TL;DR
Platform ROAS answers "what did the dashboard attribute to the ads." Incrementality answers "what would have happened without the ads." Those are different questions with different answers, and the gap between them is where budgets get wasted. You don't need a statistics degree: start with holdout tests and conversion lift studies, sanity-check with blended numbers, and stop treating attributed ROAS as profit. The advertisers who understand incrementality spend with confidence. Everyone else spends on faith.
Correlation is not causation, the ad-spend edition
Your dashboard reports that the ads generated $400,000 on $100,000 of spend. 4x ROAS. Champagne.
But here's the uncomfortable question: how many of those buyers would have bought anyway? The ones who searched your brand name after seeing the ad seventeen times. The existing customers who got retargeted into a purchase they were already making. The shoppers who were going to buy on Friday and happened to click on Thursday.
Every one of those sales got attributed to your ads. Not all of them were caused by your ads. The difference between "attributed" and "caused" is incrementality, and it's the most important number in your account that isn't in your dashboard.
This isn't a rounding error. For retargeting-heavy accounts, the incremental share of reported conversions can be a small fraction of what's attributed. You're not getting 4x. You're getting 4x minus everyone who would have converted without you, and "everyone who would have converted without you" is doing a lot of quiet work in that sentence.
What incrementality actually means
Strip away the jargon: incrementality is the lift your ads cause above the baseline of what would have happened anyway.
Imagine two identical worlds. In one, your ads run. In the other, they don't. The difference in sales between the worlds is your incremental impact. Everything else is baseline: brand demand, organic traffic, seasonality, the customers who were already yours.
Platform attribution can't see the second world. It only sees the first one and assigns credit generously, mostly to itself. That's not a conspiracy, it's just how last-click-flavored attribution works: whoever touched the buyer last gets the trophy, whether they did the work or not.
How to measure it (in ascending order of effort)
1. The blended sanity check (free, do it today). Look at total business revenue against total ad spend over time. When you scaled Meta spend 50% last quarter, did total revenue move proportionally, or did the dashboard ROAS stay pretty while the bank account shrugged? If attributed ROAS climbs while blended ROAS flatlines, you're buying attribution, not growth. This check takes ten minutes and ends more arguments than any fancy methodology.
2. Geo holdouts (cheap, slightly annoying). Split your markets: run ads in some regions, hold out others. Compare sales. The holdout regions are your second world. This works best for businesses with multiple comparable geos and enough volume per region. It's not perfect (regions differ, spillover happens), but it's real evidence, and it's dramatically cheaper than ignorance.
3. Conversion lift studies (Meta's own tool). Meta offers lift testing that splits your audience into test and control groups and measures the difference. It's the cleanest read available inside the platform. Use it on your biggest budget lines first: prospecting, retargeting, Advantage+ shopping. The results will humble you, which is the point.
4. Time-based pausing (free, terrifying). Pause a campaign for two weeks and watch what happens to total sales. If revenue barely dips, that campaign's incremental contribution was barely there. Nobody wants to do this test because everyone suspects the answer. Do it anyway, on one campaign, once. The knowledge is worth the anxiety.
5. Marketing mix modeling (expensive, enterprise). Statistical modeling of all your marketing inputs against sales. Powerful, slow, and overkill for most businesses under eight figures. Mentioned here so you know it exists, not because you need it this quarter.
What the results usually show
We won't insult you with fake benchmarks, but the pattern is consistent enough to describe: prospecting usually shows real incrementality (you're reaching people who didn't know you). Retargeting usually shows much less than its dashboard ROAS suggests (you're paying to nudge people who were already coming). Brand search-style campaigns on Meta show the least (you're taxing your own demand). Broad Advantage+ campaigns land somewhere in the middle, and the only way to know where is to test.
The typical reaction to a first lift test is grief, then clarity. Grief because the numbers were prettier before. Clarity because now you know which half of the budget is working, which is the thing everyone quotes and nobody measures.
What to do with the answer
Reallocate, don't just cut. If retargeting is 20% incremental and prospecting is 60% incremental, the move isn't to kill retargeting. It's to stop judging retargeting on the same ROAS target as prospecting and fund each according to its real contribution. Different jobs, different targets.
Set targets on blended ROAS. Platform ROAS is a diagnostic; blended ROAS is the scoreboard. When the business grows, the ads worked. When only the dashboard grows, something is wrong. Make blended the number in the Monday meeting and watch how fast the conversation gets honest.
Test the big decisions, not everything. You don't need a lift study for every ad set. Test the structural questions: does prospecting drive new customers, is retargeting incremental, does the Advantage+ campaign actually find new demand. Three good tests a year beat thirty dashboard debates.
Keep attribution for optimization, not for truth. Attributed ROAS is still useful for comparing creatives and audiences within the account. It just isn't a measure of business impact. Use the dashboard to steer the car; use incrementality to check you're on the right road.
FAQ
What's the difference between attribution and incrementality? Attribution assigns credit for conversions that happened. Incrementality measures conversions that wouldn't have happened without the ads. Attribution is about dividing credit; incrementality is about proving cause.
Do I need a conversion lift study to understand incrementality? No. Start with the blended sanity check: compare total revenue trends to ad spend trends. Geo holdouts and campaign pauses give real signal without any special tooling.
Why is retargeting ROAS usually less incremental than it looks? Retargeting targets people already close to buying: past visitors, cart abandoners, engagers. Many would convert without the ad. The ad gets last-touch credit for sales it mostly witnessed.
How often should I run incrementality tests? Test structural questions a few times a year, or whenever you make a big budget shift. It's a compass, not a speedometer: check it when choosing direction, not every minute.
Does incrementality mean I should cut my ad spend? Usually the opposite of what people fear. It typically means reallocating: less spend on campaigns that harvest existing demand, more on campaigns that create new demand. Total spend often stays similar; the mix gets smarter.
Written by the MetaMaxd team. We run Meta ads for brands that are tired of agency theater: no vanity metrics, no "hacks," just accounts engineered to make money. If your dashboard and your bank account are telling different stories, that's what our audits are for.
