In this episode of Unlocking Retail Media, Kevel CEO James Avery talks with Andrew Covato, founder of Growth by Science, about why most retail media networks are measuring the wrong things and what sophisticated brands actually need to prove incrementality.
Retail media networks face a measurement crisis. While platforms tout attribution metrics and last-click conversion data, the brands spending serious money are asking fundamentally different questions. They want proof of incrementality, not correlation. They need causal evidence, not user-level path data. And increasingly, they're walking away from networks that can't deliver it.
In this episode of Unlocking Retail Media, James sits down with Andrew Covato, a measurement marketing expert who has built and led analytics teams at Google, Facebook, Netflix, and Snap. Now through his firm Growth by Science, Covato helps brands build custom measurement systems that go far beyond what traditional attribution can provide. Here are the five biggest takeaways from their conversation about what retail media platforms must understand about modern measurement.
"Ad platforms are not in the business of optimizing advertiser outcomes. They're in the business of optimizing their own revenue."— Andrew Covato, Growth by Science
Covato opened with a blunt reality check about how advertising platforms approach measurement. Every platform, from the largest tech companies to emerging retail media networks, faces the same tension: they need to prove effectiveness to attract spend, but their primary goal is maximizing their own revenue.
"You can't not be a performing platform," Covato explained. "But you're also always trying to find an edge to showcase a story, to showcase the performance of your ads and create a narrative that attracts advertisers."
This fundamental conflict means advertisers cannot rely solely on platform-provided metrics. The major platforms support virtually any measurement approach that drives business, even ones they privately disagree with. If enough advertisers demand a particular methodology, the platform will accommodate it.
For retail media networks, this presents both a challenge and an opportunity. The challenge is building credibility with sophisticated advertisers who understand this dynamic. The opportunity is differentiating by offering genuinely objective measurement tools that these advertisers actually trust.
Modern Marketing Mix Models (MMM) analyze how changes in aggregate sales metrics correlate to changes in various inputs like ad spend, promotions, weather patterns, and competitive pricing. Critically, they do not require user-level data.
"It's a time series of aggregated data," Covato clarified. "You need it broken out by geography if you're feeding it into a hierarchical model. You may want it aggregated by different types of tactics. But it's not complicated at all."
Retail media networks often obsess over granular user-level path-to-purchase data, assuming this is what sophisticated advertisers need. Covato strongly disagrees with this focus. While some advertisers still care about attribution, the most advanced measurement approaches rely on aggregate data feeds that are actually simpler to provide.
What platforms need to deliver is straightforward: spend, reach, impressions, and proper time-series formatting. Geographic breakdowns help. Tactical segmentation (brand versus prospecting, different ad formats) adds value. But trying to track individual user journeys misses what MMM actually requires.
"By themselves, MMMs can be useful but dangerous," Covato cautioned. "They're not purely causal models. They're only as good as the assumptions you put in there."
This limitation leads directly to the next requirement.
Attribution shows correlation. Experimentation proves causation. Without true holdout tests, retail media networks cannot demonstrate incremental value to sophisticated advertisers.
"Everything we do is in service of estimating an 'i' something.Whether it's iROAS or incremental conversions or incremental whatever. That little 'i' means these are conversions you would not have gotten otherwise." – Andrew Covato, Growth by Science
Covato advocates for multi-channel holdout tests as a foundational starting point with almost all clients. The concept is simple but powerful: ask a CMO what the incrementality of their overall marketing program is, and most cannot answer. Running a broad holdout test across all channels establishes a baseline understanding of true incremental impact.
"You've got the attribution world that does not give you an 'i' anything," Covato said. "It gives you something happened and then something else happened. And then there's everything else where you've got a true experiment."
For retail media networks, this means supporting geo-level testing. Advertisers need the ability to exclude specific DMAs, zip codes, or states from campaigns. They need conversion data broken out by those same geographies. Without these capabilities, the platform simply cannot support the experimentation that proves incrementality.
Avery pressed on the practicalities: "The only way to do a global holdout is geo-based, because there's no user tie between Walmart, Facebook, and Snap, right?"
"Correct," Covato confirmed. "You pick cities or states or DMAs where you're not going to show ads."
Retail media networks typically emerge as secondary revenue streams for companies whose primary business is retail, not advertising. This creates a measurement gap that can kill otherwise promising ad businesses.
"Where I have seen measurement lag somewhat is definitely in the RMN space," Covato noted. "They're not typically the primary revenue stream. They're not caught up with what I would say are natives to ad tech."
This gap manifests in several ways. When sophisticated advertisers request MMM data feeds or exposure logs for experimentation, the requests are often misunderstood. Sales teams pivot to attribution dashboards. Tech teams struggle to provide the right data formats. Deals fall apart because the platform cannot support how the advertiser actually wants to measure performance.
"Where I find these deals fall apart is you get these asks for things like, 'Can I get data broken up this way so I can feed it into my MMM?' And the request is misunderstood," Covato explained. "If I can't measure it the way that I want to measure it, I'm not going to spend, especially if it's a relatively new platform."
The solution starts with having someone internal who deeply understands modern measurement approaches. Without that expertise, retail media networks cannot properly prioritize what to build, cannot correctly interpret advertiser requests, and cannot credibly discuss measurement with sophisticated buyers.
"It's very hard to go this road without having a leader that really understands it deeply," Covato emphasized. "Or at least a couple of folks in an organization somewhere that really get the nuances."
Many retail media networks consider building first-party lift products similar to what Meta, Google, and Snap offer. Covato supports this direction but cautions against positioning these tools as purely objective measurement.
"I would look at first-party lift as more of a sales tool versus a measurement tool," Covato advised. "Go into it fully acknowledging: look, we are doing our best to be objective. But you know, we understand that you may not look at this as purely objective because it's our own platform."
Building first-party lift requires significant investment. Done well, it can help advertisers graduate from rudimentary attribution to more sophisticated incrementality testing. The major platforms all offer these capabilities, and retail media networks should consider them as part of a measurement roadmap.
However, first-party lift has an inherent credibility problem: the platform is grading its own homework. Sophisticated advertisers understand this limitation. They may use first-party lift as directional guidance, but they want third-party validation or their own controlled experiments for high-stakes budget decisions.
"As a graduation from attribution, I think this is a great next step," Covato said. "You're helping the advertiser go along the spectrum of sophistication with the platform."
The graduated version involves third-party partnerships with measurement consultancies or SaaS platforms that can provide truly objective assessments. Some advertisers want to audit first-party lift products to verify their methodology. Others want to run their own geo tests with third-party analysis.
Platforms that acknowledge these limitations and support multiple measurement approaches will earn more trust than those claiming their first-party tools are definitive truth.
Retail media networks that cannot support modern measurement will lose access to the most sophisticated, highest-spending advertisers. These buyers no longer accept attribution as proof of performance. They demand incrementality evidence backed by real experimentation.
The technical requirements are not impossibly complex. Provide clean aggregate data feeds formatted for MMM. Support geo-level targeting and reporting for holdout tests. Build APIs that make programmatic data access straightforward. Break out metrics by meaningful dimensions like tactic type, product category, and audience segment.
More important than any specific technical capability is having internal expertise that understands what sophisticated advertisers need. Someone who can interpret requests for exposure logs, MMM feeds, and geo test capabilities. Someone who can articulate the difference between correlation and causation. Someone who recognizes that user-level attribution data, while valuable, is not what drives large budget allocations.
Covato was clear about the stakes: "If I can't measure it the way that I want to measure it, I'm not going to spend."
As retail media becomes more competitive and consolidated, measurement capabilities will increasingly separate winners from losers. Networks that acknowledge the limitations of attribution, invest in experimentation infrastructure, and hire measurement expertise will capture budgets. Those that continue pitching path-to-purchase dashboards will watch sophisticated advertisers spend elsewhere.
For more insights like these, tune in to the full episode of Unlocking Retail Media, the podcast where Kevel CEO James Avery sits down with industry leaders and innovators shaping the future of retail advertising.
Listen to this episode with Andrew Covato here.
