Retail Media Network Examples: Onsite, Offsite, and In-Store

The examples, grouped by what they actually sell

The retailers usually cited as retail media network examples are Amazon Ads, Walmart Connect, Target Roundel, Instacart Ads, Kroger Precision Marketing, Costco, Home Depot Retail Media+, CVS Media Exchange, eBay Ads, DoorDash Ads, Uber Advertising, Macy's Media Network, Sephora, Albertsons Media Collective, and Walgreens Advertising Group. What separates them isn't the brand name on the network — it's whether the ad runs onsite, offsite, or inside a physical store.

Onsite networks sell space inside the retailer's own search results and product pages. This is what most people mean by the term. Amazon's Sponsored Products, Walmart Connect's onsite placements, Target Roundel's onsite ads, and Kroger Precision Marketing's search placements all fall here — the ad sits next to or inside the thing the shopper is already looking at.

Offsite / programmatic networks take the retailer's first-party shopper data and use it to buy display, video, or audio somewhere else entirely. Amazon DSP is the largest example: it uses Amazon's shopper signals to buy inventory across Amazon-owned properties like Fire TV, IMDb, and Twitch, plus the open web. Walmart and Kroger offer a version of this too, usually routed through a partner such as Criteo or 84.51°.

In-store digital media is ad space sold against physical retail — checkout screens, shelf displays, in-store audio. Walmart and Kroger both sell this as part of their network. It's a smaller slice of spend, but it closes the loop between an ad and a purchase that never touched a screen.

Marketplace and vertical networks are smaller and narrower, but the shopper is closer to the register: Instacart Ads and DoorDash Ads sell against people actively shopping for groceries or food right now. Costco, Home Depot Retail Media+, CVS Media Exchange, and Sephora sell against loyalty bases with obvious category relevance — bulk buyers, DIYers, pharmacy customers, beauty shoppers.

A worked example: what one of these networks actually produces

Definitions are easy. Knowing what a real campaign through one of these networks looks like once it's running is harder. Here's an example scoped to a slice of our own book, not a projection: across 30 advertisers running Amazon DSP in July 2026, the book delivered 6.04x return on ad spend, 78.4 million impressions at a $4.00 CPM and a blended $1.42 cost-per-click. Blended cost per acquisition was $5.49 across 57,137 attributed purchases, 20.1% of them from shoppers new to the brand.

Those numbers only mean something if you know what they're measuring. CPM and CPC describe what the media cost. ROAS and CPA, on the platform's own dashboard, describe what got attributed to it — and attribution isn't proof the ad caused the sale. A shopper who was already going to buy that cookware set gets served a display ad on the way to checkout, and the platform credits the ad anyway. That gap exists in every retail media network's own reporting, regardless of how large the network is.

The mistake almost everyone makes with these examples

The common mistake is treating every network's reported ROAS as comparable, and as proof of causation. Amazon doesn't even have one window to compare — Sponsored Products is click-attributed and Amazon publishes its conversion columns at several lengths (1, 7, 14 and 30 days from the click), Sponsored Brands and Sponsored Display report on a 14-day click window and also credit views, and DSP is configured per campaign — so a 4x quoted against one network's window and a 4x quoted against another aren't measuring the same thing, so a 4x on one network and a 4x on another aren't the same 4x. Stacking them into one blended number and reporting it upward is a fast way to make a bad budget decision look like a good one.

We've made a version of this mistake ourselves: crediting a DSP display campaign for sales that sponsored ads had already caught, because both were reporting against the same last-click window. The fix isn't a better dashboard — it's a holdout, a matched group of shoppers who don't see the ad, reconciled in a system that can see both campaigns at once. For Amazon specifically, that system is Amazon Marketing Cloud, the only place DSP and sponsored ads stop double-counting each other.

What to do when the reported number doesn't hold up

If a network's dashboard says a campaign is delivering strong ROAS and a holdout test says otherwise, believe the holdout. Last-click attribution can't distinguish a shopper who was influenced by the ad from one who was going to buy anyway — it was never built to make that distinction. When the two disagree, the fix isn't a bigger budget on the channel the dashboard likes. It's pausing the specific tactic, re-running the matched-control test with a longer window, and checking whether the lift shows up anywhere else in the funnel.

This is the actual job description of running a DSP as a managed practice: every change carries three things before it runs — the evidence behind it, a measurement plan, and a rollback trigger. That discipline matters more on the offsite/programmatic side of a retail media network than on the onsite side, because offsite is where attribution gets fuzziest and budgets get large fastest.

Dr. DSP — the abbreviation here means Amazon DSP, the programmatic demand-side platform for display, video and audio, not the Delivery Service Partner courier program — is one way to run that discipline as a managed product. Full Circle, the company behind it, has managed more than $500M in revenue across 100+ brands.