The Inflation of Marketing Stats

In part 1 we looked at how much noise surrounds a paid media landscape. This article is about the number most teams use to see through it and why that number is bigger than the sales it describes.

The Over Reliance of Return On Ad Spend

Return on ad spend (ROAS) is revenue attributed to your ads divided by what you spent. Spend a dollar, get four back: 4:1, or 400%. It fits in one spreadsheet cell and it’s simple to digest but that is exactly why it can’t be the sole metric to run a business.

Four problems that come with ROAS:

1. It’s a ratio, not profit. A 4:1 on a $100 jacket at 25% margin loses money after shipping and returns. A 2:1 on a $60 supplement at 70% margin makes money. Same dashboard, opposite outcomes. ROAS falls short on tracking profit.
2. Platform that grade their own work. The revenue in that ratio is attributed and oddly enough, different platforms attribute more to themselves. Google provides a lopsided opinion of what it believes Google drove, and Meta’s ROAS opinion is often times in the corner of attributing more to Meta. Both will happily claim the same sale.
3. It rewards shrinking. The longterm optimization process when using solely ROAS to make adjustments, is that it often widdles away until you’re left with nothing. It will start to cut everything except brand search and retargeting, and ROAS soars initially while the business stops growing. When you take a step back and consider where those two tactics come from in the world of attribution, someone needs to already know the brand or have visited the website through other tactics – both of these are reactionary.
4. It only sees the first purchase. What about Lifetime Value (LTV) of a customer? A customer worth $400 over two years may appear to be a $60 loss on day one.

Why every number is inflated

There are two mechanisms do most of the inflating: attribution and modelling.

The first is attribution. Every platform’s attribution model is an opinion about the same sale, and adding opinions does not add revenue. Retargeting and brand search touch people last, so they claim most which is a statement about position in the journey, not about value.

The second is modelling. After Apple’s App Tracking Transparency prompt arrived in 2021, only around a quarter to a third of iOS users opted in to tracking; Meta told investors the change would cost it roughly $10 billion in 2022 revenue. In the UK and EU, where “Reject all” must be as easy as “Accept,” something like 30–50% of visitors decline cookies. The US works the other way round — opt-out rather than opt-in under state laws such as California’s CCPA/CPRA, now joined by around twenty other state privacy statutes — so measured consent gaps are smaller, but Global Privacy Control signals, Apple’s tracking prompt and browser tracking prevention remove a comparable share of the signal . Globally, roughly one internet user in three runs an ad blocker; US estimates sit a little lower, around a quarter to a third of adults . The platforms fill those gaps by modelling — estimating what the unseen users did from the ones they could see. That’s a reasonable thing to do. It also means every line-item ROAS in your account is partly an estimate, and the estimator has a stake in the answer.

The over-claim ratio

Here is the one calculation we’d ask every marketing team to run this month. It’s a Corkboard framework; the numbers below are a worked example.

1. From Google Ads, take the total conversion value for the month. Say $300,000.
2. From Meta Ads Manager, take the purchase (or lead) conversion value for the same month. Say $250,000.
3. From your order system or finance (not from any ad platform) take actual revenue for the month. Say $400,000.

Add the first two and divide by the third: $550,000 ÷ $400,000 = 137%. The platforms between them claimed 37% more revenue than existed.

When we’ve run this on real accounts the figure has landed well above 100% every time; the exact number depends on how much retargeting and brand search sit in the mix. Use GA4 as a third opinion if you like, but note it isn’t one of the claimants.

What it tells you, and what it doesn’t

A ratio above 100% is not a bug you can fix. It’s the structural consequence of several parties each attributing the same sale to themselves, on partly modelled data. What it does tell you is that line-by-line optimisation to platform ROAS is optimisation of noise pretending to have confidence and that you need at least one number that doesn’t come from anyone with a stake.

That number is Marketing Efficiency Ratio (MER), and it’s the subject of part 5 in this series. Before that, part 3 looks at what happens when the machine that produces the inflated number also decides where your money goes. 

Read next: Platform Bias in Automated Optimization (coming soon)

Related: What is MER in marketing?