What ROAS actually tells you, and what it hides

A 6x return reported inside an ad platform is a claim made by the party being paid. It is not a lie, but it is not evidence of profit either.

Return on ad spend is revenue attributed to a campaign divided by what that campaign cost. Every part of that sentence is doing more work than it appears to. “Attributed” is a modelling decision. “Revenue” is usually gross, not contribution. And “cost” typically excludes the creative, the agency and the discount you funded to close the sale.

Three reasons the platform number is high

• Attribution windows. A seven-day click and one-day view window will credit purchases that a stricter window would not. Compare two campaigns on different windows and you are comparing nothing.
• Double counting. Google and Meta will both claim the same conversion. Add up platform-reported revenue across channels and it routinely exceeds what the finance system recorded.
• Demand you already had. Retargeting and brand-term bidding intercept people who were coming anyway. Their reported return is excellent and their incremental return is often close to zero.

The number to put beside it

Marketing efficiency ratio — total revenue divided by total marketing spend — has one great virtue: it cannot be gamed by attribution, because neither term is attributed. It is blunt, it says nothing about which channel worked, and it is the honest headline. Track blended MER against platform ROAS. When they diverge, the platform is telling you a story about credit, not about growth.

Then add contribution margin. A 4x return on a product with 25 percent margin loses money. Plenty of D2C brands scaled for two years on a ROAS target that was structurally below break-even, which is an expensive way to learn what a gross margin is.

Incrementality, cheaply

The only reliable way to know whether spend caused revenue is to withhold it and watch. You do not need a research budget for this:

• Geographic holdouts. Turn a channel off in two comparable cities for four weeks. Compare against matched cities. Crude, but it answers the actual question.
• Brand-term test. Pause bidding on your own name for a fortnight. Most brands find organic absorbs the majority of that traffic. Some find it does not, which is also worth knowing.
• Spend-step tests. Move budget up or down by 30 percent and watch total orders, not attributed ones. Small changes teach you nothing; noise swallows them.

Why targets drift

A ROAS target set once and never revisited becomes a ceiling on growth. Efficiency and volume trade against each other: the efficient spend is by definition the spend against people closest to buying, and there are only so many of them. Insisting on a fixed ROAS while demanding growth is asking a team to find more of the cheapest demand than exists.

The better frame is a contribution target at a given volume, revisited quarterly. That lets you buy expensive customers deliberately when the payback justifies it, instead of pretending they are cheap ones.

What to report to a board

Blended MER and contribution, incremental results from any test you ran, cost per acquisition by cohort rather than by campaign, and payback period. Platform ROAS belongs in the channel review, not in the board pack, for the same reason you would not take a vendor’s own estimate of their value to the audit committee.

None of this makes paid media less valuable. It makes it legible, which is a precondition for spending more on it with confidence. If your tracking cannot support this, fix that first: see a GA4 setup that survives scrutiny and the measurement gap.

How far off can platform attribution be?

Far enough to change decisions. In a study published in Marketing Science in 2019, Gordon and colleagues compared fifteen randomized advertising experiments at Facebook with the observational methods advertisers commonly use, and found the observational estimates frequently failed to match the experimental ones, even with extensive controls. ROAS reported by a platform is an observational estimate made by the party being paid, which is why it belongs in its own column next to a measured number, not in the headline.

Is there a cheap way to check?

Yes, and it is getting cheaper. A geographic holdout, running a campaign in comparable markets and not in others, needs no platform cooperation and answers the only question that matters: what happened that would not have happened anyway. For a portfolio view, open-source mix models now cost nothing to run; Google made its Meridian model available to all marketers in January 2025. Neither replaces judgement, but both replace taking the platform's word for it.

A worked example

An illustrative account, to show the arithmetic rather than describe a market. A platform reports a ROAS of 6: Rs 60 lakh of attributed revenue on Rs 10 lakh of spend. Now apply the corrections this article describes. Contribution after product cost, fulfilment, returns and discounts is 35 percent of revenue, so the attributed sales contributed Rs 21 lakh. A holdout shows that half of those sales would have happened anyway, so the incremental contribution is Rs 10.5 lakh, against Rs 10 lakh of spend plus Rs 1.5 lakh of creative and fees. The campaign the dashboard called a 6x success lost money.

None of the inputs were exotic. The difference came from three questions the platform report never asks: what did each sale actually contribute, what would have happened without the spend, and what did the campaign really cost.