Few metrics are as comforting as a high ROAS.
You invest $100,000, and the platform reports $180,000 in sales. The campaign seems efficient, and the most obvious decision is to increase the budget.
But there is one question that ROAS, on its own, cannot answer: how many of those sales happened because of the campaign, and how many would have happened anyway?
This gap is putting incrementality at the center of the measurement discussion. NIQ points to incrementality and cross-channel measurement as two of the main pain points for brands in retail media today.
Start by separating attribution from causality
Attribution attempts to distribute credit. Incrementality attempts to discover what changed because the campaign existed.
These are different questions.
Imagine a repeat customer who already buys a certain brand every month. They search for the product, click on an ad, and make another purchase. The platform might attribute that revenue to the ad.
But would that sale really have disappeared without the ad?
This is why ROAS is useful for tracking efficiency within a platform, but it should not automatically be interpreted as proof of causality. NIQ itself warns that attribution explains how a purchase occurred, while incrementality seeks to identify the value that would not have existed without the investment.
Observe what happened outside the platform
Another limitation arises when the consumer does not buy in the same environment where they were reached.
They see the ad. They search for the product. They visit the store. They buy in person.
If the company only looks at website conversions, part of the impact disappears.
Google, for example, has features like Store Visits and Store Sales specifically to help eligible advertisers estimate how digital ad interactions influence physical store visits and sales.
The more omnichannel the journey, the less sense it makes to evaluate each channel as if it existed in isolation.
Compare the result with what would likely have happened without media.
This is where incrementality testing comes in.
One of the most intuitive ways is to work with comparable groups. One region receives a specific campaign. Another, similar one, does not.
Then, we compare the difference in the evolution of the results. It is not always possible to create a perfect experiment, but techniques such as geo tests, matched markets, holdouts, and econometric models help build a better estimate of the causal effect.
The most important question is not “how much was attributed to the campaign?”, but “what changed because we invested in it?”.
Don't try to find a single perfect metric.
ROAS still has value. As do conversion metrics: CPA, revenue, store visits, offline sales, and incrementality. The mistake is expecting a single metric to answer every question.
A more useful approach is to relate each indicator to a decision:
ROAS: where am I finding efficiency within the platform?
CPA: how much am I paying for a specific result?
Conversion: is the traffic generating the expected action?
Store Visits: is digital contributing to foot traffic in stores?
Incrementality: what portion of the result would likely not have happened without this investment?
MMM: how do media and other factors contribute to the aggregate result over time?
Good measurement doesn't just produce more metrics. It clarifies which decision each metric helps to make.
Before increasing investment
A high ROAS deserves attention, but also context.
If a campaign coincides with an aggressive promotion, increased distribution, or a major improvement in availability, attributing all growth to media can distort the analysis.
Mature measurement doesn't mean abandoning ROAS. It means putting it in the right place.
It is a piece of the answer, not the whole answer.
