Retail Media Solutions: The Four Layers, Explained

What the phrase actually means

The top results for this exact search prove the confusion: one is a commerce media platform selling reach across the open internet, one is an LG signage integrator managing pump toppers and window displays at c-stores, and one is a European electronics retailer's own onsite-plus-offsite ad program. None of them is using the phrase wrong. They're describing different layers of the same industry.

At its plainest, retail media is any advertising a retailer sells because it owns first-party shopper data and a storefront — physical or digital — where purchase intent is high. A "solution" is whatever tech and service wraps around that inventory to plan, book, and measure it. There isn't one product called a retail media solution. There are at least four, and a unified retail media solution is one that manages more than one of them without letting them double-count each other.

The four layers, and who actually sells each one

Below is the stack in the order most brands buy it, and which of the pages ranking for this search actually covers each piece.

• Onsite sponsored ads — search and product ads inside the retailer's own site or app. This is what most people mean by "retail media" first.
• Offsite programmatic (DSP) — display, video, and audio bought outside the retailer's own site, using its shopper data to target. This is where Amazon DSP and commerce media platforms like Criteo compete.
• In-store retail media — physical screens, pump toppers, digital menu boards. This is a real, separate category, and it's what the LG-based signage provider ranking for this term actually does.
• Measurement and reconciliation — the layer that ties the first three back to one incremental number instead of three overlapping ones. Almost nobody ranking for this term sells this layer on its own.

A worked example: why the same campaign shows two different ROAS numbers

Say a brand runs Amazon Sponsored Products and Amazon DSP at the same time. Sponsored Products reports attributed sales from shoppers who clicked a search ad. DSP separately reports attributed sales from shoppers who saw or clicked a display ad. If a shopper did both — saw the display ad, then searched and clicked the sponsored ad — both channels claim the sale. Add the two ROAS figures together and the brand is double-counting revenue that happened once.

The only honest way to see what DSP actually added is to reconcile the two in Amazon Marketing Cloud, where overlapping shoppers get deduplicated, and to test with holdouts — a matched group of shoppers who see no DSP activity — so lift gets measured against people who never saw the campaign, not against a channel's own click log.

Across 30 advertisers reconciled this way in July 2026, the book delivered a 6.04x return on ad spend, 78.4 million impressions at a $4.00 CPM, and a blended $1.42 cost-per-click, landing at a $5.49 blended cost-per-acquisition across 57,137 attributed purchases — 20.1% of them from shoppers new to the brand. Those figures held up because they were checked against holdouts, not just added on top of whatever sponsored ads already claimed.

This reconciliation step is the layer most vendors skip. Dr. DSP — Amazon DSP, the Demand-Side Platform, not the Delivery Service Partner courier franchise — runs this as the default rather than an add-on, built by Full Circle, which has managed more than $500M in revenue across 100+ brands. Orbit, the full software suite, comes with it at no extra cost.

The common mistake: buying layers instead of buying proof

The mistake we see most — including in our own early campaigns — is treating each layer as a separate purchase decision instead of asking what proof comes with it. A retailer's onsite console will always show clicks and last-click sales; it has no way to tell you what would have happened without the ad. A signage vendor can show engagement lift on a screen; it can't tell you if the shopper was going to buy anyway. Neither is dishonest. It's just the limit of what last-click and on-screen metrics can prove.

The fix isn't a better dashboard. It's a holdout: hold back a matched group of shoppers, run the campaign against everyone else, and compare. If the lift disappears once you strip out shoppers who were already about to buy, the campaign wasn't adding anything, no matter how good the platform's own ROAS number looked.

Every change to a live campaign should carry three things before it runs: the evidence behind the change, a measurement plan for checking it worked, and a rollback trigger for when it doesn't. Skip any of the three and you're back to guessing with better software.

When the answer is bad news

Sometimes the honest read is that the campaign isn't working. A few signs, and what to do about each:

• Reconciled ROAS is much lower than the channel-reported number. Usually overlap, not fraud — rebuild the number in AMC before deciding anything.
• The holdout shows no measurable lift. Don't keep spend running on hope. Pause the segment, redirect budget to the part of the funnel that did show lift, and retest with a fresh holdout before scaling back up.
• New-to-brand share is low. That's a targeting problem, not a creative problem — check whether the audience is mostly people who already buy the brand, which inflates ROAS without growing the customer base.
• CPA looks fine but varies wildly by segment. Break it down before cutting the whole campaign; one bad segment can drag down an otherwise-working one.

Clients set their own autonomy level for this — full human approval, supervised, or fully autonomous inside agreed guardrails — depending on how much risk they want carried on a given SKU. The point of the guardrail is that bad news gets caught by the rollback trigger before it becomes a quarter of wasted spend, not after.