In short
Marginal ROAS is the revenue you get from the next dollar of ad spend. Average ROAS is total revenue divided by total spend. Because ads have diminishing returns, the next dollar usually returns less than the average, so a campaign can show a healthy average ROAS while its latest spend loses money. To estimate marginal ROAS, fit a curve of sales against spend through several months, ideally with a 2× gap between your lowest and highest month, and read its slope at your current spend. The curve's shape, b, tells you how fast returns fall: at 0.5, the next dollar returns half the average. Keep raising spend while marginal ROAS is above break-even. The curve shows correlation; a holdout test measures cause.
What is marginal ROAS?
Marginal ROAS is the revenue you get from the next dollar of ad spend. Average ROAS is total revenue ÷ total spend. The two are different because ads have diminishing returns: the first dollars reach the people most likely to buy, and each extra dollar reaches people a little less likely to buy.
That's why a campaign can show a healthy average ROAS while the last thousand dollars you added lost money. The average hides it, because the early, cheap sales prop it up. When you decide whether to raise or cut a budget, marginal ROAS is the number that matters.
How to calculate marginal ROAS
You need sales at different spend levels. This calculator fits a curve through your months, then reads the slope of that curve at your current spend:
sales = a × spend^b average ROAS = sales ÷ spend = a × spend^(b − 1) marginal ROAS = slope of the curve = b × average ROAS break-even spend = the spend where marginal ROAS = break-even ROAS
- b is the curve's shape. Below 1, each extra dollar returns less. At 0.5, the next dollar returns half the average.
- a sets the curve's height. The calculator finds a and b with a least-squares fit on the logs of spend and sales.
- Break-even ROAS is 1 ÷ your margin before ads. The break-even ROAS calculator works it out from your costs.
A simpler way, without a curve: compare two months. If spend rose from $1,000 to $1,500 and sales rose by $1,000, the extra $500 returned 2.00. That works only if nothing else changed between the two months, which is rare. A curve through several months smooths out some of that noise.
Worked example
A made-up brand has 6 months of Meta spend, from $1,000 to $4,000 a month. Sales went from $4,800 to $10,600. These are the calculator's example values.
- The fitted curve is sales = 82.49 × spend^0.59. It explains 99% of the month-to-month change, so it fits well.
- At today's $4,000 a month, the curve gives $10,634 of sales: an average ROAS of 2.66.
- Marginal ROAS = 0.59 × 2.66 = 1.56. The next $100 brings in about $155.75.
- The brand's break-even ROAS is 2.29. The next dollar returns less than that, so it loses money. Marginal ROAS falls to 2.29 at about $1,577 a month.
An average ROAS of 2.66 looks safely above break-even. But the spend added above about $1,577 a month is costing more than it earns. This is the trap marginal ROAS exists to catch.
What's a good marginal ROAS?
There's no industry number to aim for. A good marginal ROAS is one above your break-even ROAS:
- Above break-even: the next dollar still makes money. Raising spend should add profit, until marginal ROAS falls to break-even.
- At break-even: you're at the most profitable spend level. More spend adds sales but not profit.
- Below break-even: the last dollars lose money, even if average ROAS looks fine. Trimming spend should raise profit.
Some brands spend past break-even on purpose, to win new customers who buy again later. If that's you, compare marginal ROAS with a break-even that counts repeat orders. The max CAC calculator helps set that limit.
Common mistakes
- Judging a budget by average ROAS. The average can stay high long after the extra spend stopped paying.
- Using months where spend barely changed. If spend is flat, there's no way to see what more or less would do. You need a spread, ideally a 2× gap between your lowest and highest month.
- Reading the curve too far out. The curve is only fitted where you've spent. Doubling spend beyond your highest month is a guess.
- Mixing seasons. If you spent more in your busy months, the curve credits ads with sales the season brought. Use months from the same part of the year if you can.
- Using ad platform revenue. Platform-reported revenue counts sales that would have happened anyway. Use total sales.
How this connects to true incrementality
This curve is a correlation: it shows that sales were higher in months when you spent more. It can't prove the ads caused the difference. Brands often spend more in months that would be good anyway, which makes ads look better than they are.
A holdout test fixes that. Pause ads in some regions and compare, and you measure what the ads cause at your current spend. Run tests at two spend levels and you have two measured points on the curve, not guesses. Plan one with the holdout test calculator, then use the curve shape in the ad budget allocator to split spend between Meta and Amazon Ads.
Spillmark runs those tests and keeps a record of each one, so every budget change starts from what your ads have been measured to do. See how lift is calculated.
Frequently asked questions
What's the difference between marginal ROAS and average ROAS?
Average ROAS is total sales ÷ total spend. Marginal ROAS is the sales from the next dollar. Because returns shrink as spend grows, marginal ROAS is usually lower than average.
How many months of data do I need?
At least 3, and 6 or more is better. What matters most is that spend varied: months at very different spend levels give a much clearer curve than many similar months.
Why is my curve shape (b) above 1?
It means sales grew faster than spend. That's rare for ads alone and usually means something else grew sales in your high-spend months, like a season, a launch or a price change.
Can I use weekly data instead of monthly?
Yes, as long as each row is the same length of time. Weekly data is noisier, so use more rows and expect a looser fit.
Should I cut spend to the break-even level right away?
Move in steps. The curve is an estimate, so cut part of the way, watch total sales for a few weeks, and confirm with a holdout test before going further.