Juliette Potuznik, VP sales at AI-powered marketing intelligence platform BillyGrace, explains why ROAS is the number everyone trusts – but shouldn’t.
ROAS is everywhere. It sits at the top of every performance report, anchors every quarterly review, and drives most of the decisions that matter in a campaign. Ask a brand what good looks like, and they’ll almost always reach for it.
The problem isn’t that ROAS is wrong. It’s that it’s incomplete in a way that systematically benefits the channels reporting it, which makes it dangerous to rely on without question. Each reports on;ly what it can see, and it can only see itself.
I spent years at Google watching this from the inside. I know how channel attribution models work, what they’re designed to measure, and what they can’t see. The number agencies were being held to was produced by a system with no view of the full picture. Not through bad intent. The limitation is structural. But the effect is the same: agencies make and defend decisions based on data that can only ever tell part of the story.
A visibility problem, not a dishonesty problem
The easy version of this argument is that channels lie to make themselves look good. That’s not quite right, and it’s worth being clear about the distinction.
Google isn’t lying when it reports a 4x return on your search spend. Meta isn’t lying when it shows you a cost-per-acquisition that looks efficient. They’re reporting what they can see. The issue is that each channel only sees its own slice of the customer journey. Google sees the search click. Meta sees the social touchpoint. Neither sees what came before, what came after, or what would have happened anyway.
So when Google attributes a conversion to a search click, it may well be right. But it can’t tell you whether that conversion would have happened without the Meta campaign that sparked the intent three days earlier. It can’t tell you whether a brand awareness campaign on YouTube shifted the behaviour that made the search happen. Each channel is, in the most literal sense, grading its own homework with no access to the test paper.
But the visibility problem runs deeper than missing data. Even if you could combine every channel’s reporting into one dashboard, you’d still only know what happened, not whether your spend was the reason, or whether it would have happened anyway. That’s the incrementality question, and it’s the one channel-level attribution was never designed to answer.
The industry has a name for what this produces: double-counting. Google, Meta, TikTok, Pinterest – every channel reports the world as it appears from its own window, claiming credit for conversions that two or three other platforms are claiming simultaneously. Incrementality research makes this concrete. Tests on Performance Max campaigns that include brand search terms regularly find that switching off the campaign reduces total sales by only 30% of what Google attributes to it. The remaining 70% represents conversions that would have happened anyway, through organic search, direct visits, or other channels Google simply couldn’t see. You’re not buying performance. You’re buying credit for performance that was already coming.
What agencies are actually being asked to prove
The pressure agencies are under right now isn’t really about ROAS. It’s about business outcomes. Clients want to know: are we growing? Are we acquiring the right customers? Is our marketing budget generating a return we can defend to the board?
ROAS is a proxy for those questions, not an answer to them. The gap between the proxy and the reality is where agency relationships break down. Clients sense it. They see ROAS improving while revenue flatlines. They see spend scaling while margins narrow. They start asking questions the channel data can’t answer, and the agency, armed only with platform reports, struggles to respond.
The numbers bear this out in a way that goes beyond measurement accuracy. Analysis of brand performance data across the Billy Grace platform found that declining brands carried a blended ROAS of 7.05. Growing brands averaged 4.00. On paper, the decliners look more efficient. In practice, they’re harvesting existing demand while defunding the channels that build new demand. Growing brands put 48% of their budget into upper- and mid-funnel activity. Declining brands managed 22%. Growers invested twice as much on Meta and nine times more per customer on TikTok, the channels that build intent rather than capture it. That gap widened quarter after quarter: from 14% in Q1 to 26% by Q4. ROAS didn’t flag the problem. It actively encouraged the decisions causing it.
Independent measurement isn’t optional any more
Third-party cookie deprecation has made this more urgent. Privacy regulation has made it more complex. The customer journey, already fragmented across devices, channels, and touchpoints, is getting harder to track, not easier. The conditions that made channel-reported ROAS a reasonable shortcut are disappearing.
What’s replacing it is incrementality testing: the discipline of asking not just “did this campaign generate conversions?” but “would those conversions have happened without this campaign?” The methodology is more straightforward than it sounds. Pause a channel in one region, run it normally everywhere else, and measure the divergence. This is called a geo-holdout test. For Google, the impact shows up directly in conversions. For Meta, it tends to show up in traffic, because the intent Meta builds often converts later and elsewhere. That distinction alone changes how you’d think about budget allocation between the two.
It’s a harder question to answer than blended ROAS. But it’s the only one that gives you an honest view of what’s actually working.
The agencies getting ahead
The agencies getting ahead aren’t defending their ROAS harder. They’re the ones who can walk into a client meeting and say: here’s what the channel reported, and here’s what we independently verified actually drove results. Now we know where to double down and where we’re wasting budget. That’s how you move from execution partner to strategic advisor, and it’s the shift the industry has talked about for years without quite achieving it.
