Incremental ROAS: The Only Return That Is Actually Yours

The short version

Reported ROAS counts every conversion the platform attributes to you. Some of those would have happened without any advertising.

Incremental ROAS counts only the revenue your advertising caused.

javascript

Incremental ROAS = incremental revenue / spend

It is almost always lower than reported ROAS, and it is the number that decides whether the spend was worth it.

Where the gap comes from

The biggest source is branded and retargeting activity. Someone who was already going to buy searches your name, clicks your ad, and converts. The platform reports a conversion. The revenue was going to arrive regardless.

This is not the platform being dishonest — attribution answers "which touchpoint preceded the sale," not "did this cause the sale." Those are different questions and only one of them is about your money.

The pattern is predictable: the tactics with the best reported ROAS are usually the ones with the largest incrementality gap, because they intercept demand rather than create it.

How it is actually measured

You cannot calculate incrementality from attribution data. It requires a comparison against a world where the ads did not run.

Geo holdout. Turn off spend in matched regions and compare against regions where it ran. The most practical method for most advertisers.

Audience holdout. Withhold ads from a random share of the audience. Cleaner, and supported natively on some platforms.

Timing test. Turn a channel off and back on. Weakest of the three — seasonality and carryover confound it — but sometimes the only option available.

All three cost you something: you deliberately forgo revenue to learn what your advertising is worth. That is the price of the answer.

Reading the result

Incremental ROAS close to reported. Genuine demand creation. Scale with confidence.

Incremental well below reported. You are harvesting demand you already had. Not necessarily wrong — harvesting cheaply is a legitimate tactic — but do not credit it with growth.

Incremental near zero. The spend is buying conversions that would have happened. This is the finding that changes budgets, and it is most often found on branded search and tight retargeting.

Incremental above reported. Possible and often missed. Under-measured channels — upper funnel, video, anything with weak attribution — can drive more than they are credited with.

When to bother

Incrementality testing costs real money and takes weeks. It is worth it when:

A channel is large enough that being wrong is expensive.
Reported ROAS looks too good, especially on branded or retargeting.
You are deciding whether to cut something.
Platform-reported and blended numbers have diverged and you need a tiebreaker.

It is not worth it on small line items or on a channel you are about to test anyway.

Practical rule

Run incrementality on your best-reported channel first, not your worst. The tactics that look most efficient are the ones most likely to be taking credit for demand you already had.