Retail media has, by every measure, become a structural pillar of the digital advertising market. US retail media spend is on track to cross $60 billion in 2024, according to eMarketer projections — a category that barely existed in its current form five years ago. Amazon Advertising alone represents roughly half that figure. Walmart Connect, Kroger Precision Marketing, Home Depot’s Orange Apron Media, Instacart Ads, and Target’s Roundel collectively represent another substantial share.

The money is real. The measurement is not.

The Network Explosion and Its Consequences

The 2021-2023 period saw an extraordinary proliferation of retail media network launches. Virtually every major retailer with meaningful transactional data — and many that don’t quite qualify — announced a retail media network, a data platform, or an advertising partnership. Industry estimates now count more than 200 distinct retail media network entrants in the US market.

This proliferation creates a structural problem that benefits no one except the retailers themselves: every network operates with its own attribution model, its own measurement methodology, its own definition of conversions, and its own data access controls. An advertiser running campaigns simultaneously on Amazon DSP, Walmart Connect, and Kroger Precision Marketing is not comparing like to like when they review performance reports from each. They may not even be measuring the same customer behavior.

The Trade Desk’s OpenPath retail media partnership expansion has been positioned as a partial solution — enabling buyers to access retail media inventory through The Trade Desk’s infrastructure and apply consistent measurement standards across retail media buys. But OpenPath access is available for only a subset of retail media networks, and the networks with the most advertiser leverage — Amazon, Walmart — are not distributing their premium onsite inventory through third-party DSPs.

The Attribution Problem in Granular Detail

The attribution methodologies across retail media networks differ in ways that aren’t always visible in the rate cards.

Amazon’s default attribution is last-touch with a 14-day click window. Sponsored Products, Sponsored Brands, and DSP ads each have different attribution defaults that can produce radically different ROAS numbers for the same underlying sales activity. A consumer who saw a DSP display ad, clicked a sponsored product ad, and purchased 12 days later attributes entirely to the sponsored product click under default settings — even if the DSP ad drove the initial product discovery.

Walmart Connect uses a 14-day view-through window in addition to click attribution, which tends to inflate attributed sales relative to networks that apply view-through more conservatively. Kroger Precision Marketing’s attribution looks at purchase data in Kroger’s loyalty database directly — which is powerful for accuracy but creates comparability issues with networks using probabilistic purchase matching.

The IAB’s retail media measurement guidelines, published in late 2023, attempt to create a common framework, but adoption is voluntary and the major networks have not committed to full alignment. The result is that advertisers who rely on network-reported metrics without understanding the attribution assumptions underlying them are comparing fundamentally incompatible numbers.

What “Closed Loop” Measurement Actually Means

The key marketing advantage of retail media — what distinguishes it from other digital advertising — is closed-loop measurement: the ability to link ad exposure directly to purchase at the product level, using the retailer’s own transaction data rather than probabilistic purchase inference.

This is a real capability that doesn’t exist anywhere else in the programmatic ecosystem. When Kroger Precision Marketing reports that a campaign drove $X in incremental sales of a specific SKU among Kroger loyalty cardholders, that number is based on actual purchase records matched to actual ad exposure, using loyalty card IDs as the linking mechanism. That’s a fundamentally more reliable signal than a pixel-based attribution or a household income-matched sales proxy.

But “closed loop” is also a marketing term that different networks define differently. Some networks apply “closed loop” to mean any attribution using internal transaction data, including last-touch attribution with no incrementality testing. Others use it to mean matched market testing or panel-based causal inference. The former produces numbers that overstate media contribution; the latter produces numbers that are methodologically closer to what advertisers actually want.

Advertisers need to ask their retail media partners specifically: “What percentage of reported attributed sales are incrementally driven, versus what would have occurred without the campaign?” Very few networks proactively answer this question. The ones that do are providing more honest measurement than the ones that don’t.

The Onsite vs. Offsite Divide

Retail media spend broadly divides into onsite (ads served on the retailer’s own properties — website, app, search results pages) and offsite (programmatic ads served across the open web using retailer first-party data). These are very different products with different inventory quality, different measurement capabilities, and different competitive dynamics.

Onsite retail media — sponsored products, display banners on retail sites, paid search within retail properties — is where retail media originated and where closed-loop measurement is most reliable. Onsite placement means the ad is served adjacent to purchase intent and the click path to conversion is direct. CPMs are high because advertiser demand for premium placement on Amazon, Walmart, and major grocery apps is intense.

Offsite retail media extends the retailer’s first-party audience data to programmatic impressions served across the web, CTV, and social. The audience signal (shoppers who bought X in the last 90 days) is valuable, but the closed-loop measurement advantage weakens significantly when the purchase path goes through multiple channels. Offsite retail media CPMs are lower, but attribution is substantially more complicated.

The Trade Desk’s retail media expansion is primarily an offsite product — it surfaces retailer audience data for programmatic activation across TTD’s DSP. The inventory quality and measurement infrastructure are different from the onsite placements that drive the bulk of Amazon and Walmart’s retail media revenue.

What Buyers Should Be Doing Differently

The operational gap in most retail media programs is the absence of a consistent measurement framework applied across networks. Buyers who are running retail media campaigns without a normalized performance view are optimizing within silos and making budget allocation decisions based on incompatible data.

The practical solution has three components. First, define a consistent attribution standard before buying: agree with your retail media partners on lookback windows, click/view weighting, and whether incrementality testing is part of the evaluation. Second, demand at least one incrementality study per network per year — matched market tests or holdout tests that establish the baseline causal effect of spend before you scale. Third, integrate retail media reporting into a unified marketing measurement model that accounts for the different attribution methodologies each network applies.

None of this is easy. It requires significant analytical resources that most brands’ internal marketing teams don’t have. It’s also the difference between knowing what your retail media spend is doing and believing what the retail media networks tell you it’s doing.

FAQ

Q: Is retail media cannibalizing traditional trade marketing spend, or is it incremental budget? Evidence is mixed. CPG brands are redirecting trade promotion budgets toward retail media in some cases, because retail media offers better measurement and more controllable ROI signals than traditional trade. But many brands are also growing retail media spend as net-new budget, particularly for brand building within retail environments. The answer depends heavily on the brand’s category and retailer relationship structure.

Q: How does Amazon’s retail media measurement compare to other networks in terms of reliability? Amazon’s measurement infrastructure is the most sophisticated in the category, largely because of the scale of its purchase data and the tight integration between DSP attribution and Amazon’s transaction records. However, even Amazon’s attribution defaults (last-touch, 14-day click) overstate media contribution relative to causal measurement standards. Amazon’s own “Amazon Marketing Cloud” clean room allows more sophisticated incrementality testing, but it requires significant analytical investment.

Q: Should mid-market brands be running retail media campaigns, or is this primarily a large CPG play? Retail media is most efficient for brands with strong existing presence in a retailer’s physical or digital shelves — because the closed-loop measurement works best when there’s sufficient purchase volume to measure. Brands with limited SKU presence or low velocity items may find retail media less efficient than programmatic alternatives. The minimum viable scale for meaningful measurement is generally $50-100K per network per quarter.

Q: What is the IAB’s retail media measurement standard and when will it be meaningfully adopted? The IAB’s retail media measurement guidelines establish definitions for terms, attribution windows, and reporting requirements. Major networks have participated in developing the standards but have not committed to compliance timelines. Meaningful adoption — where an advertiser can compare IAB-standard reports across multiple networks — is likely 18-24 months away, if it happens at all given the competitive incentive networks have to maintain proprietary measurement.