What Is Marginal ROAS? Definition, Formula and How to Use It for UA Budgets

What is marginal ROAS?

Marginal ROAS is the return on ad spend generated by an increment of budget: the additional cohort revenue you got, divided by the additional spend it took, measured at the same cohort age. If moving daily spend from $10,000 to $15,000 lifted D30 cohort revenue from $4,500 to $6,000, the marginal D30 ROAS of that step is $1,500 ÷ $5,000 = 30%.

Blended ROAS on the same numbers would read 40%. Both are correct. They answer different questions.

Why does blended ROAS mislead scaling decisions?

Blended ROAS averages everything: the cheap, high-intent users you acquire first, the campaigns and markets that work best, and the newest spend at the edge of scale. Because returns usually diminish as a campaign reaches less responsive audiences, the average moves slowly while the margin moves fast.

That creates a familiar trap. Someone asks whether to increase the budget again. The blended number still looks healthy, so the answer is yes — even though the last increase may already be buying users who will never pay back. A game can have a healthy blended ROAS and still be approaching its current scale ceiling.

How do you calculate marginal ROAS?

Pick one cohort age — D7, D30, D90 — and use it for every comparison. Mixing ages is the most common error.
Find spend steps. Periods where daily spend held at one level long enough to produce comparable cohorts, then moved to another.
Take cohort revenue at that age for each level, from matured cohorts only.
Divide the change in revenue by the change in spend for each step.

The Marginal ROAS Calculator does this for any number of steps and plots blended against marginal.

Isn’t that just blended ROAS by campaign?

No. Slicing the average by campaign, source or country gives a finer-grained average, not the return on incremental spend. Inside each slice, the budget that would have been spent anyway is still mixed with the budget you added. To isolate the increment you need a change — a spend step, a geo holdout or a controlled test — and a comparison at equal cohort age.

What should the marginal ROAS target be?

Work backwards from payback. If your D30 ROAS historically becomes 100% by day 180, and a 180-day payback fits your cash position, then the D30 ROAS that historically leads there is your hurdle. The marginal step must clear it, not just the blend. Add a margin for uncertainty when cohorts are young.

What makes a marginal ROAS read unreliable?

Immature cohorts. Revenue at D7 on a two-day-old cohort is not a D7 number. See data maturity.
Other changes in the same window. A new creative, a new market or a seasonal shift will show up as a scale effect.
Small steps on small volume. Noise can swamp the signal; label the decision a learning increase and keep its size within what the evidence can support.
Mix shift. If the step changed which countries or platforms you bought, you are comparing different audiences.

How should teams use it day to day?

Report blended ROAS to leadership. Decide budgets on marginal ROAS. Treat each increase as a capped learning step with a stated evidence bar for the next one — what the current data can decide, how much exposure you accept while slower signals mature, and what has to be true before the next step.

Do not give a campaign more budget because the average still looks healthy. Give it more budget because the next block of spend still earns an acceptable return.