAB Tech Lab’s current work on retail media standardization is open for comment and represents an industry attempt at self-regulation ahead of potential external pressure. The parallel here to brand safety and viewability is instructive: both became standardized partly through industry initiative and partly through advertiser buying pressure. Large holding group trading desks that collectively represent hundreds of billions in media spend have more leverage to demand measurement consistency from retail networks than any regulatory body currently contemplating retail media standards.

What Advertisers Should Be Doing While Standards Develop

In the absence of enforced standards, the practical approach is methodological hygiene within your own measurement framework.

Build a normalized comparison model for your retail media investments. Document the specific attribution methodology for each network you are buying on — window length, SKU scope, on/off site scope — and apply normalizing adjustments before comparing ROAS figures. A 14-day same-SKU ROAS is not the same as a 30-day total basket ROAS. The gap between them, for many CPG advertisers, is the difference between profitable and marginal investment.

Invest in independent measurement where possible. Amazon Marketing Cloud, Amazon’s clean room product, allows custom attribution analysis using Amazon’s first-party event data — you can build the attribution model you want rather than accepting Amazon’s default. Not all networks offer equivalent clean room access, but where they do, custom attribution is worth the analytical investment.

Ask each retail network for standardized disclosure before committing spend. What attribution window is the default? What events are included in reported conversions? Is off-site programmatic reporting included in network-level ROAS? These should be table-stakes questions in media planning conversations, not afterthoughts in post-campaign reconciliation.

FAQ

What is retail media advertising and why is it growing so fast? Retail media advertising encompasses paid placements on retailer-owned digital properties — sponsored search on Amazon, display placements on Walmart.com, targeted email from Kroger — powered by the retailer’s first-party purchase data. Growth is driven by the measurement advantage: purchase data enables closed-loop attribution that links ad exposure to actual purchase, which performance advertisers find more credible than browser-based attribution.

Why can’t advertisers simply compare ROAS across retail media networks? Attribution methodology differences make direct ROAS comparison invalid without normalization. Key variables include attribution window length, whether all basket purchases or only the advertised SKU are credited, and whether off-site programmatic placements are included in reported performance. The same campaign on the same product could produce very different reported ROAS numbers on two networks using different methodologies, even with identical actual sales results.

What is Amazon Marketing Cloud and how does it help with measurement? Amazon Marketing Cloud (AMC) is Amazon’s data clean room, which allows advertisers to run custom attribution and audience analysis queries using Amazon’s first-party event data. AMC lets advertisers define their own attribution logic rather than accepting Amazon’s default, enabling custom window lengths, multi-touch modeling, and cross-campaign path analysis.

How close is the industry to having enforced retail media measurement standards? The IAB Tech Lab guidelines are currently at the public comment stage and represent framework proposals rather than enforced standards. Formal measurement standards for retail media are likely several years away. In the meantime, large agency holding groups using their collective buying leverage to demand consistent methodology disclosure from networks is the most effective near-term path to comparability.