If you put a serious chunk of your digital budget into retail media and your CFO asks "is this working?", and you answer "uh… the ROAS looks good"… you're not alone. Half the CPG industry is in exactly the same place.

Here's a data point that puts it in perspective. When the Path to Purchase Institute surveyed CPG brands and agencies on the biggest challenges of working with retail media networks, the results were stark: analytics and reporting limitations were cited as the top obstacle by 53% of respondents (CPGs and agencies). You can't defend a budget you can't prove is working.

In this article we're going to unpack two things. First, why measuring retail media is genuinely hard (it's not that your team is bad — it's that the problem is hard). Second, what framework lets you honestly answer the CFO's question without falling into either the blind optimism of the retailer dashboard or the nihilism of "you can't measure anything."

Spoiler: it isn't about having a single perfect metric. It's about having four layers that complement each other. This guide is the map. The next four blog articles are the tools.

Why nobody is measuring retail media well (and nobody says it out loud)

There's a conversation that happens in every big-brand marketing team but rarely sees the light. It's this: nobody knows for sure if their retail media is working.

The brands say it: in a recent Reddit thread, a CPG brand manager described how "they moved 40% of their digital budget to retail media (Amazon, Walmart, Kroger, CVS)" and, although the ROAS "looks decent on paper," running the day-to-day is chaos. "Each platform is its own universe. Kroger Precision Marketing works completely differently from Amazon DSP. CVS Media Exchange has its own weird logic. You're not learning 'retail media' — you're learning five separate systems with nothing in common."

The agencies say it: someone from the agency side in that same thread put it this way: "We're all winging it. The biggest pain right now is proving incrementality. And of course retailers own that data, so there's basically no third-party verification." An uncomfortable but honest confession.

Why does this happen? Three reasons that tangle into one another.

The first is that each retailer is a different black box. Amazon measures sponsored impressions one way. Walmart Connect, another. Mercado Libre Ads, another. When you try to build a weekly dashboard that shows performance across all platforms, you realize Kroger's "impressions" aren't comparable with Amazon's. Forget about conversions — each retailer attributes with a different window and a different logic.

The second is the structural conflict of interest. The retailer is both the media network and the measurement system. It's like Coca-Cola counting goals in a match it's playing in. They aren't necessarily lying — they just have an incentive to show the most favorable angle. Initiatives like PrecisionView 360 from Kroger Precision Marketing are a step in the right direction, but they're still measurement produced by an interested party.

The third, and most fundamental, is that incrementality is a theoretically hard problem. To know whether a sale was caused by your ad, you'd need to know what would have happened in a parallel universe where the ad didn't exist. Since we don't have time machines, every measurement method is an approximation. The question isn't "do you have the perfect answer?" — it's "is your approximation reasonably useful for making decisions?"

The ROAS trap: why it looks good and means nothing

ROAS — return on ad spend — is the number every retail media dashboard hands you ready-to-serve. And it's almost always the wrong number for answering the incrementality question.

Why? Because ROAS counts as "ad-attributed sale" anyone who saw (or clicked) the ad and then bought. The problem is many of those people would have bought anyway.

Imagine your product is already the #1 organic result for the search "almond milk." Someone searches "almond milk," sees your sponsored ad at the top, clicks, and buys. The dashboard tells you: "ROAS 8x, congratulations!". But that person was going to buy your product regardless — they were two centimeters below in organic. You paid the retailer for a click you already had for free.

This is called brand-keyword cannibalization and it's a problem documented since 2015, when Blake, Nosko, and Tadelis ran a famous experiment at eBay (published in Econometrica) by pausing paid brand ads and discovering that the vast majority of traffic simply migrated to organic results. Total conversion barely changed. They paid millions for what they already had.

That's just one flavor of the problem. Here's another: when Uber decided to cut $35 million in Facebook and Instagram ads, nothing happened to its acquisition metrics. People kept downloading the app at the same rate. Does that mean the ads weren't working? Maybe. Or does it mean Uber already had enough brand awareness and the marginal ad effect was close to zero? Probably. But if you only look at the platform's ROAS, you'll never find out.

The ROAS trap, in summary: it's a number that looks rigorous, moves with your spend (because it measures activity correlated with your spend) and gives you false confidence. It's worse than admitting the problem is hard, because it makes you think you've already solved it.

Cat and mouse: brand vs. category incrementality

Before we go further, we need to clear up something almost nobody clarifies when talking about incrementality. There are two different definitions and people mix them up all the time.

Category incrementality asks: did the total category size grow because of my ad? Was more almond milk sold in the world thanks to my campaign? It's the economist's question. It's the question the platforms' measurement teams care about.

Brand incrementality asks: did MY sales grow because of my ad? Did I take share points from my competitor? It's the CFO's question. It's the question that pays your bonus.

For a brand, the answer to "which incrementality matters" is almost always the second. And there's a simple analogy to understand why.

Imagine your brand is a cat and sales are mice. There are two ways for the cat to eat:

1. The cat catches a mouse that was already there (you stole the sale from a competitor).
2. The cat finds a mouse that just appeared (you created new demand in the category).

To the brand's P&L, both mice count exactly the same. One unit on your truck, one peso in your books, one share point you added to your Nielsen. Pepsi doesn't get a bonus for growing the cola category — it gets a bonus for taking share points from Coca-Cola. The cat caught the mouse.

This is the angle many measurement models ignore. They obsess over proving category incrementality because it's the "pure" thing, and they end up telling the brand manager that their retail media "isn't incremental" when in reality it's moving share. The category can be flat while your brand grows — and that's exactly what you want.

This also rehabilitates an argument many critics dismiss: that retail media spend pushes organic position (because paid clicks feed the ranker's algorithm), and that the resulting organic position then captures sales that were going to competitors. Is that "real incrementality"? For the economist, no. For the brand's CFO, absolutely yes.

The right question: are we winning the digital shelf?

Once you accept that the important question is brand incrementality (not category), the operational question simplifies. You no longer need a six-month geo experiment to answer the most relevant thing. The question becomes this:

Are we winning the digital shelf?

The digital shelf is what your customers see when they search, browse, or are recommended on Mercado Libre, Amazon, Éxito, Olímpica, Jumbo. It isn't an abstraction — it's literally where the purchase decision is made. It's the modern equivalent of the linear shelf, but measured continuously and with much more data.

If your brand is improving its position on that shelf — if it's higher for more keywords, more visible across more categories, closing gaps where it was weak — you're winning. And if you're winning there, that growth is going to translate into sales, without needing a perfect experiment to prove which exact cent came from which exact ad.

This isn't surrender. It's a recalibration of rigor: instead of looking for the perfect answer to the impossible attribution question, look for the best possible answer to "are you winning or losing?". The first can't be answered well. The second can.

The 4 layers you need to answer it well

A single metric doesn't solve this. What does solve it is a system of four layers that complement each other. Each one compensates for the weaknesses of the others. Here are the four layers, and each one has its own article on this blog where it's explained in detail.

Layer 1 — The thermometer: weighted first-page organic position. Your organic position for the keywords that matter, weighted by search volume, measured weekly. It's your vital sign. It doesn't tell you what made you sick, but it tells you whether you're getting better or worse. Read more: Your organic position in ecommerce is a thermometer: learn to read it.

Layer 2 — The matrix: organic × paid. For each important keyword, simultaneously measure your organic position and your sponsored share-of-voice. The combination reveals things neither metric alone can show — starting with where you're paying for clicks you already had for free. Read more: Looking at your paid and organic position together: the trick that changes everything.

Layer 3 — The detective: observational multivariate attribution. Modeling organic position as a function of your paid spend, relative pricing, promotions, content score, review velocity, and competitor actions. It doesn't give you a perfect causal answer, but it does give you a directionally honest answer about what's moving what. Better than darkness. Read more: What's moving your ranking? How to separate noise from signals.

Layer 4 — The race: real-time operational scraping. Detecting competitor stockouts, price changes, content changes, and sponsored share-of-voice ramps, and reacting with smart bidding. This layer pays for itself every day and is the foundation feeding the other three. Read more: Stockouts, prices, and smart bidding: the real-time retail media play.

The four together give you something no retailer dashboard can give: an honest, multi-source, non-proprietary view of whether your retail media spend is building durable position on the digital shelf.

Why total darkness is the real enemy

There's a legitimate critique of this framework. If you've been following along, you saw it coming: "but this isn't pure causal attribution. Your positions could improve for a thousand reasons that aren't your retail media — a competitor went out of stock, you launched new packaging, reviews grew, there was an aggressive promo."

True. The four layers together don't give you pure causal attribution. For that, you need a controlled geo experiment, where you turn off spend in some markets and let it run in others, and compare trajectories. It's the gold standard and it exists — Meta Conversion Lift, Google Ads Geo Experiments, the Brand Lift studies from Amazon DSP. But it's expensive, slow (4-8 weeks minimum), technically demanding, and only feasible at a handful of the most mature retail media networks.

The real choice, in practice, isn't between perfect causal attribution and the four-layer framework. The real choice is between the four-layer framework and the retailer dashboard + a PowerPoint that combines non-comparable data + hope. And between those two, there's no contest.

The phrase that best captures the spirit of this approach is simple: knowing you're on a good path is better than being in total darkness. That's exactly what these four layers give you. Not darkness. Not certainty. But direction. And for making budget decisions quarter to quarter, direction is what you need.

Once a year, if your retail media spend exceeds 20% of your revenue, yes: run a formal geo experiment on your largest line of investment. It's the calibration check. The rest of the year, the four layers are your operating system.

The IAB/MRC standard: the common vocabulary the industry is adopting

If you've been working in retail media these past few years, you know that every conversation with a retailer starts with a tacit negotiation about what each metric means. That friction has a new name: in January 2024, IAB and the Media Rating Council (MRC) published the first Retail Media Measurement Guidelines, a joint standard that IAB Europe complemented with its own version in April 2024. The Executive Playbook developed with Boston Consulting Group translates the technical document into actionable steps.

What the standard puts in common agreement:

What the standard doesn't solve, and why this four-layer framework is still necessary: the structural conflict of interest. A retailer can be 100% IAB-compliant in how it counts and still report only what suits it to show. The standard aligns vocabulary; it doesn't replace independent verification. Think of it this way: IAB/MRC is the rulebook of the game — a level field for every team. But you still need an independent referee for the specific match your brand is playing.

In summary

Question Short answer
Is measuring retail media hard? Yes, and 53% of brands report measurement pain.
Is ROAS a good answer? No, it overstates due to keyword cannibalization and retailer conflicts of interest.
What question CAN be answered? "Are we winning the digital shelf?" — via brand incrementality.
How? Four layers: organic thermometer, organic×paid matrix, observational attribution, and operational scraping.
And pure causal attribution? For annual geo experiments, not for weekly management.

If you're operating retail media in Colombia or LATAM in 2026, you have two possible paths: keep the retailer dashboard and pray, or build the four-layer view that tells you whether you're winning or losing on the shelf where your customers decide. The first path is comfortable and doesn't work. The second is honest, sustainable, and is built with consistent scraping plus an analytical layer on top.

ePerfectStore.com is built around this thesis. The next four articles detail each of the layers and how they look in practice for a brand selling on Mercado Libre, Amazon, Éxito, Jumbo, Olímpica, or any ecommerce in the region.

Sources

What does your real position on the digital shelf look like? ePerfectStore.com measures the four layers across every Colombian and LATAM retailer, without depending on the retailer's dashboard.

Back to Blog