Incrementality calculator: the conversions your ads actually caused

Incrementality is the share of your results that would not have happened without the ads. Attribution answers a different question, which touchpoints led to a conversion, and credits a retargeting ad or a brand search click even when the person was coming anyway. A holdout test answers the incremental one directly: show the ads to one group, keep them from a comparable group, and the difference between the two is what the ads caused.

You need both. Attribution runs every day and tells you which campaign, keyword or creative to change; a lift test runs a few times a year and tells you how much a channel is really worth. The incrementality vs attribution post covers which question each answers, and the glossary entry defines the terms on this page.

Start from the two conversion rates, conversions divided by group size, because the groups are rarely the same size. Lift is the gap between them divided by the control rate. Incremental conversions are the same gap multiplied by the size of the test group: the conversions the ads added on top of what the test group would have done anyway.

Everything else divides by that count. Incremental cost per conversion is the spend in the test group divided by the incremental conversions; in the example above, €15,000 over 120 is €125. Incremental ROAS is the revenue of those conversions divided by spend: 120 conversions at €400 is €48,000, an iROAS of 3.2x. Your platform's ROAS for the same campaign will be higher, because it also credits the 600 minus 120 conversions that would have happened without the ads.

Two groups never convert at exactly the same rate, even when the ads do nothing. The calculator runs a two-proportion z-test, which asks how often a gap this size would appear by chance alone. Below 5 percent (95 percent confidence) the lift counts as proven; above it, the honest reading is not proven yet, which is not the same as no lift.

Sample size decides most of it. The example above, 600 against 480 out of 50,000 each, is proven well past 99.9 percent. The same rates on smaller groups, 130 against 110 out of 10,000 each, give an 18.2 percent lift at only 81 percent confidence. When a result is not proven, the calculator prints how many people per group a difference that size needs, at 95 percent confidence and 80 percent power, so you can decide whether to extend the test or accept that the effect is small.

• The z-test assumes people were split at random, as in a platform conversion lift study or a user-level holdout.
• A geo test randomises regions, not people, so a test on individual conversions overstates its confidence. Use the method the geo tool ships with; incrementality testing covers the designs.
• Decide the length and the outcome before the test starts. Stopping the day the number turns green makes chance look like lift.

A lift test is only as good as its outcome metric. In lead generation the pixel counts form fills, and a campaign can lift form fills while adding nothing to qualified leads or deals, because the extra forms come from people sales will never close. The outcome worth testing on lives in the CRM.

Long sales cycles make that harder, not optional. Read the test first on qualified leads or pipeline, which arrive within weeks, and again on won revenue once the deals have had time to close. The calculator takes any count, so run it once per stage and see where the lift holds.

LeadJourney does not run lift tests, has no holdout feature and does not build media mix models. What it does is the measurement a lift test on lead gen depends on: every visit recorded first-party and server-side, the click ids captured at the click, the lead joined to its CRM record, and qualified leads, deals and revenue reported per channel and campaign.

It also sends won deals back to Meta through the Conversions API and to Google Ads through offline conversion import, so a platform's own reporting can be read on deals rather than form fills. Whether a given lift study accepts those offline events as its outcome is set by the platform, so check your study type before you plan around it.