Designing cohorts around Songkran
Finance likes months. Travelers do not travel in months. On Thai corridors, a March install and an April departure can be the same Songkran trip — or two unrelated products. Travel App Analytics has to choose.
Corridor Seasonality Lab spends its first evening arguing with the Gregorian calendar. Not because calendars are useless, but because they are the wrong primary key for travel demand in Thailand.
Songkran is a window, not a month label
The public holiday dates move slightly; the travel window is wider. People leave Bangkok early and return late. If you cohort “April users,” you mix Songkran travelers with people booking a routine Hua Hin weekend after the roads clear. Their search behavior, price sensitivity, and cancellation risk are not cousins.
We define a Songkran window per corridor: origin, destination, and intended travel dates, not install dates. A traveler who installed in February and flies 12 April belongs to Songkran on that corridor. A traveler who installed on 3 April and flies in June does not.
Chinese New Year collides differently
Inbound demand from Greater China onto Thai destinations does not peak on the same days as domestic Songkran traffic. If your app serves both, a single “Q1 peak” chart will tell you to staff the wrong language pack. Route telemetry keeps two calendars. Yes, that is more work. No, averaging them does not save time; it creates a fake trend.
Shoulder months are not “off”
May after Songkran is often treated as a dead zone. For some island corridors it is a bargain-hunting season with longer consideration. Cohort windows should stretch when consideration stretches. A seven-day activation metric copied from a consumer social app will declare those travelers lost while they are still comparing ferries.
A cohort is a promise that two people faced a comparable market. Calendar months rarely keep that promise on Thai routes.
A practical table
In class we build a corridor calendar with: window name, start and end of intended travel, typical booking lead time, and which product surfaces matter (search, insurance attach, seat maps). That table sits next to the event schema. When someone asks for “March versus April,” we translate into windows or we refuse the chart.
If you want help building the first version, write to the desk or read the route telemetry method before you enrol.