Twenty UA terms that changed meaning

Sit two UA hires from different studios in the same meeting, ask each of them to define ROAS, and there is a real chance they mean slightly different things. The drift has been building for a few years, as measurement got harder and vendors filled the gaps with conventions of their own. A short list worth agreeing on inside a team, then. Assuming everyone already means the same thing is how a review goes sideways.

ROAS on its own now almost always needs a qualifier. Blended ROAS mixes paid and organic revenue against paid spend alone. It flatters the number. Blended ROAS calculation is one of the more common places a model quietly goes wrong, usually when organic and paid cohorts never got cleanly separated upstream.

Modelled does more work than most people notice. It once meant a specific statistical technique applied to a known gap; now it often means little more than "estimated, source unspecified," so ask what model, trained on what data, before you treat a modelled number the way you treat an observed one.

Incrementality used to mean a rigorous holdout test, and it now gets stretched to cover any comparison between a campaign period and a prior period, which is a different claim carrying nowhere near the same confidence.

SKAN postback conversion values went from a niche technical detail to something a UA manager designs personally, because the value schema a team picks shapes directly what the automated bidding layer can learn.

Value optimisation now covers a simple day one revenue proxy at one end and a genuine multi-day predicted lifetime value model at the other, and the term on its own tells you almost nothing about which one a given campaign actually runs.

Attribution window has quietly shortened in practice across several platforms even where the contractual window never moved, because measurement partners tune default lookback periods for their own reporting convenience.

A cohort used to mean players who installed on the same day. Now it frequently means players who arrived through the same campaign, which is a materially different grouping for anyone comparing numbers across teams.

A longer glossary fixes none of this. What works is a one-page definitions document each team writes for itself, naming exactly what its own dashboards mean by each term, agreed once and pulled out whenever somebody challenges a number in a review.

Related archive reading

• ROAS targets versus payback windows
• The DMA for UA operators, in plain terms