← Four arguments 02

Stop measuring TV like search.

Television builds memory over weeks; judging it on attribution windows designed for intent capture makes a working channel look broken.

26 pieces · 6 authored · related work since 2020

The argument

Somewhere along the way we started talking about TV in the language of paid search and social: outcomes, ROAS, performance. But someone watching a drama on a Tuesday night is not typing a query into a search box. One is watching content. The other is expressing intent. TV works over weeks and months, and when you judge it on attribution windows built for search, it looks like it failed when it’s doing exactly what it was built to do.

We made the same mistake with targeting. CTV is not a personal screen; it’s a household screen. A logged-in profile doesn’t tell you who is on the couch. Pile on third-party segments and supply shrinks, CPMs climb, and delivery collapses into the same five markets every time. I start somewhere else: the customer file, the ZIP codes where customers over-index, census data and lookalike markets. Then I buy direct from the roughly 20 publishers that hold most of the viewing.

I’m not arguing for less measurement. I’m arguing for measurement that fits the channel, with each tool doing the job it’s good at. Show-level CPA is an in-flight signal: it tells me which programs to lean into and which to cut while the campaign is still running. Incrementality tests and mix models answer the bigger question of whether TV is building the business over time. Read them together, and treat identity as a standard you measure, not an assumption you make. Plan CTV like linear: context first, geography informed, built for households, not IDs.

The full trail, oldest first

26 pieces