The Attribution Number Every Agency Hopes You Won't Question

Picture the monthly report. Meta says it drove a 6x return. Google says it drove a 5x. Add them up and the agency is generating an 11x return on your money, which would make them the best investment you have ever made and a small miracle of modern capitalism.

Now add up the actual revenue in your bank account. It doesn't match. It's not close. Somewhere between the dashboards and the deposits, a couple of times your real revenue evaporated.

Nobody is lying, exactly. Each platform is reporting accurately by its own rules. But the number on that report is the most flattering possible version of reality, and most agencies are happy to let you keep believing it, because the alternative is a harder conversation about how much of that return would have happened anyway.

Key takeaways

• Meta, Google, and TikTok collectively overcount conversions by roughly 20 to 40 percent, because each one claims credit for the same sales.
• Meta reports around 26 percent more conversions than analytics tools on average. Google over-attributes by 15 to 20 percent when modeled conversions kick in.
• A lot of "conversions" are modeled, meaning statistically estimated, not observed. Meta's own documentation says so.
• The honest question is not "who gets credit" but "would this sale have happened anyway." That's incrementality, and branded search is where the two numbers diverge most violently.
• Measure with blended ROAS (revenue over total spend) for a daily reality check, and incrementality tests for the truth. About 52 percent of US marketers now run incrementality testing.

Contents

1. Why the dashboards always look amazing
2. The three numbers, and which one is honest
3. Branded search: the most flattering lie in the report
4. How to find your real number
5. What an honest report looks like
6. FAQ

Why the dashboards always look amazing

The inflation isn't a bug. It's the natural result of letting each platform grade its own homework.

Start with double-counting. A customer sees your Meta ad, later searches your brand and clicks a Google ad, then buys. Meta claims that sale. Google claims that sale. If Microsoft was in the mix, it claims it too. One sale, three trophies. Ruler Analytics and others have documented how this duplication inflates reported ROAS and hides your true cost per acquisition, sometimes by close to 2x. The platforms aren't comparing notes, and they have no incentive to.

Then there's modeling. After iOS tracking limits and consent rules cut into observed data, the platforms started filling the gaps with statistical estimates. Meta's own documentation admits these modeled conversions are not observed events. Cross-platform analyses find Meta reporting roughly 26 percent more conversions than independent analytics tools, and Google over-attributing by 15 to 20 percent once enhanced conversions and Consent Mode modeling are active. Stack double-counting on top of modeling and you get reports that show 20 to 40 percent more conversions than actually happened.

Here's the part that should bother you. The platform you spend on is also the platform that tells you how well that spend worked. That's like asking a salesperson to write their own commission check. The incentive runs one direction, and it isn't toward humility.

The three numbers, and which one is honest

There are really three different things people call "ROAS," and they tell you three different stories.

Platform-reported ROAS is fine for one job: optimizing inside that platform, where its own numbers are at least internally consistent. It is useless for deciding whether the channel deserves your money, because it can't see the other channels and it counts sales it didn't cause.

Blended ROAS, sometimes called MER, is just total revenue divided by total ad spend. It's crude. It won't tell you which channel did the work. But it cannot be gamed, because it starts from money you actually received. If your blended number is healthy and stable while you scale, you're probably fine. If every platform claims a 5x and your blended number is 1.8x, you have your answer about who's exaggerating.

Incremental ROAS is the honest one. It asks the only question that matters: how many of these sales would not have happened without the ad. That's the number agencies hope you won't ask for, because it's almost always lower than the dashboard, and sometimes brutally so.

Branded search: the most flattering lie in the report

If you want to see the gap in one place, look at branded search.

A customer types your company name into Google, your ad shows up above your own organic listing, they click it, they buy. Last-click attribution records a spectacular ROAS. The agency screenshots it. Everyone feels great.

But that customer was already searching for you by name. They had decided. In most cases they would have clicked your organic result and bought anyway, for free. The ad didn't create the sale. It charged you for a click you were going to get regardless. This is exactly why analysts call branded search the channel where last-click ROAS and true incrementality diverge most violently, and the largest budget correction available per test dollar.

It gets starker on the non-branded side. In one geo holdout test, a major grocery chain found a 0 percent sales lift from non-branded paid search and moved that budget to CTV instead. Zero. The dashboard had been reporting a return the entire time on spend that was causing no incremental sales at all. That's not a rounding error. That's a strategy built on a number that was never real.

How to find your real number

You don't need a data science team to stop fooling yourself. Here's the system we use.

1. Anchor on blended ROAS first. Every week, divide total revenue by total ad spend. This is your truth serum. It ignores platform politics entirely. Watch how it moves as you change spend, because the direction of that line is more honest than any in-platform report.

2. Deduplicate your tracking. If you run client-side and server-side tracking without event deduplication, you're double-counting before the platforms even get involved. Fix deduplication and your reported conversions will drop to something closer to reality. Lower and true beats higher and fake.

3. Run geo holdout tests on your biggest line items. Turn a channel off in a set of matched regions, leave it on elsewhere, and measure the difference in actual sales. This is the cleanest causal read most businesses can run, and it's how that grocery chain found its zero. Start with branded search and any channel you're afraid to question.

4. Use the lift tools the platforms offer, with skepticism. Conversion lift and geo experiments inside Meta and Google are useful, but remember who's running them. Cross-check against your own holdouts and your blended number.

5. Reallocate on incrementality, not on the dashboard. When a test shows a channel is taking credit without driving lift, move the money. The whole point of measuring honestly is to spend differently. A number you don't act on is just decoration.

This isn't fringe practice anymore. As of mid-2025, about 52 percent of US marketers run incrementality testing, and it's become the primary KPI in retail media. The businesses doing it are quietly reallocating budget away from channels that were winning only on paper.

What an honest report looks like

At Adgo we lead with the blended number, because it's the one tied to your bank account, and we treat platform-reported ROAS as a tuning gauge, not a scoreboard. When a channel matters enough to argue about, we test it with a holdout rather than quote you the platform's self-graded figure.

That makes for less impressive screenshots. An honest report rarely shows an 11x. It shows a real blended return, a note on what each channel actually contributed, and a recommendation to move money toward what's genuinely incremental. It's a less exciting slide and a much better business.

FAQ

Why don't my Meta, Google, and GA4 numbers match?

Because each uses different attribution windows and models, and each claims credit for overlapping sales. Meta tends to report the most, analytics tools the least. None is "right," which is why your own blended revenue is the better anchor.

What is the difference between attribution and incrementality?

Attribution decides which touchpoint gets credit for a sale. Incrementality decides whether the sale would have happened at all without the ad. Attribution can make a channel look great while incrementality shows it caused nothing.

What is blended ROAS or MER?

Total revenue divided by total advertising spend across all channels. It won't tell you which channel did the work, but it can't be inflated, so it's the best quick sanity check against platform reports.

Is branded search a waste of money then?

Not always. It can protect against competitors bidding on your name and capture intent you'd otherwise lose. But its last-click ROAS is wildly overstated, so size the budget based on incremental lift, not the dashboard figure.

Do I need expensive software to measure incrementality?

No. A well-run geo holdout test, turning a channel off in matched regions, gives you a clean causal read with tools you already have. Software helps at scale, but the method is what matters.

The takeaway

The dashboard is not your friend. It's a marketing document produced by the company you're paying, designed to make the spend look as good as possible. That doesn't make it fraud. It makes it a sales tool, and you should read it like one.

The number worth trusting is the boring one: real revenue over real spend, validated by turning channels off and watching what actually changes. Ask your agency for that number. The good ones already lead with it. The rest are hoping you never ask.

Related reading

• When everyone runs the same targeting, your data is the edge
• Where a Saskatoon business should put its first $5k in ads

Sources: GoFunnel: how platform-reported conversions inflate revenue 20-40%, Ruler Analytics: conversion double-counting, EasyInsights: why conversions don't match across Meta, Google, GA4, Measured: platform ROAS to true incrementality, Lifesight: geo-based incrementality testing, eMarketer: FAQ on incrementality 2026