The Incrementality Capability Gap: Why Knowing Isn't Doing
Retail media has a strange consensus problem. Everyone agrees on the answer and almost no one can act on it.
The evidence keeps stacking up. In Skai and Stratably's 2026 State of Retail Media research, incrementality was named the single biggest measurement challenge by 75% of advertisers, while just 15% of marketers described their organization as very or extremely effective at measuring retail media performance at all. Per eMarketer and TransUnion, over half of US brand and agency marketers already run incrementality testing in some form, and more than a third plan to invest further this year.
Here's the finding buried in that research that deserves more attention: when marketers name their barriers, methodology complexity and data access barely register. The real blockers are analytics capability, cross-functional ownership, and the absence of a shared measurement language. In other words, the industry doesn't have a knowledge gap. It has a capability gap. Teams know what incrementality is, believe in it, and still can't operationalize it.
Why the gap persists
Unfortunately, what actually happens when a brand commits to "doing incrementality" often results in one of these means of failure:
The one-off test. The team runs a holdout on one big campaign. It takes six weeks, produces a solid number, and everyone learns something. But then pricing changes, a competitor launches, the media mix shifts, and the number is no longer helpful.
The unstaffed dashboard. Someone buys or builds an incrementality report and it's directionally interesting. But media buyers keep optimizing to the ROAS they see in-platform because that's what refreshes daily and that's what their tools accept as an input. The dashboard gets opened before QBRs.
The ownership vacuum. We see it time and time again: measurement sits with analytics, activation sits with media, budget sits with brand or finance, and incrementality is everyone's priority and no one's responsibility. When the incrementality number disagrees with the retailer's attributed number (and it will), there's no one with the standing to say which one number the organization believes. The more flattering number wins.
To be clear - none of these are measurement failures. The math worked every time. What failed is the connection between the math and the decision about where budget goes.
Closing the gap is an infrastructure problem
A handful of the largest advertisers can genuinely build a data science team, stand up the pipelines, and run the models in-house. For everyone else the work is brutal, because it means ingesting sales, media, and merchandising data across every retailer continuously, continuously retraining models so the answers stay current, and pushing outputs into the systems where buying decisions actually happen. That's a permanent engineering function, not a project.
The alternative is treating causal measurement as a data signal that gets plugged into the stack. What that has to mean in practice, concretely:
Always-on rather than episodic, so the numbers are as fresh as the ROAS numbers they're competing with for attention.
Comparable across retailers and channels, because a per-platform incrementality number recreates the same silo problem that made ROAS untrustworthy in the first place.
Connected to activation, delivering recommendations where buyers already work instead of adding a dashboard to the pile.
Brands running this way typically find 15-20% more sales in the same budget, which is what happens when reallocation finally follows cause instead of credit. Seven of the ten largest CPG brands in the world now run their commerce media on this model.
The uncomfortable takeaway
If your organization believes in incrementality and still runs its budget on ROAS, the problem isn't conviction and another webinar won't fix it. The gap between knowing and doing is closed by infrastructure that makes the causal number the easiest number to use.
That's a testable claim. Request a Commerce Media Optimization Assessment and we'll show you, on your own data, where credit and cause diverge.
