Measurement

Measurement is the category for how a marketing team learns whether anything it did worked, and what changes once agents act on the numbers. The author's view, which the pieces here will test: a holdout or geo test beats a model, a model beats platform attribution, and platform attribution beats a dashboard that counts clicks, and that order matters more once an agent runs the media, because an agent optimizes whatever it is given, the wrong thing included, and faster than a person would.

Three pieces are planned. The first will ask what an agent does when it is pointed at a platform metric, and whether the evidence hierarchy holds up as the thing that stops it. The second will ask whether incrementality testing can stop being a quarterly project and become the fuel for an agent-run learning loop, and what that costs a media team in holdouts. The third will ask what has to change in marketing mix modeling when allocation is continuous instead of annual: how often the model is read, where its inputs come from, and how small the decisions it informs become.

The reading order below puts the hierarchy first, the tests second, and the model third, because the second two assume the first.