Journal · 21 January 2026
Cohort windows for UK fintech
A seven-day retention chart imported from a Californian consumer app will lie to a United Kingdom payments product, politely and every week.
Payday is not evenly sprinkled through a month. Faster Payments can make a Friday afternoon look like product-market fit. A bank holiday can empty Monday. If your cohort window is “week 1 / week 2” in UTC, you are measuring the calendar of the warehouse, not the cashflow of the user.
Write the window as a rule, not a chart setting
Example we use in studio: a new payer enters the cohort on first successful inbound payment, not on KYC-passed. The first return window is “next expected payday ± two working days”, with working days defined as England & Wales banking days unless the product is explicitly Scottish-payroll. Ugly. Defensible. Reproducible in six months when someone asks why March looks odd.
SCA step-ups and frozen accounts are not drop-off in the moral sense. Record them as exclusions or as a parallel ledger. Mixing “could not pay” with “chose not to return” is how App Analytics becomes a weapon in a credit-risk meeting it does not belong in.
What we refuse to automate
A vendor toggle labelled “smart retention” will not know your regulator. The paper ledger in the Atelier is slower on purpose. Once the rule is written, automate it. Not before.