A lexicon for incrementality

Incrementality is a causal claim about conversions that would not have happened without the intervention, in a named population and window. It is not a platform metric with the same label.

Conversion lift is the estimated difference between treated and untreated groups after you have stated the unit of randomisation. If you cannot name the unit, you are looking at a delta, not lift.

Holdout is an operationally untreated group that still exists at the end of the window. A group that was supposed to be untreated and then received CRM is not a holdout. It is a leak.

Contamination is any path that moves treatment into the control, or control behaviour into the treated, including shared households and retailer media. We treat it as a design problem, not a footnote.

Geo-experiment is randomisation or matched assignment at a geographic unit you can actually buy against. A coloured map is not the method.

The In-house Measurement Desk exists largely to force this lexicon onto a wall everyone can see. Until the words pin to operations, conversion lift measurement stays a slogan.