Retail media is the advertising category in which brands buy placements on a retailer’s owned digital surfaces, with the retailer acting as both the media owner and the primary attribution source. Amazon Ads is the dominant example by spend; Walmart Connect, Target Roundel, Instacart Ads, Kroger Precision Marketing, and the major marketplace ad systems sit in the same category. The unifying feature isn’t the inventory type, it’s the operating model: a retailer monetizing its shopper data and its surfaces, and reporting back on what converted.
The category splits into two halves.
Onsite retail media
Placements a shopper sees on the retailer’s own properties: sponsored product listings on search results, banners on category pages, display units in the checkout flow.
Offsite retail media
Activates the retailer’s first-party purchase data on inventory off the retailer’s site, typically programmatic display, connected TV, and social. Walmart Connect’s offsite display network is a common reference point — a brand can target Walmart shoppers across the open web using purchase signals the retailer holds.
A practical distinction worth knowing: endemic brands sell on the retailer; non-endemic brands (a financial-services or auto brand using a retailer’s audience signals without selling on the retailer, say) buy the audience without selling there.
The measurement story is what makes retail media structurally different from open-web paid media. Because the retailer owns the conversion event, click-to-purchase inside the walled garden is deterministic in a way the open-web attribution story can’t match — the retailer logs the click and the order against the same shopper ID. That’s more closure than open-web paid media, not total closure. View-through measurement still exists on retail media DSP buys (Amazon Ads, for instance, attributes view-through conversions within a defined time horizon), and offsite inventory leans on the same modeled measurement techniques as the open web. The honest framing is that retail media has a larger deterministic surface than other paid channels, not that it eliminates probabilistic measurement.
That partial closure is where the operator trap lives. Retailer-reported ROAS often looks attractive because the surface is high-intent: the shopper is already on the product page or search results when the sponsored listing appears. But a meaningful share of that attributed revenue is organic-listing cannibalization rather than incremental — purchases the shopper would have made from the unpaid result on the same page. The retailer’s view is walled: closed within their data, blind to whether the brand also got the sale via search, email, or organic elsewhere. Read retailer ROAS as a measurement of within-retailer click capture, not as channel efficiency. The diagnostic is a holdout or incrementality test inside the retailer’s measurement environment, such as Amazon Marketing Cloud’s lift tools or Scintilla (formerly Walmart Luminate).