What Is Incrementality? Measuring the Sales Your Marketing Actually Causes

Your best-performing campaign might be doing almost nothing. I say that to clients more often than they’d like, usually about the retargeting campaign with the gorgeous ROAS or the branded search campaign that “never misses”. Those campaigns look brilliant because they stand directly in front of people who were already on their way to buy. They collect the credit. Whether they caused the sale is a different question, and most ad reports never ask it.

Incrementality is the discipline of asking it. It measures the part of your results that exists only because of your marketing, and it regularly disagrees with the attribution numbers people use to set budgets. If you’ve ever cut a campaign and watched sales barely move, you’ve already run an accidental incrementality test. This guide shows you how to run one on purpose.

What incrementality actually means

In plain terms, incrementality is the difference your marketing makes. It’s the sales that happen with the marketing, minus the sales that would have happened without it.

The formal term is incremental lift: the increase in a desired outcome caused by exposure to marketing, measured against a comparable group that wasn’t exposed. That comparable group is called the control or holdout group, and it’s the heart of the whole idea. Without it, you have no way to know what “would have happened anyway” looks like.

This is different from attribution, which is what most dashboards show. Attribution takes a sale that happened and divides credit among the touchpoints the customer passed through. It never asks whether the sale depended on those touchpoints. A customer who searched your brand name, clicked your branded ad, and bought would have found your website through the organic listing sitting right underneath. Attribution gives the ad full credit anyway.

Incrementality is the correction. It’s the closest thing marketing has to a controlled experiment, and that’s why it has become central to how serious advertisers judge channels as tracking has become harder.

Why incrementality matters more now

For years, many advertisers could get away with trusting attribution because tracking was fairly complete. Cookies followed people across sites, ad platforms saw most conversions, and the gaps felt small.

That has changed. Browser privacy protections, consent requirements and app tracking limits mean platforms now see a smaller share of the journey and fill the rest with modelled estimates. Each platform also still wants credit, so the sum of what Google, Meta and your email tool claim can add up to more sales than you actually made.

When the data underneath attribution gets weaker, a controlled comparison becomes more valuable, not less. A holdout test doesn’t need to track every click. It only needs to count outcomes in two groups. That makes incrementality one of the few ways to judge a channel that doesn’t depend on the tracking that’s breaking.

How incrementality is measured

Every method follows the same logic: create two comparable groups, expose only one of them to the marketing, and compare the results.

Audience holdouts. You take a target audience and randomly withhold a portion of it from the campaign. Because the split is random, the two groups should behave alike except for the ads. Both Google and Meta offer conversion lift studies that run this kind of split inside their platforms.

Geo experiments. You switch marketing on in some regions and off, or down, in comparable ones, then compare sales by region. This works when you can’t split audiences cleanly, and it captures effects that platform-level tests miss, like people who saw an ad and later bought in a shop.

Pause tests. You switch a channel off for a period and watch what happens to total sales. They’re simple but noisy, because seasonality and everything else you’re doing also change during the pause. Use them to generate questions, not final answers.

Whatever the method, the arithmetic is the same:

Incremental lift = (Test conversion rate − Control conversion rate) ÷ Control conversion rate

And the number that should replace platform ROAS in your budget decisions:

Incremental ROAS = Incremental revenue ÷ Ad spend

A worked example: the retargeting holdout

Say you run a D2C apparel brand spending ₹3,00,000 a month on retargeting. This is a hypothetical example with example numbers, but it follows the pattern I see most often when retargeting gets tested.

Your ad platform reports ₹24,00,000 in revenue from retargeting, an 8x ROAS. It’s your best campaign on paper, so the plan is to double it.

Instead, you hold out 20% of your retargeting audience for four weeks. Both groups are people who visited your site recently. The test group sees your ads. The control group doesn’t.

At the end of the test, the test group converts at 5.0% and the control group converts at 4.0%. So the lift is (5.0 − 4.0) ÷ 4.0 = 25%. Most of the people in your retargeting audience were going to buy anyway. The ads nudged some extra conversions, but only about a fifth of the conversions the platform credited to retargeting were genuinely caused by it.

That puts incremental revenue at roughly ₹4,80,000 against ₹3,00,000 of spend, an incremental ROAS of about 1.6x. Whether that’s profitable depends on your margin. It’s certainly not the 8x campaign that deserved double the budget.

The right response isn’t to kill retargeting. It’s to spend on it only up to the point where its incremental ROAS still clears break-even, and move the rest of the budget to campaigns that find people who weren’t already coming.

Reading a lift result honestly

A test result is only useful if you know how sure it is. Two groups will never convert at exactly the same rate, even if your ads do nothing, simply because of random variation. The question is whether the gap you see is bigger than that random noise.

This is what statistical significance and confidence intervals are for. A good lift study reports a range, not only a single number. For example, it might say lift sits somewhere between 10% and 40%. If that range includes zero, the honest conclusion is that you can’t tell whether the campaign worked. That’s a legitimate result. It usually means the test needs more time or a bigger holdout.

Two habits keep you honest. Decide the test length and the size of effect you care about before you start, so you aren’t tempted to stop the moment the numbers look good. And judge the business decision on the low end of the range, not the high end. If the campaign is still worth funding at the pessimistic end, fund it.

Testing brand and upper-funnel campaigns

Incrementality isn’t only a tool for exposing retargeting. It’s also the best way to defend campaigns that attribution treats unfairly.

Brand video, sponsorships and broad prospecting campaigns rarely get the last click. Attribution reports make them look expensive, so they’re the first to be cut when budgets tighten. A geo test can show what they’re really doing: regions where the brand campaign ran should see more branded searches, more direct visits and more sales, including sales that attribution credited to other channels.

The effects are slower and smaller per rupee than retargeting’s apparent returns, so these tests need to run longer. But they’re often where the most genuinely incremental growth lives, because they reach people who weren’t already on their way.

Incrementality vs attribution

	Incrementality	Attribution
Question it answers	Would this have happened without the marketing?	Which touchpoints were involved in this sale?
Method	Controlled comparison against a holdout group	Rules or models that assign credit to touchpoints
Needs a control group	Yes	No
Tends to favour	Channels that reach new, undecided people	Channels closest to the purchase, like retargeting and branded search
Speed	Slow: weeks per test	Fast: continuous, in the dashboard
Best used for	Deciding how much a channel really deserves	Day-to-day optimisation within a channel

These aren’t rivals. Attribution is fast and granular, so it’s how you optimise ads and keywords day to day. Incrementality is slow and expensive, so it’s how you check whether attribution is telling the truth about a whole channel. Use incrementality to calibrate how much you trust each channel’s attribution, then keep using attribution for daily decisions with that calibration in mind.

How to run your first incrementality test

1. Pick the channel you trust least. Retargeting, branded search and campaigns aimed at existing customers are the usual suspects, because they sit closest to purchases that were already happening.
2. Write down the decision the test will inform. “Should we double retargeting?” is a good test question. “Let’s see what happens” isn’t.
3. Choose the method. Use the platform’s lift study if one is available for that channel. Use a geo split when audiences can’t be separated cleanly or when offline sales matter.
4. Size the holdout and duration. The control group needs enough conversions for a real difference to stand out from random variation. Smaller accounts need longer tests, not smaller conclusions.
5. Freeze everything else. Don’t change creative, budgets or offers in only one group mid-test. Anything that affects one group and not the other breaks the comparison.
6. Calculate lift and incremental ROAS. Compare incremental ROAS with your break-even ROAS, not with the platform’s reported figure.
7. Act, then retest. Rebalance budget based on the result, and retest the same channel every few months, because incrementality changes as your audience and competition change.

Common incrementality mistakes

Trusting a pause test as proof. Switching a channel off for two weeks tells you something, but seasonality, promotions and other channels also move in those two weeks. Treat pause tests as a reason to run a proper holdout, not as the answer.

Stopping tests too early. It happens because early results look decisive and everyone wants an answer. Small samples swing wildly. Decide the test length before you start and don’t peek your way into a conclusion.

Contaminating the control group. If the holdout group sees your ads through another campaign, or gets the same email offer, you’re no longer measuring the difference the ads make. Keep the groups genuinely separate for the length of the test.

Testing only the channels you already doubt. It’s natural to test the suspect channels, but prospecting and brand campaigns also need testing. They’re often undervalued by attribution, and incrementality is how you prove they deserve their budget.

Treating one result as permanent. Incrementality depends on audience, creative, competition and season. A channel that was barely incremental last year can be valuable now, and the reverse. Build testing into your calendar.

How to tell if your measurement is healthy

There’s no universal incrementality benchmark worth quoting. Lift depends entirely on the channel, the audience and how much demand already exists. These signals tell you whether you’re measuring well:

• Every major channel has been through at least one controlled test, and the results are written down with dates.
• Budget decisions for your largest channels reference incremental ROAS, not only platform ROAS.
• Your retargeting and branded search budgets have a ceiling set by test results, not by how good their dashboards look.
• Prospecting campaigns that attribution undervalues have been tested, and funded according to what the tests showed.
• Tests are repeated on a schedule, because you expect the answers to change.

If most of that is true, you’re making budget decisions on evidence rather than on whichever channel stood closest to the checkout.

Key takeaways

• Incrementality is the part of your results your marketing genuinely caused, measured against a group that didn’t see it.
• Attribution tells you who was involved in a sale. Incrementality tells you whether the sale depended on them.
• Retargeting and branded search usually look far better in attribution than in incrementality tests.
• Judge channels on incremental ROAS against break-even, not on platform-reported ROAS.
• Test regularly. Incrementality changes as your audience, creative and competition change.