GUEST COMMENT Proving that retail media sales are truly incremental

Juan Baron, head of international and global accounts at Decentriq, explores one of the key challenges in retail media – how to prove incrementality.

Retail media has grown from an experiment into a budget line that sits beside search and social in many marketing plans. Scrutiny has followed, and brand leaders and finance teams now ask a fair question of every report they receive: how many of these sales would have happened anyway?

Why attributed sales overstate impact

Retail media has a built-in measurement advantage. For onsite ads, the retailer sees both the ad exposure and the purchase, so closed-loop reporting is easy to produce and return on ad spend often looks excellent. Offsite and in-store sales rely on identity matching where it exists and on probabilistic modeling where it doesn’t, but the same attribution logic applies.

The weakness lies in what those figures count. Retail media reaches shoppers close to the point of purchase, when intent is already high. A shopper who searches for a product and then buys it from a sponsored listing is credited to the ad, even though many such purchases were going to happen anyway. Loyal customers and promotional periods inflate the picture further. Attribution records which purchases followed an exposure. Incrementality asks what changed because the ad ran.

Measuring what would have happened anyway

Incrementality rests on a counterfactual: an estimate of what sales would have been had the campaign not run, compared with what actually happened.

The cleanest approach is a randomised holdout. A share of the target audience is deliberately withheld from the campaign, and purchases in the exposed and withheld groups are compared. Where randomisation is impractical, geo experiments and matched-market tests compare regions with different levels of spend, and methods such as synthetic controls can approximate the missing baseline.

Each approach has traps. Holdout groups need enough scale to detect realistic effects, and test and control groups must be comparable before the campaign starts. Timing matters too, because a promotion that pulls purchases forward can look strong in week one and flat by week six.

Brands and retailers also define incremental differently. A brand may count a sale when a shopper chooses its product after seeing an ad. The retailer may see the same sale as a switch from another brand on the same shelf, adding nothing to its revenue. Both views are legitimate, so the definition needs to be agreed before launch rather than argued over afterward.

The data problem behind the method

Good measurement needs information that neither side holds alone. The retailer has exposure records and transactions. The brand has sales in other channels and at other retailers, pricing and promotion calendars, and customer data of its own. Joining these views allows better control groups and captures purchases made outside the retailer’s own channels.

Each side has good reasons to keep its data close. Shopper-level behavior is protected by data protection law and commercially valuable, and a brand’s margins and channel performance are not something it wants a retail partner to see in detail. Many partnerships settle for aggregated reports prepared by the retailer selling the media, which asks the seller to mark its own homework.

How privacy-preserving collaboration helps

Privacy-preserving techniques now let organisations analyse combined data without exposing the underlying records. Data clean rooms are the best-known method, built on or alongside techniques such as secure multi-party computation and federated analysis. The shared principle is that each party keeps control of its own data, analysis runs under rules both sides have approved, and only aggregated results come out.

In practice, a brand can match its own sales records against retailer exposure data to build test and control groups, then measure lift across channels. The retailer never sees the brand’s pricing, and the brand never sees individual shoppers. The same setup can support longer measurement windows and comparisons across several retail partners, which no single retailer’s report can offer.

Technology covers only half the job. A clean room cannot repair a weak experiment. The harder work is governance: agreeing on the question being asked, the queries that are permitted, the minimum group sizes below which results are suppressed, and which parties can see the output.

Four habits worth adopting:

1. Agree the definition of incremental before launch, including whether the benchmark is brand-level or retailer-level sales.
2. Build the test into the campaign plan, with a holdout designed up front rather than reconstructed afterward.
3. Report incremental results alongside attributed ones, with confidence intervals, so that the gap between them starts a conversation instead of a suspicion.
4. Write down the rules of the collaboration, including which analyses are permitted and what minimum output thresholds apply.

Proof will fund the next phase of growth

Retail media’s growth rests on a real strength: retailers know their shoppers and can connect advertising to sales better than most media owners can. The next phase will be funded by proof. Retailers that show brands which sales their media actually caused, using a method the brand helped design, will earn larger and more stable budgets. Those that rely on attributed figures alone will face harder questions each quarter.

The tools to do this while protecting commercially sensitive information already exist. What remains is the willingness to agree on a shared definition of success and accept results that may land below attributed figures.

Stay informed

Our editor carefully curates two newsletters a week filled with up-to-date news, analysis and research. Click here to subscribe to the FREE newsletter sent straight to your inbox. Why not follow us on LinkedIn to receive the latest updates on our research and analysis?