What Is Amazon Retail Media, Really?

What Amazon Retail Media Actually Includes

Amazon retail media covers three things people usually lump together: sponsored ads (Sponsored Products, Sponsored Brands, Sponsored Display, on search and product pages), Amazon DSP (programmatic display and video that runs on Amazon-owned properties like Fire TV and IMDb, plus the open web), and Amazon Marketing Cloud, the clean-room layer that measures both without letting them double-count each other.

There's a fourth piece now, and it's why the term is getting bigger, not smaller: Amazon Retail Ad Service. Amazon has started licensing its own ad tech to other multi-brand retailers in the US — Asda in the UK was the first outside the country — so a brand advertising through 'Amazon' infrastructure might now be running on a completely different retailer's site. That's a genuinely new wrinkle and most explainers on this term haven't caught up to it yet.

If someone says 'we do Amazon retail media' they could mean any one of those four things, or all of them stitched together badly. The rest of this page is about telling those apart, and about what the numbers from each one actually mean once you stop reading them separately.

How the Money Actually Moves: A Worked Example

Numbers move through the funnel in a specific order, and it's easier to trust an explanation once you can watch the arithmetic hold up. Here's one real book: 30 advertisers, reMKTR's managed accounts, July 2026.

The book ran 78.4 million impressions at a $4.00 CPM, which works out to roughly $313,600 in impression spend. Clicks came in at a blended $1.42 CPC — worth flagging, because the $0.41 CPC figure that gets quoted constantly in this category is online-video only, not a whole-book number, and mixing the two makes campaigns look cheaper than they really run. At $1.42 a click, that spend buys around 220,000 clicks.

Blended cost per acquisition was $5.49 across 57,137 attributed purchases — which, worked backward, lands close to the same total spend, meaning the CPM/CPC math and the CPA math describe the same book rather than two different ones. That's a basic sanity check most reporting decks skip. Roughly one in four clicks converted. Across the whole book, blended ROAS was 6.04x — measured across all 30 advertisers, not the best line item pulled out to make a slide look good. And 20.1% of those purchases came from a shopper new to the brand, which matters more than the ROAS figure if the campaign's job was growth rather than harvesting existing demand.

That's the whole chain: impressions, CPM, clicks, CPC, purchases, CPA, ROAS, new-to-brand rate. If you can't walk your own numbers through those steps in order, you don't have a working report — you have a dashboard.

Where the Number Lies to You: Attribution vs. Incrementality

A 6.04x ROAS tells you what Amazon's reporting attributed to the ads. It does not tell you whether those purchases would have happened anyway. Last-click attribution takes credit for the last touchpoint before a sale, whether or not that touchpoint caused it — and when a shopper sees a sponsored ad and a DSP ad in the same week, standard reporting often counts the sale twice, once under each channel.

Amazon Marketing Cloud exists specifically to fix that second problem: it's a clean room where DSP and sponsored ads get reconciled against the same purchase, so a sale only counts once. It does not fix the first problem. The only way to know if display added incremental sales — sales that wouldn't have happened without it — is a holdout: run the campaign to part of the audience, withhold it from a matched group, and compare. That takes longer than pulling a dashboard number and it's less flattering more often than agencies like to admit, but it's the only method that actually answers the question.

This is a mistake we've made ourselves in earlier engagements: reporting a strong blended ROAS to a client without separating what sponsored ads would have delivered alone from what DSP added on top. The number wasn't wrong. It just wasn't answering the question the client thought it was answering.

The Mistakes That Show Up Most Often

• Treating ROAS as proof of incrementality. It's a ratio of attributed revenue to spend. It says nothing about what would have happened without the spend.
• Quoting a video-only CPC as the category average. The $0.41 figure circulating in reports is online video specifically — a blended book runs closer to $1.42. Using the lower number to set budget expectations sets you up to look like you're overspending when you're not.
• Scaling DSP before testing whether it does anything. Brands see a sponsored-ads ROAS they like and add display spend expecting the same multiple, without checking whether display sales are additive or just restating sponsored-ads demand under a different line item.
• Assuming there's a rate card. Amazon retail media — sponsored ads and DSP both — runs on auction pricing set by competition for the placement, not a published price list. If a vendor quotes a flat CPM or CPC as 'the going rate,' ask what date and what category that came from.

What to Do When the Number Is Bad News

Sometimes the honest answer is that the campaign didn't work, or the fix you tried didn't move it. Three situations come up often enough to name directly.

ROAS looks fine but new-to-brand share is low. That usually means the campaign is well-targeted at people who were buying anyway. It's not a wasted budget, but it's not growth either — and if growth was the brief, the fix is audience and placement, not more spend on the same targeting.

CPCs are rising and conversion isn't. Before cutting budget, check whether the increase is coming from more competitive auctions in your category or from your own campaigns cannibalizing each other's keywords — a common self-inflicted problem once a brand runs several overlapping campaign structures.

A holdout test comes back flat. This is the outcome nobody wants to report up, and it's exactly the finding a holdout is supposed to be able to produce. A flat result on a specific placement or audience segment is information — it tells you to reallocate, not to abandon measurement in favor of a friendlier number next quarter.

Amazon's Ad Tech Is Now Showing Up on Other Retailers' Sites

Amazon Retail Ad Service is the newest layer, and it changes what 'Amazon retail media' will mean going forward. Amazon is licensing the same ad tech that runs Sponsored Products on Amazon.com to other multi-brand retailers — Asda was the first outside the US, and Amazon has named Oriental Trading Company, iHerb, Weee!, and Tilly's among early US partners. Brands don't need to sell on Amazon to advertise through it.

The practical effect for a brand like a kitchenware or grooming label — think HexClad, Beardbrand — is that 'buying Amazon retail media' and 'buying media on a retailer's site' are starting to overlap even when the retailer isn't Amazon. Budgets, reporting, and creative specs may increasingly run through the same console regardless of where the ad actually appears. Whether that consolidation helps or just adds another line to reconcile is still an open question industry-wide, and worth revisiting each time a retailer you sell through announces it's adopted the service.