Geo lift uses geographic regions — DMAs, metros, or states — as the unit, not users. A channel runs in a treatment set of markets and is held out in matched controls; the revenue gap between them, adjusted for the pre-period trend, is the incremental contribution. Output is incremental revenue or incremental ROAS (incremental revenue ÷ treatment-geo spend).
The mechanic hinges on the match. “Matched” usually means a synthetic control — a weighted blend of control markets built to mimic the treatment market’s pre-period revenue path — or a simpler pre-period correlation match. A brand testing YouTube might run it in Chicago, Houston, and Phoenix while holding it out in Dallas, Atlanta, and Philadelphia — controls whose pre-period revenue tracked the treatment DMAs. Tests typically run two to six weeks, after a pre-period long enough to establish the match.
DTC operators reach for it because it sidesteps the user-level identity problems that ATT and privacy shifts created — no IDFAs or cookies to lose when the unit is a metro. It is the brand-side, channel-agnostic counterpart to platform-run conversion lift studies, and it answers what would have happened without the spend — a counterfactual that deterministic attribution cannot.
Constraints are practical: geographic spend variance, baseline volume per market for statistical power, and a clean pre-period. Geo lift is often pitched as the option for brands too small for marketing-mix modeling, but the same preconditions decide both — a brand at very small scale, or one running the same spend pattern everywhere, can fail both. A national promo or weather shock can swamp the read.