[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"glossary-incrementality::en":3,"gloss-cluster-incrementality::en":23,"gloss-next-incrementality::en":9},{"slug":4,"category":5,"name":6,"definition":7,"meta_desc":8,"faq":9,"schema_markup":9,"related":10},"incrementality","analytics","Incrementality","Incrementality is the share of an outcome that would not have happened without the thing you are measuring. It is the question attribution cannot answer: a report that credits a channel with a thousand signups is telling you those people touched the channel on the way in, not that the channel caused them to arrive. Some of them were already coming. The gap between credited and caused is often large, and it is largest exactly where attribution looks best, because the channels that intercept people close to the moment of purchase — branded search, retargeting, a discount email to an active user — are the ones most likely to be collecting credit for demand that already existed. Measuring it requires a comparison group that did not get the treatment, which is what separates an incrementality test from a report. Where you can randomise at the user level, an ordinary controlled experiment answers the question directly. Where you cannot — because ad platforms decide who sees what, or because the treatment is a brand campaign that reaches whole markets — the usual designs work at a coarser grain: hold a set of regions out while comparable regions receive the campaign, or switch spend off and on in alternating periods and compare against the forecast the earlier periods imply. Both are blunt, both need the holdout to be genuinely comparable, and both cost real money while running, which is the honest reason they are rare. The practical stance is proportionate. You do not need an incrementality test for every line of spend, but you should have run one for the largest lines, because a channel that looks efficient under attribution and turns out to be mostly harvesting demand is the most expensive mistake in the budget. It is worth being clear that a null result is a real result here. A test that cannot distinguish the treated group from the held-out one has told you something expensive and useful, and the common mistake is to rerun it with a smaller holdout until the difference reappears.","Incrementality is what would not have happened anyway. Why attribution credits demand it did not create, and the holdout designs that measure the difference.",null,[11,14,17,20],{"slug":12,"name":13},"ab-testing","A\u002FB Testing",{"slug":15,"name":16},"attribution-model","Attribution Model",{"slug":18,"name":19},"channel-market-fit","Channel-Market Fit",{"slug":21,"name":22},"statistical-significance","Statistical Significance",[24,26,29,32,35,38,42,45,48,51,54,57],{"slug":12,"category":5,"name":13,"updated_at":25},"2026-09-08T21:32:36+00:00",{"slug":27,"category":5,"name":28,"updated_at":25},"active-user","Active User (DAU, WAU, MAU)",{"slug":30,"category":5,"name":31,"updated_at":25},"autocapture","Autocapture",{"slug":33,"category":5,"name":34,"updated_at":25},"cost-per-resolution","Cost per Resolution",{"slug":36,"category":5,"name":37,"updated_at":25},"customer-data-platform","Customer Data Platform (CDP)",{"slug":39,"category":5,"name":40,"updated_at":41},"data-sampling","Data Sampling","2026-09-08T21:32:37+00:00",{"slug":43,"category":5,"name":44,"updated_at":25},"deflection-rate","Deflection Rate",{"slug":46,"category":5,"name":47,"updated_at":41},"double-counting","Double Counting",{"slug":49,"category":5,"name":50,"updated_at":41},"feedback-signal","Feedback Signal",{"slug":52,"category":5,"name":53,"updated_at":41},"goodharts-law","Goodhart's Law",{"slug":55,"category":5,"name":56,"updated_at":25},"guardrail-metric","Guardrail Metric",{"slug":58,"category":5,"name":59,"updated_at":25},"identity-resolution","Identity Resolution"]