Journal
When ROAS hides a weak retention loop
A subscription app in our 2025 cohort posted a cheerful D7 ROAS. Trials were converting in the MMP because the trial start was the conversion event. Refunds after day eight sat in a finance spreadsheet nobody joined.
Once trial refunds entered the same window as media, two networks remained merely acceptable. A third became a delay machine: it filled the top of the funnel with people who liked the offer and disliked the product.
Retention is not a UA KPI until you force the join. We use a simple overlay: cohort retention curves beside spend. If day-14 retained users per £1,000 fall while ROAS holds, you are buying a story.
Product still owns the loop. Courses cannot invent a reason to open the app. What we can do is stop UA from laundering a product problem through a platform tile.
That is why Media Mix for Subscription Apps includes a finance liaison hour. The argument needs a person who cares about refunds in the room, not only a person who cares about CPI.