The Kernel · 3 February 2026
Why retention curves flatten on day three
Day three is where a lot of United Kingdom consumer apps stop pretending. The push copy has been used up, the widget has been tried, and the person who only needed a one-off job has left. In Cohort Retention Studio this flatten is treated as a clue, not a tragedy.
Blended time zero
If your cohort mixes first-open with first-purchase, day three is where the two populations disagree. Purchasers may still be in a delivery window. Browsers have already learned the app cannot do the job they brought. A single curve averages those stories into a polite plateau.
We ask for an exclusion list before we look at the shape. Returning customers from a previous year do not belong in a new-user retention cut, even if they reinstalled after a campaign. That mistake produced the grocery-delivery case we still teach: a “healthy” D7 that was almost entirely known shoppers.
Onboarding as debt, not a funnel
A permission dialog, a card scan, and a tutorial that cannot be skipped are not steps toward value. They are debt due by day three. If the first completed job happens after that pile, the curve will look like a cliff then a floor. Dressing the floor as “habit” is how teams commission the wrong feature.
Module six of the studio is deliberately short on inspiration. Students bring the actual screens between time zero and first value, then mark which ones would survive a bus-factor of one. Most flattens become obvious once the screens are in the ledger beside the numbers.
If the plateau remains after an unblended cut and a honest time zero, then — and only then — we talk about product. App Analytics is not a licence to skip that order of operations.