Attribution tells you how to divide credit. Incrementality testing tells you whether the credit was real to begin with.
Those are different exercises, and I think the second one is becoming more important than the first.
Every ad platform reports its own conversions. Most of them inflate. Meta is built to take credit for any conversion it touched, Google does the same with branded search, and platforms like YouTube get unfairly overlooked in last-click reporting.
Incrementality testing is how you cut through that. It’s a truth-seeking exercise that asks one question: if I turn this channel off, what actually changes?
What Incrementality Actually Means
Incrementality measures the additional revenue or conversions a channel drives that wouldn’t have happened otherwise.
Attribution apportions credit across all the conversions that did happen. Incrementality asks how many of those conversions only happened because of a specific channel.
The two work together. I use incrementality tests to build conviction in an attribution model. If I run a test on Meta and the lift matches what HubSpot and Funnel are reporting, then I trust the attribution. If they disagree, I trust the test.
Why Incrementality Testing Is Having a Moment
Marketers have always known that platform numbers aren’t fully trustworthy. What changed is that the gap got too big to ignore.
iOS14 broke pixel-based tracking. ChatGPT and other LLMs are starting to drive purchases that bypass web analytics entirely. Modern marketing mix modelling is going mainstream as last-click models look less and less useful.
Add in ChatGPT’s Instant Checkout, where users buy without ever visiting a website, and the case for measuring incrementality directly gets stronger every quarter.
The Three Types of Incrementality Tests
There are three common approaches. They each answer a slightly different version of the same question.
Geo holdouts
The most common, and usually the most practical. You run ads in some geographies and not in others, then compare sales between them.
The simplest version: turn off Meta in California for two months, then compare California sales to the rest of the country. If California declines relative to the baseline, the drop is your incremental lift.
You can also run it the other direction. If you aren’t running ads anywhere in Canada, turn them on in BC only and compare lift against the other provinces.
Geo holdouts are simple to explain and almost any team can run one. They’re my default recommendation for mid-market brands.
Ghost ads
Ghost ads serve a placeholder to one group of users while the test group sees the real ad. You compare conversion rates between the two groups.
The methodology is solid but harder to set up. Some platforms offer it as a native feature, like Meta’s Conversion Lift studies.
Conversion lift studies
Conversion lift studies are platform-run tests where the platform itself holds out a portion of your audience. Meta and Google both offer this.
The catch is you’re trusting the platform to measure itself, which is exactly the conflict of interest incrementality testing is meant to solve. They’re still useful, but I treat their results as one input rather than the answer.
The Channels That Get Undervalued by Last-Click Reporting
What used to surprise me, but no longer does, is how much value sits in channels that don’t look like direct response.
Meta and Google Search ads are built to drive immediate action. A click or a conversion, fast. That’s what they’re best at, and they take credit accordingly.
YouTube isn’t that. Neither is Spotify. They build awareness, which leads to demand, which eventually leads to a conversion on a different platform.
For years, growth marketers, including me, dismissed YouTube because they couldn’t measure it directly. The reality is that incremental tests routinely show YouTube driving meaningful top-line revenue.
Side by side with display ads in a platform conversion report, YouTube and display look the same: both report almost no conversions. But run an incrementality test and they look completely different. YouTube drives real lift. Display rarely does.
The biggest payoff from incrementality testing has been at the two ends of the funnel: top-of-funnel awareness channels that look terrible on last click, and bottom-of-funnel channels that look great but aren’t actually adding revenue.
The Channels That Fail Incrementality Tests Most Often
Two patterns come up over and over.
Branded search
Google branded search, where you bid on your own domain or brand name, reports incredible ROAS. The conversion rate is high, the cost per click is low, and the platform tells you it’s working.
Run an incrementality test, and most of those conversions would have happened anyway. The customer already knew your brand. They were going to type it into Google whether or not you bid on the keyword.
Branded search isn’t always wasteful. Defending against competitor bids on your terms can be worth it. But it’s one of the first places I look when a team wants to free up budget.
Meta retargeting
Retargeting ads on Meta show the same pattern. The audience is people who already visited your site and were already considering a purchase. The platform happily takes credit for the conversion.
An incrementality test usually shows a lot of that revenue was going to happen anyway.
How to Explain Incrementality Results to a CMO or CFO
When I present incrementality results to executives, I bring it back to the income statement.
Revenue is at the top. Advertising is a line in overhead somewhere below it. The ratio between those two numbers is what we’re trying to improve.
Incrementality testing is how I make sure we’re improving that ratio with the same cost. Same advertising spend, more revenue at the top.
Executives understand income statements. They’re skeptical of ROAS because they have been burned by inflated platform numbers before. Framing the work around the financial outcome they actually care about gets buy-in much faster than walking them through statistical methodology.
Where to Start
If you have never run an incrementality test, start with a geo holdout on a channel you’re uncertain about.
If you have a YouTube budget that you can’t defend in last-click reporting, test it. If you have branded search spend you suspect is non-incremental, test that.
The goal isn’t perfect measurement. The goal is to build enough conviction to make better spend decisions than you would have made using platform reports alone.
Frequently Asked Questions
What is incrementality in marketing?
Incrementality is the additional revenue or conversions a channel drives that wouldn’t have happened otherwise. It’s different from attribution, which divides credit across the conversions that did happen.
How do you measure incrementality?
The most practical method is a geo holdout. Turn a channel off in one region, keep it running everywhere else, and compare sales between the two over a defined test window. The difference is your incremental lift.
What is the smallest budget where incrementality testing makes sense?
Geo holdouts work for most mid-market brands, even with modest budgets, as long as you have enough sales volume in each region to see a clear signal. If you’re spending less than $20K per month per channel, the noise tends to drown out the signal.
What channels usually fail incrementality tests?
Branded search and Meta retargeting are the most common. Both report strong ROAS, but the conversions they claim credit for are often the ones that would have happened anyway.
