The short version
Incrementality testing measures what a channel actually caused, by comparing a group that saw your marketing with a similar group that didn't. Attribution only shows which channels touched a deal.
It's the most reliable way to answer "if we switched this off, what would we lose?" That makes it the right test before cutting or doubling a big budget line.
B2B firms can run it, with adjustments: test big channels, measure an earlier signal such as demo requests or qualified pipeline, and run long enough to cover the early stages of the sales cycle.
The two most common surprises are branded search and retargeting. Both often look excellent in attribution reports and turn out to be far less incremental.
Every attribution report answers the same question: which channels were present before a deal closed? It doesn't answer the question a founder actually needs answered before moving money: which channels made the deal happen?
A buyer who already knew your name, searched for you on Google and clicked your brand ad would probably have found you anyway. The ad gets the credit. The budget follows the credit. Incrementality testing is how you find out whether that credit was earned.
What is incrementality testing?
Incrementality testing is an experiment that measures the extra results a marketing activity causes. You split an audience or a set of regions into two comparable groups, run the activity for one group only, and compare the outcomes. The difference between the groups is the lift the activity actually produced.
Incremental lift = (Test group result − Control group result) ÷ Control group result
If regions that saw your LinkedIn campaign produced 60 demo requests and comparable regions without it produced 50, the campaign caused 10 extra requests, a 20% lift. If LinkedIn's own report claimed 30, two-thirds of its credit was for buyers who would have come anyway.
The group that doesn't see the activity is called the holdout. It's what makes this a measure of cause rather than coincidence.
Why attribution can't prove causation
Attribution models, from last touch to multi-touch, assign credit to the touches a buyer had. They can't tell you what would have happened without those touches, because nobody saw the version of events where the touch didn't happen.
That gap matters most for channels that reach people who are already on their way to you:
Branded search captures buyers who typed your name. Many would have clicked the free result below the ad.
Retargeting shows ads to people who already visited your site, who were always the most likely to return.
Late-stage email lands just before a decision that was already made.
These channels look brilliant in attribution reports and often turn out to be much less incremental when tested. Meanwhile, the channels that created the interest in the first place, such as content, podcasts and founder posts, look weak because they rarely get the last click. We cover the reporting side of this in common attribution mistakes.
// The Rule
Attribution tells you who was in the room. Incrementality tells you who closed the deal.
Three ways to run an incrementality test
Method | How it works | Best for | Watch out for |
|---|---|---|---|
Platform lift study | The ad platform splits its own audience into people who can see your ads and a control group who can't | Large paid programmes on Google, YouTube or Meta | Not every account is eligible, and results only cover that one platform |
Geo holdout | You switch a channel off in some regions and keep it on in similar ones, then compare | Branded search, paid social, and any channel you can target by location | Regions must be genuinely comparable and big enough to produce results |
Account holdout | You split a target account list in two and only market to one half | Account-based programmes on LinkedIn or outbound | Accounts talk to each other, and lists need to be large |
Google Ads offers Conversion Lift studies based on either users or geography, built to show "the causal impact of ads", though it notes the tool isn't available to all accounts (Google Ads Help). Geo experiments have a longer research history: Google's own researchers described randomly assigning non-overlapping regions to control and treatment groups to measure ad effectiveness (Google Research).
A simple "switch it off for a month and see what happens" test is the weakest version. Seasonality, sales activity and everything else that changed that month gets mixed into the result. If you do run one, compare against the same period last year and keep everything else as steady as possible.
How to make incrementality testing work in B2B
Most incrementality guides assume thousands of conversions a week. A B2B firm with 40 demo requests a month needs a different approach.
Test big channels only. A channel that spends a few hundred dollars a month won't move the numbers enough to measure. Start with the largest or most questioned line in the budget.
Measure an earlier signal. Closed revenue takes too long. Use demo requests, qualified meetings or qualified pipeline created, and check closed revenue later.
Run long enough. Four to eight weeks is common. Shorter tests catch noise, not effects.
Accept bigger questions. You won't detect a 3% lift with B2B volumes. You can tell whether a channel is doing a lot, a little or nothing, and that's usually the decision you need.
Pair it with what buyers say. Self-reported attribution shows where demand started, which a lift test on a capture channel can't.
Read the result with care. A difference between two small groups can easily be chance, so look at whether it holds week by week, and learn the basics of statistical significance before calling a winner.
How to run one, step by step
1. Pick one decision. For example: "Should we keep spending $6,000 a month on branded search?"
2. Choose the metric. Demo requests, qualified meetings or qualified pipeline, measured the same way in both groups.
3. Build two matched groups. Regions or account lists that looked similar on that metric over the previous few months.
4. Agree the rules before you start. Test length, what counts as a meaningful difference and what you'll do with each outcome.
5. Switch the activity off for the holdout only. Keep everything else, including sales outreach, the same in both groups.
6. Don't stop early. Early results swing. Let the test run its full length.
7. Act on it. Cut, keep or scale the channel, then retest once a year or after big changes.
A worked example: branded search
An illustrative B2B services firm spends $6,000 a month on branded search. Google Ads credits it with 30% of all demo requests, so it looks like the best-performing channel the firm has.
The team splits its eight sales regions into two matched groups of four and switches branded search off in one group for six weeks.
Group | Branded search | Demo requests in 6 weeks |
|---|---|---|
Control regions | On | 96 |
Holdout regions | Off | 90 |
Switching branded search off cost about 6 demo requests in six weeks, roughly a 6% drop, where the attribution report implied it should cost close to 30. Most of those buyers found the firm through the organic listing instead.
The firm cut branded search to cover only the terms where competitors bid on its name, saved about $4,000 a month, and moved the money to LinkedIn, the channel buyers kept naming on first calls. Its marketing efficiency ratio improved the following quarter without any drop in demo requests.
Incrementality vs attribution vs media mix modeling
Method | Question it answers | Use it for |
|---|---|---|
Attribution | Which channels touched the deals we won? | Day-to-day optimisation and seeing the buyer journey |
Incrementality testing | What did this one channel actually cause? | Big budget decisions about a specific channel |
How much does each channel contribute over time? | Large budgets with years of spend and revenue data |
Most B2B firms under $20M should run attribution continuously, use incrementality tests a few times a year on their biggest questions, and skip media mix modeling until they have far more data. Our comparison of attribution and media mix modeling explains why.
Common mistakes
Groups that leak into each other. A buyer in a holdout region who sees the ad while travelling, or an account on both lists, blurs the result.
Tests that are too short. One or two weeks mostly measures random variation.
Testing a tiny channel. If the channel produces two demo requests a month, no test will find its effect.
Changing other things mid-test. A new offer, a price change or a sales push in one group ruins the comparison.
Never acting on the answer. A test that doesn't change a budget decision was a research project, not a test.
Want to know which of your channels is actually driving pipeline? In a 30-minute Growth Diagnostic we'll look at your spend and attribution data and pick the one channel worth testing first.
Book a 30-min Growth Diagnostic
Frequently asked questions
What is incrementality testing in marketing?
It is an experiment that compares a group exposed to a marketing activity with a matched group that was not, to measure how many extra results the activity caused. It answers what would be lost if the activity stopped.
How is incrementality testing different from A/B testing?
An A/B test compares two versions of something, such as two landing pages, to see which performs better. An incrementality test compares running an activity with not running it at all, to see whether it adds anything.
How long should an incrementality test run?
For most B2B companies, four to eight weeks, and long enough for each group to produce a meaningful number of demo requests or qualified meetings. Ending a test early because the first week looks promising is a common way to reach the wrong answer.
Can small B2B companies run incrementality tests?
Yes, if they test their largest channels and measure an early signal such as demo requests rather than closed revenue. Small firms can detect whether a channel matters a lot or not at all, even if they cannot measure small effects.
Should a company use incrementality testing or attribution?
Both. Attribution is useful for ongoing optimisation and understanding the buyer journey, while incrementality testing answers occasional, high-stakes questions about whether a specific channel is worth its budget.