iROAS vs ROAS: why your best campaign might be your worst

Return on ad spend (ROAS) is the most quoted number in performance marketing and one of the most misleading. It tells you how much revenue happened near your ads. It doesn't tell you how much revenue happened because of them. Incremental ROAS (iROAS) does, and the gap between the two is often where the biggest budget mistakes hide.

The two formulas

ROAS = attributed revenue ÷ ad spend.

"Attributed" means the revenue an ad platform or analytics tool assigns to a campaign under its attribution rules: for example, purchases within seven days of a click or one day of a view. A ROAS of 5 means the tool credits the campaign with $5 of revenue for every $1 spent.

iROAS = incremental revenue ÷ ad spend.

"Incremental" means revenue that would not have happened without the ads. You can't read it off a dashboard; you estimate it by comparing a group exposed to the ads with a comparable group that wasn't, in a controlled experiment such as a geo (matched market) test or a platform lift study.

Two related measures are worth knowing:

• Cost per incremental acquisition (CPIA) = ad spend ÷ incremental customers (or conversions).
• Marginal iROAS = incremental revenue from the next block of spend ÷ that extra spend. This is the number budget decisions should use, because the average return of a channel tells you little about what an extra dollar will do.

Why the two numbers differ

Attributed ROAS counts every conversion that meets the attribution rule, whether or not the ad changed anything. Three patterns inflate it:

1. Selection. Ads target people already likely to buy: past visitors, loyal customers, people searching your brand name. Their purchases get credited to the ad even though many would have happened anyway.
2. Overlapping credit. Google, Meta, TikTok and your email platform may each claim the same order. Add up platform-reported revenue and it often exceeds your actual revenue.
3. View-through credit. An impression scrolled past on a phone can earn credit for a purchase made a day later for unrelated reasons.

The research record is blunt about this. In a large controlled experiment at eBay, switching off branded keyword ads left almost all of the traffic intact, because users simply clicked the organic result instead (Blake, Nosko & Tadelis, Econometrica, 2015 (opens in a new tab)). A study of 15 large advertising experiments at Facebook found that common observational methods, the kind attribution relies on, frequently failed to recover the effects measured by the randomized tests, and often overstated them (Gordon et al., Marketing Science, 2019 (opens in a new tab)).

None of this means attribution is useless. It is fast, granular and fine for comparing ads within a campaign. It just can't answer "is this channel worth the money?" on its own.

A worked example

Take three campaigns from a hypothetical online retailer, each spending $20,000 in a month. The attributed numbers come from the ad platform; the incremental numbers come from tests.

Illustrative numbers only, chosen to show the mechanics. Your results will differ, which is the point of testing.

On ROAS, branded search is the star and prospecting video looks marginal. On iROAS, the ranking flips. The video campaign has an incremental share above 100% because it drives sales the platform never sees (people who watch, then buy later through search or direct visits). Branded search mostly harvests demand that existed already.

If this retailer moved budget toward the highest ROAS, it would shift money from the campaign that creates demand to the one that captures it, and total sales would fall even as reported ROAS rose. That is the trap in the title.

How to measure iROAS

You need a control group. In order of how independent the measurement is:

• Geo experiments. Change spend in some regions and not others, then compare sales in your own systems. Works for almost any channel and needs no user-level tracking. Open-source tools such as Meta's GeoLift (opens in a new tab) and Google's Trimmed Match (opens in a new tab) handle design and analysis. See our step-by-step guide for Canada.
• Platform lift studies. Meta Conversion Lift and Google Ads Conversion Lift (opens in a new tab) randomize users or regions inside the platform. Convenient and often free for eligible advertisers, but they measure one platform with that platform's conversion data.
• Holdout audiences. For owned channels like email or SMS, keep a random slice of customers out of a campaign and compare. Simple and powerful.
• Natural experiments. An outage, a regional launch, a budget freeze. Useful, but rarely clean. Treat them as clues, not proof.

To compute iROAS from a test, divide the incremental revenue by the spend difference between test and control. For a holdout, that's the spend you would have had in the held-out regions. Always report a confidence interval: "iROAS 2.1 (90% interval 1.2 to 3.0)" is far more useful than a single number.

What's a "good" iROAS?

It depends on your margins, not on industry benchmarks. Revenue-based iROAS needs to be compared with your gross margin. If your contribution margin after product, shipping and payment costs is 40%, an iROAS of 2.5 is roughly break-even on the first purchase (2.5 × 40% = 1.0). Above that, the campaign is profitable on first orders. Below it, it only pays if repeat purchases or lifetime value make up the difference.

So the useful thresholds are:

• Break-even iROAS = 1 ÷ contribution margin.
• Target iROAS = break-even iROAS adjusted for lifetime value and your growth goals. A company buying market share may accept iROAS near break-even; one optimizing cash flow won't.

Using iROAS day to day: calibration factors

You can't run an experiment every week. The practical approach is to use each test to calibrate the attributed numbers you already have:

Calibration factor = iROAS from the test ÷ platform ROAS over the same period.

In the example above, branded search has a factor of 0.1, retargeting 0.25 and prospecting video 1.2. Multiply the platform's daily ROAS by the factor to get an estimated iROAS for reporting and bidding decisions. Re-test every six to twelve months, or after major changes in creative, targeting or competition, because the factors drift.

Calibrated targets also make automation safer. Google's Target ROAS and Target CPA bidding optimize toward the conversion values and goals you give them. If those goals reflect attributed rather than incremental value, the algorithm will chase the easiest-to-credit conversions. Feeding it calibrated targets, and excluding conversions that tests show aren't incremental, points it at real growth. We discuss this further in AI agents for Google Ads and Meta.

Average vs marginal: the next dollar

Even a correct iROAS is an average across all spend in the test. Returns usually diminish as spend rises: the first $10,000 reaches the most responsive people; the next $10,000 reaches people who are less likely to buy. A channel with an average iROAS of 3 may have a marginal iROAS of 1.5 at current spend.

To estimate marginal returns, run scale-up tests at different spend levels, or test a large increase and a large decrease in different regions. Then allocate budget so the marginal iROAS is roughly equal across channels. That's the core idea in our guide to budget allocation by marginal return.

How to explain this to a CFO

Finance teams tend to like iROAS once they see it, because it matches how they think about investments: what did we get that we wouldn't otherwise have had? A short way to put it:

"Platform ROAS tells us which sales our ads were near. Incremental ROAS tells us which sales our ads caused. We test the big channels a few times a year to measure the difference, and we use that ratio to correct the daily numbers."

Pair it with a simple table of channels, platform ROAS, latest iROAS, test date and confidence interval, and budget conversations get shorter.

Quick checklist

• Know your contribution margin and break-even iROAS before looking at any campaign.
• List your top five channels by spend and note when each was last tested.
• Start with the channel most likely to be over-credited: often branded search or retargeting.
• Run a geo holdout or a platform lift study with a pre-agreed decision rule.
• Turn the result into a calibration factor and update your reporting.
• Re-test on a calendar, and test scale-ups before big budget increases.

Frequently asked questions

Want help measuring what your ads really add? See our incrementality testing service or contact us.