There’s a precise moment in monthly meetings when the conversation breaks down. The media buyer opens Ads Manager and shows a return on ad spend of 4.2. The business owner listens, nods, then opens the company’s management system and says that monthly revenue is essentially the same as it was in March, when they were spending half as much. From that point on, the meeting becomes an argument over who’s right, and the uncomfortable answer is that they’re both right. They’re looking at two different numbers that answer two different questions, and for years we’ve pretended they were the same number.
This overlap worked as long as tracking was deterministic. When the pixel could actually see the user’s journey, the data returned by the platform and the data recorded by the business tended to match, or at least be close enough that nobody had to feel awkward about it. That world no longer exists, and it didn’t disappear with an official announcement. It eroded slowly, through denied consent, App Tracking Transparency, declining cookie coverage, the Privacy Sandbox being shelved without the promised replacement, and finally modeled conversions, meaning numbers the platform estimates rather than observes. Today, when you see a conversion in a dashboard, in a far from insignificant share of cases you’re looking at a statistical prediction presented with the same appearance as a fact.
The Problem Isn’t That Platforms Lie
It’s worth being precise here, because the wrong accusation makes the right argument useless. Meta and Google aren’t inflating numbers to scam you. They’re doing exactly what their systems are designed to do: claim every conversion they can reasonably attribute within their own perimeter. The problem is that the perimeter is closed. Each platform sees only its own touchpoints, applies its own attribution windows, and concludes, in complete good faith, that it generated that sale. Then the other platform does the same thing with that very same sale.
The result is arithmetically observable, and anyone managing more than one channel has seen it at least once: if you add up the revenue the platforms claim to have generated, you get a number that’s higher than what you actually collected. The independent attribution platforms that have emerged in recent years, from Northbeam to Triple Whale, are built around this exact problem. Northbeam puts the systematic over-attribution of native dashboards at the center of its value proposition: every platform takes credit for the same conversion, and without an external layer to remove the double counting, you don’t have a single source of truth. You have three sources in conflict.
This is where what I believe is the real key to the entire issue comes into play, and it’s a distinction of roles before it’s a distinction of methodology.
The Technical Person Thinks Vertically, and That’s a Good Thing
Anyone optimizing an account needs to think vertically, and there’s nothing wrong with that. If I’m moving budget between two ad sets inside Meta, the data I need is Meta’s data, read according to Meta’s rules and compared against itself. I don’t need to know what Google did that week. I need to know which creative can sustain the acquisition cost and which one can’t. Platform ROAS is perfectly adequate for that decision. It’s a tactical signal, and as a tactical signal, it still works well.
The disaster happens when that number leaves the advertising account and enters the meeting room. Because the question the business is asking isn’t which creative is performing better. The question the business is asking is whether marketing produced more this month than it cost, all together, including the things that no dashboard measures. That’s an aggregate question, and you can’t answer an aggregate question with a vertical number, no matter how accurate that number is within its own context.
It’s exactly the same mistake as confusing the thermometer with the fever. Meta’s ROAS is an optimization tool that has been promoted in the field to the position of business health indicator, and nobody noticed the promotion.
What’s Taking Its Place
The answer the market has found isn’t another magic number, and that’s perhaps the hardest part to get across to clients who are used to asking how much a campaign generates. The direction shared by serious operators in 2026 is triangulation: multiple independent readings that check one another, each with a specific job.
At the highest level is blended data, which in its most common form is called MER, Marketing Efficiency Ratio: total revenue divided by total marketing spend, calculated using your numbers rather than the platforms’ numbers. It doesn’t depend on attribution, doesn’t need a pixel, and can’t be inflated by anyone because your revenue is what it is. It’s the view that can be reconciled with a profit and loss statement.
Below that is marketing mix modeling, which has made a major comeback for a very practical reason: it works with aggregated data and is therefore immune to the loss of individual-level signals. Google released Meridian as an open-source framework in 2025, and Meta has maintained Robyn since 2021. The free availability of these tools has removed a barrier that previously kept econometrics largely confined to six-figure consulting engagements. According to a TransUnion survey cited by eMarketer, nearly half of U.S. marketers say they plan to invest in MMM in the following year. It’s worth keeping a small light of skepticism on, though, and AdExchanger turned that light on in July 2026 with an article asking the right question: are models offered for free by the very platforms being measured a gift or a Trojan horse? Meridian’s value depends to a significant extent on data that only Google possesses, and that should give you pause.
Then there are incrementality tests, which remain the only available source of causal truth: turn something off in one geographic area and not another, keep a control group, and measure the difference. It’s difficult, unpopular, and often politically impossible to propose to a client who has already had bad experiences with campaigns being turned off. But it’s the only thing that truly answers the question of whether that spending is generating something that wouldn’t have happened without it.
And finally, at the bottom, there’s platform attribution. It isn’t disappearing, and it shouldn’t disappear. Its role is simply changing. From verdict to clue.
What Changes Tomorrow Morning
The operational consequence is simpler than all the reasoning that came before it, and it concerns a decision that most agencies have never made consciously: which number leaves the advertising account and which one stays inside it.
Inside the account, you keep platform ROAS, cost per result, frequency, and all the other tools needed by whoever is optimizing the campaigns. Outside, toward the client and the meeting room, you present an aggregated figure that the client can reconcile with what they see in their own management system. Once the two worlds are separated, that meeting where nobody understood what the other person was saying simply stops happening, because everyone is looking at the number that answers their own question.
The paradox is that this makes life harder for people in my line of work, not easier. An aggregated figure is more honest and far less flattering than a 4.2 ROAS. But it’s also the only ground on which a budget discussion can be a serious discussion rather than a competition over who has the prettiest slide.
In the next articles, I’ll go into detail on the two pieces that support this architecture: how to actually calculate MER and why the benchmark you find everywhere is probably wrong for you, and why the sum of the ROAS figures shown by the platforms is structurally larger than what you actually sold.
