In Q1 2025, an e-commerce fashion brand ran $50,000 in monthly ad spend across Meta and Google. The platform dashboard reported $250,000 in attributed revenue. A 5.0x ROAS. The performance marketing team celebrated. The CFO approved the budget expansion.
There was only one problem: that number was wrong by 426 percent.
A Celerian Digital audit — correcting for cross-channel attribution overlap, deducting organic baseline sales, accounting for COGS, fulfillment, and payment processing fees — revealed the true ROAS was 0.95x. The brand was spending $67,300 all-in to generate $64,220 in profit contribution. They were profitable on less than one dollar returned per dollar spent. The dashboard was showing them something that had not happened.
This is not an isolated case. Celerian Digital's 2025 performance audit across multiple e-commerce advertisers found an average 38 percent difference between platform-reported and true ROAS. High-COGS businesses — those above 60 percent margin compression — showed a 52 percent gap. Fashion and apparel, with generous returns policies built into their unit economics, showed a 45 percent gap. Wicked Reports, in December 2025, confirmed the phenomenon with a broader study finding platform ROAS inflated by 20 to 50 percent across categories, a pattern they termed the "Performance Trap" — dashboards showing 4 to 6x ROAS while P&L statements show flat revenue.
The fashion brand's situation is the industry's situation in miniature. The metric that determines where $258 billion in annual US digital ad spend gets allocated is, on average, off by between a third and a half. And the audit that would reveal this gap — if it exists at all — happens rarely, inconsistently, and usually only after significant budget damage.
The gap between reported and true ROAS is not a bug in the measurement system. It is a feature of how attribution works — specifically, of how platforms assign credit for conversions they did not exclusively cause.
The dominant mechanism is attribution overlap. Rockerbox research found that the average e-commerce conversion involves 6.8 touchpoints across 3.2 different platforms. When each platform claims full credit for the entire conversion, the sum of claimed revenues exceeds actual revenue by 142 percent. Meta and Google overlap in 25 to 45 percent of conversions. Retargeting overlaps with organic intent in 60 to 80 percent of cases. Advertisers running five or more platforms see their attributed revenue inflated by 180 to 200 percent — the same dollars being credited multiple times by multiple systems that never coordinated on the accounting rules.
The average e-commerce conversion involves 6.8 touchpoints across 3.2 platforms. When each platform claims full credit, the sum of claimed revenues exceeds actual revenue by 142 percent.
Platforms also claim credit for baseline organic sales — users who would have searched for the brand, found the website, and purchased regardless of the ad. Brand search inflation runs 70 to 90 percent — those users were already in the funnel, already typing the brand name. Retargeting inflation runs 60 to 80 percent — those users had already demonstrated intent, often had the product in a cart, and would have returned. The ad receives credit for a conversion that the customer journey was going to produce anyway.
View-through conversion windows compound the problem. Facebook's default includes a 1-day view-through attribution. Google Display and YouTube run up to 30-day view-through windows. Without a proper holdout group — a control that did not see the ad — there is no way to distinguish between users who converted because of the ad and users who were going to convert anyway. Industry research estimates 40 to 60 percent of view-through conversions are non-incremental. They are baseline behavior that the platform is claiming as induced behavior.
Repeat buyer blending creates another layer. When existing customers click retargeting ads, platforms count them as new conversions. This makes prospecting look more expensive and retargeting look more efficient than it is. New customer CAC appears artificially low. The model that emerges from this data systematically undervalues brand-building and overvalues retention-focused tactics — even when the business needs new customers to grow.
The attribution model at the foundation of most platform reporting is last-click — the convention of assigning full conversion credit to the final touchpoint before purchase. Last-click is familiar, intuitive, and deeply wrong.
Research from Meta's own internal analysis — published in the European Journal of Computer Science and Information Technology in 2025 — found last-click attribution gives search advertising 94 percent more credit than is warranted by actual customer behavior. Display advertising is undervalued by approximately 36 percent under last-click. In some cases, the discrepancy reaches 143 percent when comparing last-click attribution to more sophisticated approaches.
A landmark study by Randall Lewis and Justin Rao, published in the Quarterly Journal of Economics in 2016, ran 25 large field experiments across $2.8 million in ad spend and found that "the median confidence interval on ROI is over 100 percentage points wide." The statistically small impact of profitable advertising means that selection bias is, in their words, "a crippling concern for widely-employed observational methods."
The median confidence interval on ROI is over 100 percentage points wide. Selection bias is a crippling concern for widely-employed observational methods.
The most rigorous comparison came from Gordon, Moakler, and Zettelmeyer's 2024 study using 2,226 randomized Meta ad experiments. Last-click attribution achieved an R² of 0.19 for predicting incremental conversions per dollar — meaning it explains less than a fifth of the variance in actual causal impact. A methodology called Predictive Incrementality by Experimentation (PIE) achieved R² of 0.88. The average prediction error from last-click was 491 percent. In 12 to 20 percent of campaigns, last-click attribution would lead to the opposite budget decision from what incrementality testing would recommend.
Last-click doesn't just misrepresent ROAS. It actively misdirects budget. A retargeting campaign showing 8x ROAS and a prospecting campaign showing 2.5x ROAS will always recommend shifting budget toward retargeting — unless someone knows that the retargeting's true incremental ROAS is 2x (mostly claiming organic conversions) while prospecting's true ROAS is also 2x (actually acquiring new customers). The platform recommendation is rational given what it can see. What it can see is not the truth.
The structural cause of ROAS inflation is incentive misalignment. Platforms sell advertising. Their revenue depends on advertisers continuing to spend. The platform that reports lower ROAS loses budget to the platform that reports higher ROAS, all else being equal. This creates a systematic pressure toward the most optimistic reasonable attribution methodology.
As one industry analyst at Cresva put it: "Platforms have a structural incentive to show you the best possible numbers. Their revenue depends on you continuing to spend." Google and Meta trained the industry to grade their own homework. The agencies that managed the money were incentivized to believe the numbers — good numbers kept the budgets flowing. The CFOs who approved the budgets saw the dashboards, not the underlying economics.
The Meta whistleblower complaint filed in August 2025 by former product manager Samujjal Purkayastha illustrated the dynamic with unusual clarity. Purkayastha alleged Meta artificially inflated Shops ads ROAS by 17 to 19 percent by counting shipping fees and taxes as revenue — revenue that did not represent actual margin contribution. Meta included these fees in revenue calculations for Shops ads but not for other ad products or competitors like Google, which use net revenue. Internal Meta reviews in early 2024 discovered the inflation. The complaint was filed. The question of how widespread this practice might be across product lines remains unresolved.
Google and Meta trained the industry to grade their own homework. The agency that managed the money was incentivized to believe the numbers. Good numbers kept the budgets flowing.
The industry knows this is happening. Sixty-five percent of marketers still use platform-provided attribution as their primary methodology, according to an October 2025 survey. But 46.9 percent plan to increase investment in Marketing Mix Modeling over the next 12 months. Thirty-four point seven percent plan increased multitouch attribution investment. The measurement infrastructure is slowly, reluctantly, beginning to catch up to the problem.
Google lowered the minimum cost for a conversion lift study from $100,000 to $5,000 in May 2025 — a democratization of incrementality testing that signals the measurement arms race has reached the mid-market. Meta and Google have both released open-source MMM tools — Robyn and Meridian respectively — making enterprise-grade measurement accessible to brands without eight-figure media budgets. Incrementality testing is no longer exclusively for companies that can afford to run $500,000 experiments.
The CFO is becoming the de facto CMO. This is not a compliment. It is an observation about which metric has won. CFOs do not care about attributed revenue. They care about bank balance. And when the dashboard shows 10x ROAS while the bank account shows flat revenue, a reckoning arrives that the performance marketing ecosystem has been structurally avoiding.
The brands that have run proper incrementality tests — the kind that use holdout groups and randomized control trials rather than platform-reported conversion data — discover they are wasting an average of 23 percent of marketing spend on non-incremental activities, according to Marketing Science Institute research published in 2025. One CPG brand found through testing that heavy TV advertising drove only 8 percent incremental sales despite attribution models showing 35 percent of conversions from TV-influenced views. A $2.3 million annual budget reallocation followed, producing a 31 percent improvement in marketing ROI within two quarters.
The fashion brand's dashboard still shows 5.0x ROAS when the meeting room screen shares the campaign review. The CFO still approves the budget. The platform still takes credit for $250,000 in revenue that was always going to arrive. Somewhere in the building, the person running the incrementality test knows the number is a reflection — accurate in its shape, inverted in its substance. The gap between what the mirror shows and what is actually there has a name now. It is a 426 percent problem. And it is the foundation on which $258 billion in annual spend is allocated.
· · ·