Incrementality vs attribution comes down to the question. Attribution answers "who gets credit for this sale?". Incrementality answers "what would have happened without this spend?". If the decision is daily steering inside a channel, attribution is enough. If the decision moves budget, ask for incrementality.
Both are useful. The trouble starts when one is used to answer the other's question. That happens in a lot of budget meetings.
Attribution is a rule for splitting credit for a sale between the ads and touchpoints that came before it. An attribution model is the specific rule: last click, data-driven and so on. Google Ads now offers last-click and data-driven attribution, and has retired its older rule-based models1.
Last-click attribution gives all the credit to the final click before purchase. It shows who was standing nearest the till. Whether they made the sale is a different question.
Attribution does not show that an ad caused a sale. It shows that the ad was there.
Incrementality is the effect that would not exist without the activity. A vendor that sells incrementality tests frames the difference as correlation versus causation2. Another vendor's FAQ defines incrementality as the extra result that would not have happened without the marketing3. Both sell measurement. The definitions are still the plain ones.
You estimate incrementality by comparing what happened with a baseline: what would have happened anyway. Usually that means an experiment, such as a holdout test or a geo test.
Neither column is the winner. They do different jobs.
Often not. In a study of advertising lift experiments at Facebook, researchers found that observational methods often failed to reproduce the experimental results4. Attribution is observational too. It sees who clicked. It cannot see who would have bought anyway.
A measurement vendor suggests using platform ROAS to steer inside a platform and incremental ROAS to compare investments and set budgets5. It sells incrementality measurement, so read that as a view. The same explainer says the gap can run either way, depending on the channel, so there is no fixed correction to apply.
Incremental ROAS (iROAS) counts only the revenue the ads caused. Platform ROAS counts the revenue they were credited with. For a worked comparison, see platform reported ROAS vs incremental ROAS.
Label the number every time: which platform, which model, which attribution window.
The budget is large enough that steering and proof both matter. Attribution runs the daily work. A test checks, from time to time, whether the attributed numbers hold up. Where they disagree, believe the test for that question and adjust how you read attribution.
A CFO is not asking who gets credit. A CFO is asking whether the money came back. That is an incrementality question, judged against contribution margin: what is left per order after product, delivery, payment, returns and discounts.
So for board-level decisions, attribution is context, not evidence. That is not a criticism of attribution. It was never built for that job.
Attribution answers "who gets credit?". Incrementality answers "what would have happened without it?". Both are useful. Just don't use the first to answer the second.
In our view, most brands don't need a test every week. They need attribution labelled honestly, a total-revenue trend that keeps everyone grounded, and a test when a big budget call is coming. We compare all three main methods in MMM vs MTA vs incrementality.
For ZFD: We use attribution to steer and incrementality to decide. The figure that goes to the board is the one checked against your sales and margin.
No. It answers a different question. Attribution is better for fast steering inside a channel. Incrementality is better for deciding whether spend causes sales.
It shows which ad was closest to the sale. Research comparing observational methods with experiments found they often give different answers4.
No. Data-driven attribution is a smarter way of sharing credit across tracked touchpoints. It still does not compare against what would have happened without the ads.
Start smaller. A before-and-after comparison, a trend in total revenue against total marketing spend, or a post-purchase survey can help. Each has a known bias, so name it before you rely on the result.
“Their engagement is impressive.”
See how far your platform numbers could be from reality. Try the Is your ROAS real? calculator.
