Marginal ROAS

Average ROAS divides all revenue by all spend. Marginal ROAS divides the change in revenue by the change in spend between two budget levels. The first describes the account you have. The second describes the decision in front of you, which is almost always whether to spend more or less than you spend now.

The two diverge because returns diminish. The first dollars in an account reach the most responsive buyers: people searching the brand, recent visitors, the audiences the algorithm finds cheapest. Each further increment has to reach someone less ready to buy, so each increment returns less than the one before it. An account can therefore report a healthy average while its last increment loses money.

A worked example, as arithmetic. At $40,000 a month an account returns $140,000, a 3.5x average. Raised to $60,000, it returns $172,000, a 2.87x average that still looks comfortable. The marginal ROAS on the extra $20,000 is $32,000 divided by $20,000, or 1.6x. If contribution margin is 45%, breakeven ROAS is about 2.22x, so the last $20,000 destroyed contribution even though the average never dropped below 2.8x.

The stop rule follows directly. Keep adding spend while marginal ROAS is above breakeven ROAS (one divided by contribution margin), and stop when it falls below. Scaling on the average instead is the most common way a profitable account becomes an unprofitable one without any single metric looking alarming.

Marginal ROAS is an estimate, not a reading. The cleanest estimates come from deliberate spend steps held for long enough to see the response, or from holdout tests; day-to-day correlation between spend and revenue is noisier and is confounded by promotions and seasonality. State which method produced the number, and treat a single week as a hint rather than a verdict.