You run a campaign on Google. You run one on Meta. Both dashboards show a conversion for the same customer. You made one sale. Two platforms claimed it. Congratulations — you’ve just experienced the central lie of modern digital advertising attribution.
This isn’t a bug. It’s the architecture.
Every major ad platform operates a closed attribution system where the incentive to show you favorable numbers is baked into the business model. They grade their own homework. And the gap between what those dashboards report and what your CRM actually shows keeps getting wider.
Platforms Grade Their Own Homework
Flor Zanetic at ECD Digital Strategy laid this out plainly in October 2025: Google and Meta each run attribution systems designed to maximize credit for themselves, not to give you an accurate picture of what’s actually driving revenue. When a user clicks a Google ad, sees a Meta ad, and then buys something, both platforms count a conversion. Your CRM counts one sale. The math doesn’t resolve.
This isn’t cynicism — it’s structural. Ad platforms earn revenue when you spend more. Showing you inflated conversion counts makes you spend more. The incentive to report accurately is basically zero. What you’re trusting, every time you look at a ROAS figure in Ads Manager or Google Ads, is a number produced by an entity with a direct financial interest in making that number look good.
The gap between platform reporting and backend reality has gotten bad enough that marketers at serious organizations now treat platform conversion data as a rough signal, not a source of truth. The ones still optimizing off platform ROAS alone are flying blind with a very confident co-pilot.
The Numbers Are Actually Damning
Lee Riley at Funnel.io surfaced a stat from Cannes Lions 2024 that deserves to be repeated more than it is: only about 6% of advertising drives any meaningful value. Six percent. The attribution problem isn’t just that platforms double-count — it’s that even perfect attribution couldn’t identify which 6%.
The ROI illusions compound. Keen’s econometric analysis found Display advertising delivering $2.02 ROI on a platform dashboard while the actual marginal return had fallen below $1. Meaning every incremental dollar spent was losing money, while the reporting said it was printing it.
One DTC brand Riley cited scaled Performance Max aggressively after watching ROAS climb from 300% to 800%. Impressive, right? Actual new customer growth stayed flat. The algorithm had found an efficient way to claim credit for customers who would have bought anyway — brand searches, email clickers, loyal repeat buyers. ROAS went up. Business value did not. The platform called it a win.
GA4’s Own Models Can’t Agree With Each Other
Even within a single platform, the attribution math breaks down. Killian Walsh at Silverback Labs documented this in April 2025: comparing Data Driven attribution to Last Click attribution inside GA4 regularly produces discrepancies of 40% or more on the same conversion data.
That’s not a rounding error. That’s a philosophical disagreement happening inside one tool on one dataset.
And here’s the part that should bother you: Data Driven Attribution is now the default in GA4. Google switched to it without asking. It’s a machine learning model. It’s a black box. Walsh’s point is that marketers can’t audit how it assigns credit — you can’t see the weights, you can’t challenge the logic, you can’t verify anything. You just have to trust the output from the same company that sells the ads you’re measuring.
Apple Quietly Broke Mobile Attribution in 2021 — The Effects Are Still Compounding
The App Tracking Transparency framework Apple launched in 2021 didn’t just inconvenience Facebook. It broke the foundational data layer that mobile attribution depended on. Lifesight’s 2024 analysis put the numbers in stark terms: IDFA opt-in rates dropped from 70-80% pre-ATT to around 27% afterward.
When you can’t track 73% of the audience, you’re not tracking. You’re sampling badly and calling it measurement.
What fills that gap? Modeled data. Platform estimates. Statistical inference from the small slice of users who opted in and the assumption that they behave like the ones who didn’t. The resulting ROAS numbers look real. They’re not. Lifesight’s research found that more than 70% of purchase decisions complete before buyers ever engage a trackable channel — meaning the attribution models aren’t just inaccurate, they’re measuring the wrong part of the funnel entirely.
The “Messy Middle” Was Always Unmeasurable
Here’s what the industry calls it: the Messy Middle. That stretch between first awareness and final purchase where real buying decisions get made, across a dozen channels, over days or weeks, with no linear path.
Funnel.io has documented repeatedly how GA4 struggles to track multi-channel journeys coherently. And why would it succeed? The customer sees a YouTube pre-roll, reads a Reddit thread, gets a retargeting ad on Instagram, Googles the brand name, sees a review on a blog, maybe clicks an email, and then converts. Attribution models look at that chain and try to assign credit according to a set of rules — first touch, last touch, linear, time decay, data-driven — as if any ruleset could accurately represent a human being making a real decision.
It can’t. Attribution models aren’t describing reality. They’re choosing a story to tell about it. The story you pick determines who gets budget.
Bad Data Breaks the Algorithm
There’s a feedback loop here that most marketers aren’t thinking about carefully enough. Verde Media’s 2025 research on Smart Bidding laid it out: flawed conversion data goes into the bidding algorithm, the algorithm miscalibrates based on that data, performance gets worse, the worse performance generates even worse data, and the cycle tightens.
Verde’s characterization was blunt: the algorithm optimizes “confidently in the wrong direction.” Smart Bidding is powerful when it has accurate signal. Feed it platform-reported ROAS figures that have double-counted conversions, modeled in missing mobile data, and credited the wrong touchpoints? The machine gets very good at doing the wrong thing.
The optimization is working exactly as designed. The inputs are garbage.
SEO Attribution Is Collapsing Too
It’s not just paid. Search Engine Land’s reporting in 2025 called out something that anyone doing SEO has felt: AI-generated answers in search results are eating click-through rates, and the traffic that once provided the data trail for SEO attribution is evaporating.
What’s left is a situation where SEO’s influence is real and measurable through business outcomes, but the specific attribution chain — keyword to click to session to conversion — is breaking apart. You can see that organic visibility drives revenue in aggregate. You can’t always prove which content, which keyword, or which touchpoint did it.
“A phase where we know SEO’s influence is real, we just can’t prove it,” is how the situation was described. That’s honest. Most attribution frameworks aren’t.
Attribution Was Always a Philosophical Problem
The attribution crisis isn’t new. It’s just more visible now.
MarTech Series documented the shift in 2023-2024: serious practitioners have largely given up on the search for a perfect attribution model. Not because they’re lazy — because the problem is philosophically unsolvable.
Les Binet put it best in The Drum back in February 2023. Binet, whose work on marketing effectiveness is about as rigorous as this field gets, said: “If you say this ad generates a million in sales, the true answer could be anything between zero to a million.” That range isn’t uncertainty due to bad data. It’s the inherent indeterminacy of causal attribution in a complex system.
There’s a concept in measurement theory called the Fallacy of Immediacy — the tendency to attribute outcomes to the most recent, visible, measurable cause rather than the actual driving forces. Digital attribution has this baked in structurally. What gets measured is what’s trackable. What’s trackable is the last click, the final touchpoint, the bottom of the funnel. The years of brand equity, the word of mouth, the content consumption that shaped the purchase intent — none of that shows up in a dashboard.
So it doesn’t get credited. And therefore it doesn’t get funded. And eventually it doesn’t get built.
What This Means for Anyone Running Ads
The attribution model you’re using right now is telling a story. It’s a story with real consequences for where you allocate budget, how you evaluate campaigns, and how you think your marketing is performing. That story was written by the platforms selling you the ads, using models they built, with rules they set, that they’re not required to explain or justify.
That’s not a conspiracy. It’s just a conflict of interest at industrial scale, running silently in the background of every marketing decision made in digital advertising today.
The question isn’t whether your attribution is broken. It is. The question is whether you’re building your strategy on top of that fact or pretending it away.