In short
Incremental ROAS, or iROAS, is the sales your ads caused divided by what you spent on them. An iROAS of 1.5 means every $1 of ads brought in $1.50 of sales that wouldn't have happened otherwise. From a geo holdout test, take the ratio of ads-on to holdout sales before the test, multiply holdout sales during the test by that ratio to get the expected ads-on sales, and subtract that from what the ads-on regions actually sold. The difference is incremental sales. Divide it by the Meta spend in the ads-on regions only, and use before and during periods of the same length. A good iROAS is above your break-even ROAS. Below 1, the ads cause less in sales than they cost, before any product cost.
What is incremental ROAS (iROAS)?
ROAS, return on ad spend, divides sales by ad spend. The catch is which sales. Meta's ROAS uses the purchases it credits to your ads. Amazon Attribution's uses sales after tagged clicks. Both count shoppers who would have bought anyway, and both miss some who wouldn't.
Incremental ROAS (iROAS) uses only the sales your ads caused: what you'd lose if you switched them off. An iROAS of 1.5 means every $1 of ads brought in $1.50 of sales that wouldn't have happened otherwise. It's the number to compare with your break-even ROAS when you decide whether to spend more, less or nothing.
You can't read iROAS off a dashboard. You measure it with a test, and the most direct test for Meta ads that sell on Amazon is a geo holdout: pause ads in some regions, keep them on in others, and compare Amazon sales.
How to calculate incremental ROAS
pre-test ratio = ads-on sales before ÷ holdout sales before expected sales = holdout sales during × pre-test ratio incremental sales = ads-on sales during − expected sales lift = incremental sales ÷ expected sales iROAS = incremental sales ÷ Meta spend in ads-on regions caused vs. credited = incremental sales ÷ Amazon Attribution sales
- Ads-on regions: where Meta ads kept running during the test.
- Holdout regions: where Meta ads were paused.
- Before: a period the same length just before the test, when ads ran everywhere. It tells you how the two groups normally compare, since regions are never the same size.
- Meta spend in ads-on regions: only the spend that could have caused the extra sales.
The idea: the holdout regions show what sales did without ads. Scale them by the usual ratio between the groups, and you get what the ads-on regions would have sold without ads. Everything above that line is what the ads caused.
Worked example
A made-up brand runs a six-week test. These are the calculator's example values.
- Before the test, the ads-on regions sold $45,000 and the holdout regions $40,000. Pre-test ratio = 1.125.
- During the test, the holdout regions sold $40,000. Expected ads-on sales without ads = $40,000 × 1.125 = $45,000.
- The ads-on regions actually sold $54,600. Incremental sales = $54,600 − $45,000 = $9,600, a lift of 21.3%.
- The brand spent $6,000 on Meta in the ads-on regions. iROAS = $9,600 ÷ $6,000 = 1.60.
- Amazon Attribution credited $3,600, an attributed ROAS of 0.60. The ads caused 2.67× what Attribution credited.
Judged by Amazon Attribution alone, these ads look like they lose money. Measured, they return 1.60. Whether that's profitable depends on the brand's margins.
What's a good iROAS?
A good iROAS is one above your break-even ROAS: the ROAS where ad-driven sales just cover the ads after product cost and fees. Below it, each extra sale costs more than it makes. Find yours with the break-even ROAS calculator.
- Below 1: the ads cause less in sales than they cost, before any product cost. Almost always unprofitable.
- Between 1 and break-even: the ads bring in sales but lose money on them, unless repeat orders make up the difference.
- Above break-even: each ad dollar earns its keep. Consider spending more, and test again at the higher level, since returns usually fall as spend rises.
Break-even ROAS varies a lot by category, price and margin, so there's no single good iROAS across brands.
Common mistakes
- Using total spend. Divide by the spend in the ads-on regions only. Using national spend makes iROAS look worse than it is.
- Unequal periods. The before and during periods must be the same length, or the ratio is off.
- Reading one number as certain. A test gives an estimate with a range. A 1.6 with a range of 0.4 to 2.8 is a different decision from 1.6 with a range of 1.3 to 1.9.
- Ignoring stockouts and promotions. A stockout or sale in only one group during the test shows up as fake lift, or hides a real one.
- Reading the result too early. Shoppers can buy days after seeing an ad. Wait a week or two after the test ends.
How this connects to true incrementality
iROAS is incrementality expressed as a return. Before you run a test, check it can see the lift you care about with the holdout test calculator. After, compare the result with what Meta and Amazon reported using the attribution gap calculator, or see the sales Attribution missed with the halo effect calculator.
This page gives a point estimate. Spillmark measures the same thing from your daily order data, with an uncertainty range, and keeps every test so the next one starts from what you learned. See how lift is calculated.
Frequently asked questions
What does iROAS mean?
Incremental return on ad spend: the sales your ads caused, divided by what you spent on them. It leaves out sales that would have happened without the ads.
What's the difference between ROAS and iROAS?
ROAS uses the sales an ad platform credits to your ads. iROAS uses only the sales the ads caused, measured with a test. It can be lower than Meta's reported ROAS and higher than Amazon Attribution's.
Why compare to a period before the test?
Regions never sell the same amount. The pre-test ratio shows how the two groups normally compare, so a difference during the test can be put down to the ads.
Can iROAS be negative?
A single test can come out negative by chance, especially a small one. A clearly negative result usually points to a problem with the test, such as a stockout in the ads-on regions.
How do I get an uncertainty range for iROAS?
You need the daily or weekly sales for each group, not just totals, to see how much they normally move. Spillmark reports the range alongside every result.