App traffic in Thailand rarely arrives as a flat line. Campaign bursts, payday cycles, and festival weeks reshape who shows up and how they behave.
Compare like with like
If variant A collected most of its exposures during a promotion and variant B during quieter days, the measured lift may describe calendar effects more than product changes.
Segment the story carefully
Cohorts by acquisition channel can help, but only when assignment still holds inside each slice. Splitting after the fact without checking assignment balance is a common measurement mistake.
Document the window in the readout
Stakeholders deserve to see the experiment window next to the lift. A Results Readout Facilitation session keeps that context attached to the decision.