Incrementality

Why platform ROAS overstates it: platforms count conversions their ads touched. View-through attribution credits an impression someone scrolled past; retargeting claims customers already en route to purchase; branded search claims people who typed your name because they already wanted you. Each mechanism books revenue that would have arrived anyway. The gap between touched and caused is not a rounding error; it is routinely the majority of reported conversions in bottom-funnel tactics, which is why incrementality testing exists as an industry.²

Three estimation routes, in increasing order of statistical machinery. Holdout tests suppress ads for a random user or region group and compare outcomes. Geo-lift tests, the product built by vendors like Haus, do this with matched markets: dark in Denver, live in Phoenix, difference measured against a synthetic control.² MMM estimates incrementality statistically from spend variation across two-plus years of history.¹ Experiments are ground truth for one question at a time; the model covers every channel continuously. The modern stack uses experiments to calibrate the model, covered properly in triangulation.