Is Your ROAS Real? How Ad Platforms Inflate Attribution by 134%

The Question You Can't Answer

Your Meta dashboard says ROAS is 4.2x. Your Google Ads dashboard says 6.8x. Your CFO asks: "So we made 4-7x on every dollar?" And you say yes, because the platforms say so.

But did you?

Every ad platform measures its own performance. That's like letting students grade their own exams. And a 792-model analysis of real marketing data says the grades are inflated.

How Much Do Ad Platforms Over-Report Conversions?

In 2025, Cassandra published the most comprehensive public analysis of platform attribution accuracy: 792 marketing mix models across hundreds of brands.

Source: Cassandra, 792-model analysis (2025)

These aren't worst cases. These are medians. Half of the campaigns measured were inflated more than this.

For every dollar of ROAS Meta reports, roughly $0.43 of it was actually caused by the ad. The rest was organic traffic, brand recognition, other channels, or people who would have bought anyway.

Why Meta Inflates More Than Google

The gap between Meta (134%) and Google (18%) isn't random. It's structural.

Google captures high-intent users. Someone searching "buy running shoes" was probably going to buy running shoes. Google Ads gets credit, but the ad's causal contribution is small. The inflation is lower because many of those conversions would have happened organically.

Meta creates demand from scroll. Someone scrolling Instagram wasn't planning to buy anything. When they convert days later, Meta claims credit—but did the ad cause it, or did the person see 12 other touchpoints in between? Meta's 7-day click window and 1-day view window are designed to capture the widest possible credit.

Neither platform is lying. Both are structurally incentivised to take maximum credit.

What Did Dropbox Find When They Measured Incrementality?

In March 2026, Dropbox's Marketing Data Science team published a peer-reviewed paper in IEEE Access: "From Attribution to Causality in Digital Advertising."

Their methodology: month-long geo-blackout experiments across the US, using Difference-in-Differences, Bayesian Structural Time Series (CausalImpact), and GeoLift analysis.

Click-based attribution overstated actual performance by 2-10x depending on the channel.

This wasn't a vendor selling incrementality testing. This was Dropbox's own PhD team running controlled experiments with millions in ad spend.

Honesty matters here: Dropbox's approach required a PhD-level data science team, millions in monthly ad spend, and month-long channel blackouts. Most companies spending $60K-$1.2M/year can't replicate this. That doesn't mean the data is irrelevant to you. It means you need a different approach to the same problem.

What Is ROAS Inflation? (And Why It's Getting Worse)

ROAS inflation is the gap between what ad platforms report and what actually happened. It's driven by three structural forces—and two recent changes are making it worse.

The Three Structural Causes

Double-counting. A customer clicks your Google ad on Monday, sees your Meta retargeting ad on Wednesday, buys on Friday. Google claims the sale. Meta claims the sale. You made one sale. Platforms claimed two.

View-through attribution. Meta's default: if someone saw your ad (didn't click, just scrolled past it) and converted within 1 day, Meta claims it. On a high-traffic brand, thousands of "conversions" are people who happened to scroll past an ad before doing what they were already going to do.

Self-serving attribution windows. Google Ads uses 30-day click, Meta uses 7-day click, LinkedIn uses 90-day click. Each platform sets its windows to maximise its own credit.

Two Changes Making It Worse Right Now

These changes mean platform-reported ROAS is becoming less tethered to reality, not more. The modelling that replaces lost signal is a black box. You have less visibility into what's real.

What Percentage of Ad Budget Is Wasted?

The 792-model analysis found that 20-35% of ad budgets flow to channels producing zero incremental return.

Not low return. Zero.

The usual culprits: branded search (bidding on your own name for clicks you'd get organically), retargeting oversaturation (every platform claiming credit for the same warm lead), and view-through attribution on high-impression campaigns.

The 5-Minute ROAS Reality Check

You don't need a PhD team to spot inflation. You need two numbers and a calculator.

Step 1: Pull Platform-Reported Revenue (2 minutes)

Open each ad platform. Pull revenue attributed to ads for last month.

Step 2: Pull Actual Revenue (2 minutes)

Open your CRM, Stripe, or backend. Pull total revenue for the same period. Estimate what percentage came from paid channels (vs organic, direct, referral).

If you can't estimate the paid percentage, use total revenue. The ratio will be conservative—reality is even worse than the number you get.

Step 3: Calculate the Ratio (1 minute)

Inflation ratio = Total platform-claimed revenue ÷ Paid-attributable revenue

Example:
  Platforms claim: $480,000
  CRM paid-attributable: $210,000
  Ratio: $480,000 ÷ $210,000 = 2.29x

Step 4: Read Your Ratio

How to Fix Inflated ROAS Reporting

Three levels, depending on your budget and where you are.

Immediately: Separate Your Metrics

Use platform ROAS for one thing only: within-platform optimisation. Which ad creative works? Which audience converts? Which keywords are efficient? Platform data is good at this.

For cross-channel decisions ("should we spend more on Meta or Google?"), switch to CRM-based revenue:

Real ROAS = CRM revenue from paid ÷ Total ad spend

Not: Platform-reported revenue ÷ Total ad spend

This Month: Add Independent Measurement ($29-299/mo)

Deploy a neutral attribution tool that sees all touchpoints and applies consistent logic. Server-side tracking captures 30-40% more data than client-side because it doesn't depend on browser JavaScript that ad blockers and iOS 26 strip out.

When Meta says ROAS is 4.2x and your independent tool says 2.1x, you know the truth is closer to 2.1x. The goal isn't perfection—it's a second opinion.

This Quarter: Run an Incrementality Test ($5K+)

Turn off a channel in a geographic region. Measure whether conversions actually drop. If they don't, the channel wasn't incremental. Google's Conversion Lift now starts at $5K minimum (down from $100K in 2024). Meta and TikTok have similar tools.

This is the gold standard but requires enough spend per region for statistical significance. If you're spending $60K-$1.2M/year, start with independent measurement first. You can't run geo-holdout experiments at that spend level, but you can absolutely measure whether your platforms are double-counting.

Related Reading

• Why Your Platform Reports Don't Match — the technical reasons platforms disagree
• Why Google Ads and GA4 Show Different Numbers — the specific Google Ads vs GA4 discrepancy explained
• Why Did GA4 Remove Most Attribution Models? — what changed and your options
• MTA vs MMM: Which Do You Actually Need? — choosing the right measurement approach for your spend level

Holly Henderson

Co-Founder, mbuzz

Holly Henderson is Co-Founder of mbuzz. With 10+ years in marketing including roles at Westpac, Avon, and Forebrite, she's obsessed with making measurement actually useful.

Harvard Extension School Forebrite Westpac Avon