What marginal ROAS measures
Return on ad spend, as normally reported, is an average: all the revenue attributed to a campaign divided by everything that campaign cost. Marginal ROAS asks a narrower question. If you add a little more budget to this campaign tomorrow, what does that extra money bring back on its own?
The two are almost never the same, because advertising returns diminish. The first money spent buys the cheapest, most obvious demand — people already searching for exactly what you sell. Extending beyond that means bidding higher for the same clicks, or reaching people with weaker intent, or both. Each additional unit of spend therefore returns less than the one before it, while the average, dragged upwards by all the cheap demand bought first, stays comfortably high.
Nobody reports marginal ROAS directly. It is inferred, by changing spend by a known amount, holding other things steady, and comparing the change in revenue with the change in cost.
Why marginal ROAS matters
It is the only figure that answers the question a budget decision actually asks. Average return tells you whether past spending was worthwhile; marginal return tells you whether the next spending will be. A campaign with a strong average may be entirely saturated, so adding money to it produces almost nothing extra, while a campaign with a weaker average may still be turning demand away.
This is also where scaling plans quietly fail. A business sees a healthy average, doubles the budget expecting the same ratio, and finds the additional spend performed far worse than what came before. The average did not lie — it simply described money already spent, at a level of demand that had already been picked over.
Where marginal ROAS goes wrong
The first error is treating it as a reported metric and hunting for it in the interface. It has to be worked out from a deliberate change, and that means changing one thing at a time and waiting long enough for delayed conversions to arrive.
The second is reading it from a period that contained something else. A seasonal peak, a competitor pausing, a stock problem or a price change will all move revenue independently of budget, and any of them can make an increase look brilliant or ruinous. Compare like with like, and prefer a longer window over a clean-looking short one.
The third is forgetting that the platform’s own attribution sits underneath all of this. If the campaign takes credit for sales it did not cause, marginal return will be overstated at every budget level, which is why an incrementality view is the honest companion to it.
What to do about it
Test in steps. Raise the budget on one campaign by a set amount, leave everything else alone, run it long enough to cover your usual lag between click and sale, then compare the extra revenue with the extra cost. Repeat until the extra return falls below the point where your margin still works. That point, not the average, is your real ceiling for that campaign.
Use the result to drive budget allocation: money belongs wherever the next increment returns most, which is often not the campaign with the best-looking report. And when the marginal return runs out everywhere, growth has to come from a better conversion rate, a higher order value or a new market rather than from more spend.
