Journal · 12 March 2026
Cohort decay in Thai subscription brands
A subscription that pauses in April is not the same object as a subscription that cancels in August. Thai operators still fold both into a single “month-three retention” tile, then wonder why the Customer Lifetime Value Analytics model disagrees with the finance forecast.
Songkran empties warehouses and inboxes. School holidays change who is home to receive a meal kit. Mid-year beauty promotions pull in buyers whose second box was never intended. If your cohort table ignores those calendars, decay looks like a personality trait of the brand. It is often a calendar artefact.
What we ask students to plot first
Before any survival model, plot active status by acquisition week, not by a rolling 90-day repeat. Mark public holidays and the brand’s own promo windows on the same axis. In the last 61 working cohorts on our desk, the April intake for replenishment brands routinely showed a steeper drop than the February intake — until pauses were separated from cancels. After that split, the two curves were almost cousins.
Pauses are not polite churn
A pause that resumes in May still belongs in the population if your contract allows it. Dropping paused accounts from the denominator makes month-three look heroic and makes CLV look like a vanity metric. Keep them, label them, and let contribution during pause sit at zero rather than inventing a “loyalty score.”
A limitation worth writing down
If you only have one Songkran inside your extract, you do not yet know whether last year’s dip will repeat. Say that in the memo. Predictive CLV that pretends a single holiday is a law of nature is worse than a historical ranking with an honest window.
The flagship module on cohort grammar walks this with a Thai beauty replenishment extract. If you want the studio pressure test, bring your own pause codes — we will not invent them for you.