Here is a conversation that happens in every ecommerce business, roughly monthly. Meta reports a 4.1x return. The team proposes increasing budget. Someone senior asks whether that will hold at a higher spend. Nobody can answer, so the budget goes up, the return drifts down, and three months later the same conversation happens with smaller numbers.
The reason it repeats is that the number being discussed cannot answer the question being asked.
What average ROAS is actually measuring
Average ROAS divides all attributed revenue by all spend. That figure blends together two very different things: demand you created, and demand that already existed and would have converted anyway. Branded search is the purest example. It posts spectacular returns while largely intercepting people who had already decided to buy from you.
So the average is dragged upward by your cheapest, least incremental inventory. Scale the budget and the new money goes into progressively less efficient territory, while the average, anchored by that efficient base, barely moves. By the time the average visibly deteriorates, you have been overspending for months.
The number that does answer the question
Marginal ROAS asks: what return did the last increment of spend produce? Not the average across everything, but the slope at the point where you currently sit.
You measure it by moving budget deliberately and watching what happens. Raise spend 15%, hold everything else stable, and measure the change in total revenue against the change in total spend, ideally with a geo-holdout running underneath so you can separate your increase from the season, the promotion and the competitor who happened to go dark that week.
Why this changes how accounts get built
Once marginal return is the metric, several familiar habits stop making sense.
- 01Campaign proliferation becomes expensive. Eleven campaigns competing for one audience fragment learning and inflate CPMs without adding reach.
- 02Creative supply becomes the binding constraint. Marginal return decays as frequency climbs, so the only way to hold the slope while scaling is a steady flow of genuinely new angles.
- 03Budget changes become experiments. Each increase carries a pre-registered hypothesis and a stop rule, so a failure produces a decision rather than an argument.
- 04Channel cuts get faster. When a channel fails its own test, it gets cut that week, not at the quarterly review, by which point the money is gone.
The uncomfortable part
Measuring marginal return honestly usually reveals that a brand's true efficient spend ceiling is lower than its current spend. That is a difficult finding to deliver and a more difficult one to hear.
But the alternative is worse. A business scaling on average ROAS is not growing on the strength of its advertising. It is growing on the strength of its reporting, and reporting does not deposit funds.