Ultimate Guide of Incrementality Testing in 2026: Why accuracy matters
by
Brian Plant
Your attribution model says Meta drove 40% of revenue last month. But did those sales really come from Meta, or would those customers have bought anyway?
That's the question incrementality testing answers.
Attribution watches what happens (clicks, views, purchases) and hands out credit to the touchpoints along the way. What it can't do is prove cause and effect. Plenty of customers would have bought without ever seeing your ad, and attribution has no way to tell them apart from the ones your ad actually convinced. Incrementality testing can.
If you're deciding where to spend based on attribution alone, you're guessing at which channels actually move the needle. Incrementality testing replaces that guess with an answer: did your marketing cause this sale, or was it going to happen anyway?
What is incrementality?
In marketing, incrementality is the marginal contribution, or the true impact, any specific marketing effort has on driving purchases, leads, or any other important customer action. If a sale is incremental, it means it wouldn't have happened if the marketing did not take place. Incrementality testing answers the question: did the marketing cause the customer to purchase or would they have purchased anyway?
Why attribution alone falls short?
Attribution is the backbone of digital marketing measurement. Meta, Google, TikTok, and analytics tools like Google Analytics all use it to report results. But it has three real limits.
It's missing more of the journey every year.
Tracking every touchpoint for every customer across every channel keeps getting harder. iOS privacy changes, the decline of third-party cookies, and walled gardens all leave gaps in your view of the customer journey.
It shows correlation, not causation.
This is the bigger problem. A customer who clicked your Meta ad before buying didn't necessarily buy because of the ad. All you know for sure is that the click came first.
Each platform only sees itself.
Meta's attribution doesn't know about your Google campaigns, and Google's doesn't know about Meta. Each one reports its own slice, so you end up with a set of silos instead of one complete picture.
That's where incrementality testing comes in. It measures whether a marketing action actually changed what customers did. Understanding what incremental revenue is, and how it differs from attributed revenue, is the first step. Think of incrementality as the audit layer that keeps your attribution honest.
How incrementality testing works
Incrementality testing works like any controlled experiment. You split an audience into two matched groups, show your marketing to one, hold it back from the other, and measure the difference.
The test group sees your ads. The control group doesn't, because ad spend in their regions is paused for the length of the test.
After the test runs (usually 21 to 28 days), you compare purchases, revenue, or whichever metric matters most between the two groups. The gap between them is your incremental impact: the sales your marketing actually created.
There are two ways to split the audience. Random assignment at the individual level is the textbook approach, but it's very hard to pull off across channels. You'd need to identify every user on every channel to split and track them, which most brands can't do and most platforms won't allow.
Geo-based matching is how it's done in practice. You find geographic markets (DMAs, zip codes, or cities) with similar purchase history and pair them up, one test market and one control market per pair. Running the test across dozens of matched pairs makes it both statistically rigorous and practical to run.
Setting up an incrementality test
Here's how a test comes together, step by step.
Step 1: Pick what channel or tactic you want to test
Pick a channel (Meta, Google, TikTok) or a tactic (Performance Max, Reels-only spend). The narrower you go, the bigger the sample you need. Testing a single ad usually isn't practical. Testing a whole channel is.
Step 2: Define control and test geo pairs
Look at about 180 days of purchase data by geography. Find markets that have performed similarly over that period, then pair them up. This step takes data science know-how, because you need a statistical power analysis to confirm the sample is big enough to trust the result.
Step 3: Define additional test details
Decide how long the test runs (usually 21 to 28 days), how much to hold out, and how long the cooldown lasts (typically 7 days after the test ends, to catch late conversions). Holdouts can be smaller than many marketers expect. With the right design, a pause-to-measure test can hold out as little as 5% of orders. A launch-to-measure test avoids a full holdout entirely by putting a fraction of budget into test markets and measuring what it adds. A data scientist sets these details based on your order volume and how much your sales vary.
Step 4: Launch the test and measure
Set up the test in your ad platform. With automatic setup on Meta, Google, and TikTok, geo exclusions are applied for you and removed when the test ends. On other channels, you pause spend in the control markets manually.
Once the test wraps, you measure the gap.
Keep testing
For brands with enough scale, a regular testing rhythm matters. Plan to retest each major channel about every 3 months. Incrementality changes over time as audience behavior, competition, and creative performance shift, so a Q2 result won't tell you what Q4 looks like.
Make sure you have the volume
Geo testing needs enough orders to produce a reliable read. As a rough guide, brands doing fewer than 3,000 orders a month usually don't have the volume for matched geo pairs to deliver a statistically meaningful result.
What channels should you test for incrementality?
Not every channel is a good fit. Channel incrementality testing works best when two things are true: there's enough volume, and you can control spend by geography.
Good candidates have high volume and geo-targeted spend: Meta, Google, TikTok, CTV, and Snap. These give you the scale and control a test needs.
Poor candidates are low-volume channels or channels you can't target by geography, like affiliate and podcasts. There's no reliable way to hold out an audience on them.
Most brands start with their biggest channels: Google and Meta, and often TikTok. That's where the most budget sits, and where attribution tends to be furthest off in either direction.
Testing incremental attribution on Meta often reveals a big gap from what Meta reports on its own, and other channels show similar gaps. Testing your top two or three channels gives you a clear incrementality map of your core media mix.
Incrementality testing is also one of the best ways to measure the true ROI of CTV. Once you measure causally, brand-focused and conversion-focused CTV spend often show very different returns.
How do you measure incrementality?
Three numbers matter most: incremental lift, incrementality, and incremental ROAS. The first two examples use the same test: 140 purchases in test markets and 100 in control markets.
How do you calculate incremental lift?
Incremental lift shows how much your marketing increased purchases, as a percentage.
((Test Purchases − Control Purchases) ÷ Control Purchases) × 100 = Incremental Lift
(140 − 100) ÷ 100 × 100 = 40% lift
Lift tells you how big the effect was compared with not advertising. It doesn't tell you how much of the test group's sales your marketing was responsible for. That's the next metric.
How do you calculate incrementality?
Incrementality shows what share of the test group's purchases your marketing actually caused.
((Test Purchases − Control Purchases) ÷ Test Purchases) × 100 = Incrementality
(140 − 100) ÷ 140 × 100 = 28.6% incrementality
Put simply, about 29 of every 100 purchases in test markets happened because of your marketing. The other 71 would have happened anyway, just as they did in the control markets.
How do you calculate incremental return on ad spend (ROAS)?
Incremental ROAS divides incremental revenue by ad spend.
Incremental Revenue ÷ Ad Spend = Incremental ROAS
Spend $100,000 in test markets, generate $150,000 in incremental revenue, and your incremental ROAS is 1.5x. That's your true return on the channel, not the inflated number attribution reports.
These are simplified versions of the math. WorkMagic's methodology goes further, with confidence intervals, regression adjustments, and halo effect measurement, but the core logic is the same. For a deeper walkthrough, see how to calculate incremental lift.
How do you divide control and exposed groups for incrementality tests?
There are two ways to divide audiences:
Randomly assigned audiences - This is the standard in scientific controlled trials, but is not really practical for advertising. The ability to split the testing population at a user level into two randomly assigned audiences is not available for most channels and is often impossible to truly execute as complete user level identity resolution is needed to accurately measure incrementality using this method.
By geography (Geo-based Match Market Incrementality Test) - For geo-based match market incrementality tests, also known as geo-lift testing, individual geos (defined at DMA, ZIP code, or other level) are divided into two comparable groups based on the historic performance of each geo. This approach is the best to use as it's both statistically rigorous and feasible to execute.
Why brands hold off on incrementality testing
The most common worry is losing revenue in the holdout.
In practice, the cost is smaller than most people expect. A well-designed pause-to-measure test can hold out as little as 5% of orders in a single channel, and a launch-to-measure test doesn't pause any existing spend at all. And because you also save the ad spend you pause, the hit to profit is usually smaller than the hit to revenue.
There may be a short-term dip. Over the long run, it's small next to the payoff of optimizing to real results instead of inflated attribution.
The real barriers are different:
Scale. Matched geo pairs need enough orders to produce a reliable result, roughly 3,000 a month as a starting point. Many DTC brands aren't there yet.
Results go stale. One test can't predict the future. Channel performance shifts with seasonality, competitor spend, creative fatigue, and audience saturation. Testing quarterly keeps your incrementality map current.
Granularity. Tests work at the aggregate level. You learn that a channel drove lift, but not why or where. You need a way to translate those results into campaign-level decisions.
Resources. Running incrementality tests in-house takes data scientists and engineers working closely with marketing. For most brands, building that capability internally doesn't pay off.
Why incrementality testing matters right now
Attribution is getting less reliable. iOS privacy changes, the decline of third-party cookies, and walled gardens all chip away at what it can see. Brands that rely on attribution alone drift further from the truth every quarter.
Incrementality testing brings that signal back. It measures what attribution can't see, and it lets you check your measurement against what actually caused sales.
Geo incrementality testing has become table stakes for serious measurement. For brands with enough volume, it should be a foundation, not a one-time audit. Make it a quarterly habit.
Conclusion
Measuring for incrementality has never been more important, or easier to do. Brands that optimize for incrementality measure for causation, restore the signal strength of their data, and overcome the natural limitations of standard multi-touch attribution. For brands with sufficient volume, incrementality testing should be a foundational component to their measurement solution.
If you want to get started with testing and optimizing for incrementality, schedule time with a WorkMagic Measurement Expert to learn more about how your business can leverage incrementality testing to make better decisions.