Incrementality

In advertising the question is always the same: would this conversion have happened if the ad had not run? A conversion that would not have happened is incremental. One that would have happened anyway is not, even if the person clicked or saw the ad on the way, and even if every report credits the ad with it.

• Attribution asks: which touchpoints led here?It starts from the conversions that happened and shares the credit among the clicks and visits before them, by a rule such as last click or position based. See attribution model.
• Incrementality asks: would it have happened anyway?It needs a comparison: a group that got the marketing and a group that did not. Without that second group there is no incrementality number, only a credited one.

The gap between the two is largest where the ad reaches people already on their way: retargeting, brand search and view-through credit. A retargeting campaign can look like the best in the account on attribution and add little in a test, because the people it reaches were coming back. Incrementality vs attribution works through that case and when each answer is the one to use.

Incremental lift is how much higher the conversion rate is in the group that got the marketing than in the group that did not, as a share of the second rate.

The example in full. An advertiser splits 250,000 people at random. 200,000 are the test group and can be shown the ads; 50,000 are held out and are not. Over four weeks the test group produces 1,000 conversions, a rate of 0.50%. The holdout produces 200, a rate of 0.40%. The holdout's 0.40% is what the test group would have done without the ads, so the ads lifted the rate by 0.10 points, which is 25% on top of the baseline.

Lift is a ratio. The number a budget owner needs is the count behind it: how many conversions the ads produced that would not otherwise exist.

So 800 of the 1,000 conversions would have happened without the ads. If the platform attributed 600 conversions to the campaign, it credited the ads with three times what they caused. That is not the platform lying: its attribution counts conversions near an ad, and a test counts conversions because of one.

Incremental CPA is the true price of an extra conversion. It is always at least as high as the attributed cost per acquisition, and the distance between the two is how much of the attributed figure was conversions you would have had anyway.

Return on ad spend divides the revenue attributed to the ads by what they cost. Incremental ROAS divides only the revenue the ads caused.

In the same test, the test group closed €180,000 of revenue. The holdout closed €30,000, which scaled to the test group's size (four times as many people) is €120,000: what the test group would have closed without the ads. Incremental revenue is the difference, €60,000. The platform attributed €150,000 to the campaign, a ROAS of 5.0 on €30,000 of spend; the iROAS is 2.0.

An iROAS of 2.0 means each euro returned two in revenue the business would not otherwise have had. Whether that pays depends on margin: on revenue with a 40% gross margin it returns 0.80 of profit per euro and loses money, however good the 5.0 looked.

A holdout group, also called a control group, is the part of the audience deliberately kept from the marketing so its results show the baseline. Everything an incrementality number means rests on the holdout being comparable to the test group.

• Random assignment. People or regions are assigned by chance, not by choice, so the two groups differ only in whether they got the ads. A holdout made of the people who happened not to be reached is not a holdout: they were not reached for a reason.
• Held out for the whole test. A person who moves between groups, or a region that sees the campaign through a neighbouring market's ads, blurs the comparison. This is called contamination.
• Big enough to read. The holdout costs reach, so advertisers keep it small. Google Ads lets a Conversion Lift study hold back from 1% to 50% of users; the smaller the holdout, the longer the test needs to run before the difference means anything.

Every method builds a group that did not get the marketing. They differ in what is held out: people, places or time.

• User-level holdout (conversion lift)The platform randomly splits the people it could reach and withholds the ads from one part. The cleanest design, run inside one platform and read on the conversions that platform can see.
• Geo lift (matched markets)Regions are the unit: the ads run in some, pause in comparable others, and the outcomes per region are compared. Works for any channel and any outcome you can count by region.
• On/off testA channel is paused for a period and the results compared with before and after. Cheap, and the weakest design, because everything else that changed in that period is in the result too.
• Ghost ads and PSA testsVariants of the user holdout: the control group sees an unrelated public service ad, or the platform records where your ad would have shown without showing it, so both groups are comparable at the moment of exposure.

On the platforms, as read on their own documentation on 7 October 2026: Google Ads runs Conversion Lift based on users or on geography, with a minimum campaign budget of $5,000 and at least 1,000 observed conversions, and not every account has access. Meta's Conversion Lift splits Accounts Center accounts at random into a test group that sees the ads and a control group that does not; Meta says access is currently limited and goes through a Meta representative. Meta also publishes GeoLift, an open source package for geo experiments built on synthetic control methods.

Marketing mix modelling estimates incrementality from history instead, by relating spend to outcomes over many weeks; MMM vs MTA sets it against attribution, and incrementality testing walks through running a test step by step.

Two random groups never convert at exactly the same rate, even with no ads at all. Significance asks whether the gap you measured is bigger than the gap chance alone would produce. A test read at 95% confidence means a gap that size would turn up by chance less than one time in twenty if the ads did nothing.

Same rates, same lift, opposite verdicts. The small test cannot tell a 25% lift from chance, so it proves nothing either way: not that the ads work, and not that they do not. That is why platforms set minimum conversion volumes, and why lead generation accounts, which count tens of deals rather than thousands of purchases, often read a test on qualified leads first. The incrementality calculator runs the test for your own numbers.

A lift test is only as good as the outcome it counts. A platform's lift study counts the conversions the platform sees, and on a lead generation account that is the form fill. A campaign can lift form fills by 25% and lift deals by nothing, if the extra leads are the ones sales disqualifies.

• The outcome lives in the CRM. Whether a lead was qualified, became a deal and closed is recorded by sales, often weeks after the click. No ad platform sees it unless you send it back.
• Send it back, then read it. Won deals returned through Meta's Conversions API or Google Ads' offline conversion import are conversions the platform can count. Meta's lift study objectives can name offline event sets; check whether your study type accepts the event you send before you plan around it.
• For a geo test, count by region yourself. Export deals with their source and the customer's region from the CRM, and compare test regions against held-out ones on qualified leads, pipeline and revenue.

LeadJourney does not run incrementality tests, lift studies or geo experiments. What it supplies is the outcome: every visit recorded first-party and server-side, the click ids captured at the click, each lead's journey written onto the CRM record, and qualified leads, deals and revenue reported per channel and campaign. Won deals with their value go back to Meta, Google Ads, LinkedIn and Microsoft, so a platform's own test or model can work on deals rather than form fills. See multi-touch attribution for the attribution side.

Incrementality is the part of a result that would be gone without the marketing, measured against a holdout that did not get it. Lift is the relative gap in conversion rate, incremental conversions the count behind it, incremental CPA and iROAS the price and the return of what the ads actually caused. All of them depend on a comparable holdout, enough volume to beat chance and an outcome that matters to the business. For the decisions, read incrementality vs attribution for when to use which and incrementality testing for how to run a test on CRM revenue.