Commerce Media and ROAS: What Retail Media Changes

Commerce media is advertising bought on a retailer's own property, where the retailer sells you access to shoppers already browsing its site. Amazon Ads, Walmart Connect and Instacart Ads are the familiar examples. It changes your ROAS in one structural way that outweighs every tactic: the party selling you the placement is also the party measuring whether it worked, inside a loop you cannot audit. Last updated: September 2026.

Omniconvert has measured how storefronts acquire and convert customers across the CROBenchmark dataset of 7,000+ websites in 15+ industries, against 248+ audit criteria, over 13 years in eCommerce. The channel-level tactics for lifting return on ad spend are covered in how to increase ROAS. This piece is about what happens to the measurement itself once a retail network is in the mix, which is a different problem and the one that catches teams out.

What commerce media is

The formats look familiar. Sponsored results in a category listing, promoted placements on a product page, display units in an app. Nothing about the creative is unusual and nothing about the buying interface is exotic.

What is unusual is the ownership. On a search or social platform you buy attention and then send it to a destination you control, where your analytics observe what happens. On a retail network you buy attention on the retailer's shelf, the shopper adds to a basket the retailer owns, and checkout completes without ever touching your infrastructure.

That is worth stating plainly because it is the source of everything else in this article. You are not running ads to your store. You are running ads inside somebody else's store, on their terms, measured by their instruments.

Why retailers built it

Understanding the retailer's incentive is not cynicism, it is the fastest route to reading their reporting correctly.

A retailer operating on ordinary trade margins found a product with a far better one attached to inventory it already had. Shelf attention is close to free to sell again, which means every additional advertiser competing for the same placement improves the retailer's economics without costing it anything.

The consequence for a brand is easy to miss because it arrives gradually. Placement that used to be earned by selling well is increasingly bought, so the cost of appearing where you already appeared rises over time. Budget spent defending existing visibility is not growth spending, and it is worth accounting for it separately from the spending that is meant to find new customers.

The closed loop, and what it does to ROAS

Closed-loop attribution is genuinely the strongest thing about the channel, and it is also where the difficulty lives.

It is strong because the retailer really does see the whole path. The ad, the click, the basket and the purchase happen on one property, with none of the identity loss that breaks measurement everywhere else. Compared with inferring whether a social impression caused a sale three days later, this is a far tighter observation.

The difficulty is that every judgement inside that loop belongs to the retailer. How long the attribution window runs, whether a view without a click earns credit, whether a shopper who searched your brand name and then bought counts as influenced by your ad: each is a methodology choice made by the company selling you the placement. None of it is available for you to check, and all of it can change.

This is a structural conflict rather than an accusation of bad faith. Any party that both sells and grades has an interest in the grade, and the correct response is not suspicion but treating the number as what it is: a vendor-reported metric, useful for pacing, insufficient for a budget decision.

The three ROAS numbers you now have

The table sets the three against each other.

The middle row is where the damage happens. Blended ROAS is a useful discipline precisely because it refuses to trust any single platform's self-report, and it works by dividing total revenue by total spend using numbers you control on both sides. Drop a closed-loop channel into it and one input is now somebody else's opinion, so the resulting average is neither blended nor comparable to last quarter's.

The distinction between a return figure and what an order actually earns is the same one set out in ROAS versus true profit versus contribution margin, and commerce media makes it sharper, because the retailer's commission and fulfilment fees sit between the reported return and the money you keep.

Where commerce media actually works

The channel has a real and narrow advantage, and it is worth buying for the right reason.

The advantage is proximity. A shopper browsing a category on a retailer's site has already decided to buy something in that category, and the distance between the ad and the purchase is a few seconds rather than a few days. Very little else in the media landscape gets that close.

The limit is that proximity captures intent rather than producing it. Nothing about a sponsored placement makes somebody want a category they did not already want, which means commerce media scales with existing demand and stops when that demand runs out. Teams that shift demand-creation budget into it typically see efficient numbers for a quarter followed by a ceiling nobody predicted.

The arithmetic test is unglamorous and decisive. Take the placement cost, add the retailer's commission, fulfilment and returns, subtract the lot from the revenue, and see whether the remainder justifies the effort. A strong reported ROAS on a product whose margin was already thin after commission can be a loss with a good scorecard attached.

Running it without corrupting your reporting

Four rules cover most of it.

• Give it a separate line, permanently. Not a sub-line under paid media. A channel whose attribution you do not control is a different kind of thing from one you measure yourself, and merging them costs you the ability to reason about either.
• Compare it against itself. Month over month on the same platform under the same settings is a fair comparison. Against your own paid social is not, and no amount of adjustment makes it one.
• Record the settings with the numbers. Attribution window, view-through treatment, brand-term handling. When the figure moves, the first question is whether the measurement moved, and you can only answer that if you wrote down what it was.
• Decide on contribution, not on the scorecard. Placement cost plus commission plus fulfilment plus returns plus cost of goods, subtracted from revenue. Slower, entirely yours, and the only version that survives a methodology change.

Nexus by Omniconvert is the AI eCommerce growth engine: it unifies customer data, segments buyers by behavior and value, predicts churn, and ranks the next-best action, so the contribution left after every fee can be read per segment rather than as one blended average. A channel that looks marginal overall is often strongly positive on one segment and negative on another, and the average hides both.

What you do not get

This is the limitation that matters most in the long run and gets discussed least, because it does not show up in any channel report.

When a customer buys from your own store you get a relationship: an identity, permission to contact them, a repeat history and eventually a lifetime value you can measure. When they buy your product through a retail network, most of that stays with the retailer. You get the margin on one transaction.

The consequence is that commerce media cannot be judged by the standards used for acquisition elsewhere. A channel with a modest first-order return that produces customers who come back is frequently worth more than one with a strong reported return that produces anonymous single purchases, and the reasoning behind that comparison is in what customer lifetime value measures.

None of which makes the channel a mistake. It makes it a specific instrument: efficient at converting existing intent into transactions, on somebody else's property, in exchange for the relationship. Bought knowingly, that is a reasonable trade. Bought as a substitute for building your own demand and your own customer base, it is a slow transfer of the most valuable asset you have.

FAQ: commerce media and ROAS

What is commerce media?

Commerce media is advertising bought on a retailer's own property, where the retailer sells you access to shoppers already on its site or app. Amazon Ads, Walmart Connect and Instacart Ads are the familiar examples. It is also called retail media, and the defining feature is that the placement and the measurement both belong to the retailer.

How is commerce media different from other paid channels?

The audience is already shopping, and the transaction happens on the seller's own property. That gives it very high intent and it also means the retailer owns the click, the basket, the checkout and the reporting. On other channels you at least control the destination and the analytics; here you control neither.

Why do retailers sell advertising?

Because retail margin is thin and advertising margin is not. Selling shelf attention costs the retailer almost nothing per additional unit, so the same shopper visit can be monetised twice: once on the sale, once on the placement that influenced it. That is a genuinely good business for them, and it is a cost line for you.

Why can I not compare commerce media ROAS to my other channels?

Because it is measured under different rules by a party with an interest in the result. Attribution windows, what counts as an influenced sale and how much view-through credit is granted are all set by the retailer, and none of it is auditable by you. The number is real inside its own definition and it does not share a definition with anything else you measure.

Should commerce media go into blended ROAS?

No. A blended average of measurements made under different rules produces a figure that cannot be reasoned about and that will drift as the retailer changes its methodology. Keep commerce media on its own line, with its own reported figure and its own contribution calculation, and compare it against itself over time.

Does commerce media help customer lifetime value?

Usually not directly, because on most retail networks you do not receive the customer. You get a sale, a fee and, at best, aggregate reporting, while the identity, the email address and the repeat behaviour stay with the retailer. That makes it a channel for buying transactions rather than for building relationships.

When is commerce media the right buy?

When you are already selling on that retailer, the shopper intent is high and the margin after all fees survives the placement cost. It works closest to the transaction, which is where its advantage is real. It is a poor substitute for demand creation, because it captures intent that already exists rather than producing any.

What should I measure instead of platform ROAS?

Contribution after every fee: the placement cost, the retailer's commission, fulfilment, returns and the cost of goods. That number is yours to calculate and nobody else controls it. Platform ROAS is worth tracking as a directional signal for pacing, and it is not the number a budget decision should rest on.