Retail media: the seller of the space measures its own effect.

SHORT ANSWER

Decision questionHow do you measure the real effect of retail media spend and set its cap?

Retail media covers the advertising space a retailer sells on its own surfaces: site, app, store, and sometimes audiences resold elsewhere. Its distinctive feature is not the format, it is the measurement: the retailer sells the space, observes the purchases and publishes the result itself, matching exposure to transaction inside its own records. That matching produces attributed sales, not incremental sales: some of those purchases would have happened without the advertising, because the shopper was already in the aisle or on the product page. The spend also covers three different things — media that creates demand, a commercial action that steers it, and access to visibility that behaves like a distribution cost. The decision therefore means separating those three, then measuring the effect outside the tool that bills it. Treat the sales attributed by the network as a billing signal, measure incrementality through a holdout on a scope written in advance, and cap the buy at the share whose effect is demonstrated.

SUPPORTED DECISIONTreat the sales attributed by the network as a billing signal, measure incrementality through a holdout on a scope written in advance, and cap the buy at the share whose effect is demonstrated.

Key distinctions

Three conditions before calculating.

An attribution, not an incrementality

The network matches an exposure and a purchase inside its own window, often fourteen or thirty days, click and view combined. That matching is a billing convention, not an estimate of causal effect: it also attributes purchases that would have happened without advertising.

Three kinds of spend under one name

A single budget line funds demand creation, the steering of choice at the shelf, and access to visibility. The first compares with media, the second with a promotion, the third with a distribution cost. Mixing them makes any return incomparable.

The net price falls by the same amount

Retail media fees are deducted from the price collected, like a commission or a promotional contribution. A margin calculated before those fees overstates the profitability of the channel and distorts the comparison with other routes to market.

Method

Measuring retail media outside the tool that bills it

01Write down the scope and the window
Set what counts as an attributed sale, the window retained, the place of view and click, and the surfaces included, from site and app to store and external audiences. Without that written convention, two reports from the same retailer cannot be compared.
02Classify the spend before judging it
Split the budget between demand creation, commercial action and access to visibility. Each share compares with its own benchmark, whether media return, promotional return or cost of shelf access.
03Measure through a holdout
Suspend the spend across a set of comparable stores or regions, keep the rest live, and read the sales gap over a period defined in advance. Where the retailer's data is not available, use a geo-experiment on the brand's total sales.
04Set the cap and the review date
Derive the incremental share from the test, turn it into a buying cap, and record the date of the next measurement. Treat the share that is not demonstrated as a cost of access, to be negotiated, not as profitable media.

ORIGINAL ASSET

From attributed sale to contribution actually earned

The same campaign, read by the retailer's network and then by a holdout test. The gap between the two readings is the share nobody bills but the advertiser pays.

Step	Amount	Reading
Sales attributed by the network	€3.2M	14-day window, click and view combined
Sales that would have happened without advertising	−€2.1M	Gap read on the control stores
Incremental sales	€1.1M	34% of attributed sales
Contribution margin (38%)	€418k	After retail media fees
Retail media spend	−€600k	Billed on the attributed basis
Net contribution	−€182k	Cap the buy at the demonstrated share

Break-even: €1.58M of incremental sales, 49% of attributed sales.

The reported ROAS of 5.3 says nothing about that threshold: it also counts the sales the holdout showed were not caused.

ILLUSTRATIVE EXAMPLE · SIMULATED DATA

A reported ROAS of 5.3 that leaves €182k of loss

01 · SITUATION

Illustrative example: a brand invests €600k on a retailer's network. The network's report attributes €3.2m of sales to it, a reported ROAS of 5.3. An eight-week holdout across a third of the stores, compared with control stores, serves as the independent measurement.

02 · CALCULATION

The test measures €1.1m of incremental sales, 34% of the attributed sales. At a 38% contribution margin, they return €418k for €600k spent: net contribution is −€182k. Breaking even would require €1.58m of incremental sales, 49% of the attributed sales.

03 · DECISION

The spend is not profitable as media at its current level. It is capped at the demonstrated share, and the remainder is renegotiated as a cost of access to visibility, whose counterpart is discussed in assortment or placement rather than in ROAS.

Acceptance conditions

What must be true to act.

01
Sales attributed by the network are not a basis for decision without independent measurement.
02
The spend is classified as media, commercial action or access to visibility before any comparison of returns.
03
Incrementality is measured through a holdout, on a scope and a period written before the campaign.
04
The share whose effect is not demonstrated is treated as a cost of access and negotiated as such.

Limits

What this analysis does not prove.

A holdout test assumes comparable stores or regions, which the retailer's footprint does not always provide.
Not every retailer gives access to the data an independent measurement requires.
An awareness effect that materialises outside the store escapes a holdout measured on the retailer's sales.
Measurement conventions change with industry standards and network versions, which limits comparison over time.

References