Retail Media Platforms: How They Work, With Real Numbers

What Is a Retail Media Platform?

A retail media platform is advertising space a retailer sells inside — and increasingly outside — its own store, priced and targeted using its own shopper data. Amazon Ads (Sponsored Products, Sponsored Brands, Sponsored Display, and Amazon DSP), Walmart Connect, Target Roundel, Instacart, Kroger Precision Marketing, and Costco Retail Media Network are the platforms most people mean when they say the phrase. Each runs the same basic trade: the retailer knows what people actually buy, and it sells access to that knowledge through ads.

A second category sits alongside them: connective platforms like Criteo, which don't own a store but plug into many retailers' inventory and data at once, letting one team buy across several retail media networks from a single seat. These are real retail media advertising platforms too — they just broker the shelf instead of owning it.

Amazon's version splits in two, and this is where confusion usually starts. Sponsored ads are auction-priced placements on search results and product pages. Amazon DSP — the Demand-Side Platform, Amazon's programmatic buy for display, video and audio across Amazon-owned and third-party sites — is a separate system with separate reporting. It is not the Delivery Service Partner courier franchise that shares the same three letters. If you typed "DSP" expecting a van route, this page is about the other one.

How the Stack Actually Works: Five Layers

Every retail media platform, whoever owns it, is built from the same five layers stacked on top of each other. Sponsored ads sit closest to the register — someone searches, an ad wins, they buy or they don't, and the retailer can tell you the exact SKU. Display and video sit further back, showing someone a brand while they're still deciding whether to look for it at all. In between sits the plumbing: identity, offsite reach, and measurement that's supposed to connect the two.

The table below breaks out what each layer actually buys, where it shows up, and — the part most guides to retail media platforms skip — what it can and cannot prove on its own.

A Worked Example: What the Numbers Actually Mean

Numbers without context are how retail media gets oversold in both directions. Here's one real block of them, scoped honestly: across 30 advertisers running Amazon DSP in July 2026, the book delivered 78.4 million impressions at a $4.00 CPM, a blended $1.42 cost-per-click, and 57,137 attributed purchases at a blended $5.49 cost-per-acquisition — 20.1% of those purchases from shoppers new to the brand. Reported return on ad spend was 6.04x.

Walk the math forward and the number means something specific: at a $4.00 CPM, $100,000 of impression budget buys roughly 25 million impressions. At the blended click and purchase rates above, that spend should land inside a predictable band of purchases — and if a campaign comes in wildly outside that band, something changed. Not just "performance."

The part every retail media platform's dashboard will quote, and the part almost none of them explain, is that last-click ROAS and incrementality are not the same measurement. Last-click tells you which ad fired closest to the sale. It cannot tell you whether that sale would have happened anyway — through search, direct traffic, or a sponsored ad the same shopper saw five minutes later. Only a holdout, or a clean read inside a reconciliation layer like Amazon Marketing Cloud where DSP and sponsored ads stop double-counting each other, can answer that.

The Mistake Almost Every Advertiser Makes (Including Us, Early On)

The common failure is judging a display or video campaign by the same last-click rules built for search ads. Display's job is usually to start a journey, not finish it — the shopper sees the ad, closes the tab, searches the brand name three days later, and clicks a sponsored ad instead. Last-click hands 100% of the credit to sponsored, marks the display campaign a loser, and someone cuts the budget that was actually driving the search volume in the first place.

We have made this mistake ourselves. Before reconciling DSP and sponsored reporting properly, campaigns we ran looked worse on paper than they were — the display spend showed up as cost with no attached credit, because the credit had already gone to a sponsored click downstream. It's an easy trap: the dashboards make it look like a clean answer when it's actually two systems counting the same shopper twice, once each.

The fix isn't to stop measuring last-click. It's to stop treating it as the only measurement, and pair it with a holdout or matched-control comparison before deciding a channel isn't working.

What to Do When the Number Looks Wrong

Sometimes the ROAS is genuinely bad, and sometimes the number is lying to you. Before pulling budget, check three things.

• Overlap first. Pull a path-to-conversion report and check how many "sponsored" conversions were also exposed to display. If it's a large share, your sponsored ROAS is partly borrowed from display, not independent of it.
• Frequency second. A CPM that climbed sharply mid-campaign usually means you're bidding against your own sponsored campaigns, or a competitor moved in — not that the audience got worse.
• New-to-brand third. A dropping new-to-brand percentage inside a flat ROAS often means the platform is spending more of your budget re-showing ads to people who were already going to buy.

If all three check out and the campaign is still underperforming, that's real — turn it off. Every change should carry a stated reason, a way to measure whether it worked, and a point at which you reverse it if it didn't. That discipline matters more than which platform you're running it on.

Where Dr. DSP Fits

Dr. DSP is Amazon DSP — the Demand-Side Platform — run as a managed product by Fable 5, from Full Circle, which has managed more than $500M in revenue across 100+ brands, with 70+ brands currently active across the Full Circle and reMKTR group. The argument underneath it is the one above: last-click can't prove incrementality, holdouts and AMC reconciliation can, and every change should carry the evidence behind it, a measurement plan, and a rollback point before it runs. There's no published price — a demo, a free first 30 days, and pricing set on the call against your actual media budget — and clients choose their own autonomy level, from full approval to fully autonomous inside agreed guardrails. Whether or not you ever run DSP through us, the reconciliation step is worth doing wherever you run it.