2 October 2025

Cohort charts from Thai mobile gaming

A Bangkok mid-core studio sat Cohort Signals with IAP timestamps, session lengths, and guild chat logs. They wanted a spend model. They left with a social one. This note is the public version of that sitting, with identifiers removed.

Payday is a calendar, not a persona

Thai salary cycles still imprint on mid-core spend. Cohorts acquired just after payday look “high value” for ten days and then resemble everyone else. If you train a churn model on calendar recency of purchase without a payday index, you will call ordinary post-payday quiet a death spiral. We now keep a simple month-progress feature in gaming stores. It is crude. It beats another recency column.

Guild mute as an exit cue

Muting guild chat predicted voluntary churn in the 20–35 day window better than days since last purchase. Spenders who mute are not always leaving the economy; they are leaving the social room, and the economy follows. The studio shipped a lighter re-engagement quest instead of a discount. Revenue stayed flat. That window’s churn fell. Finance was unimpressed; production was not.

Shared devices

Younger players on family Android phones create false uniqueness. Two people, one Google account, staggered session hours. “Churn” was often a sibling going back to school. Household-level windows are ugly. Person-level windows were lying more.

What we would not generalise

This was one title, one year, one city. Hyper-casual ads-only games do not behave like this. Neither do hardcore PC ports. The transferable lesson is methodological: ask which relationship is decaying, not which wallet. IAP recency is a loud column. It is rarely the first one that tells the truth.

Guest instructor Niran Vong still uses this case in module eight when store-review sentiment and social decay disagree. They often do. Sentiment spiked after a balance patch; mute rates told the slower story that mattered.