12 May 2026 · Mali Khemthong

Why paid install quality often collapses after week three

Bangkok UA desks still celebrate a network that “wins” CPI in the first ten days. By day twenty-one the retained users look like a different product. That is not mysterious. It is sampling.

Most paid sources front-load people who were already close to installing: retargeting bleed, brand search leakage, and users who click anything with a reward. Those people convert cheaply. They also leave. The durable audience — the one that should determine the score — arrives later and costs more. If your scorecard freezes at week one, you are scoring the sample, not the channel.

What we ask Studio students to plot

Take one source. Plot Day-7 retained users by install week, not by report date. Keep CPI on a second axis if you must, but do not let it lead. When week-three retained quality drops more than 25% while CPI stays flattering, the score moves down even if blended ROAS still looks polite.

We also ask whether creative changed in that window. If the asset calendar is stable and quality still falls, you are looking at audience rot or inventory quality, not fatigue. That distinction is the whole of a later note.

A limitation, stated plainly

Week-three decay does not prove incrementality. A holdout might show the “decaying” source was still incremental to store listing. Without a test, the Channel Score records a quality defect and refuses to call it causal lift. Finance partners prefer that honesty once they have been burned by a composite index.

If you want the full critique sequence, it lives in Channel Quality Scoring Studio, module four.