3 Apr 2026 · 13 min
Choosing cohort windows when checkout lag is 9–14 days
A seven-day CAC in Thailand is often a story about payment method, not about media quality. The window is a product decision pretending to be a tracking setting.
COD, bank transfer, and “pay at 7-Eleven” do not fail because the pixel is shy. They succeed on a human schedule: confirm with family, wait for salary, collect after Songkran travel. If your User Acquisition Cost Analytics window closes before that behaviour finishes, you will report a CAC that finance cannot find in the month.
Observed versus modelled
We split the window into two parts. Observed: the period where you wait for actual orders to appear, no modelling. For many Bangkok DTC brands that is 14–21 days. Modelled tail: a conservative estimate of late conversions you will not wait for in the weekly operating meeting. The tail should be written down, including the error you accept. If you cannot defend the model, drop it and live with a slower observed number.
Festivals are not outliers to delete
Chinese New Year, Songkran, and mid-year sales change both spend and lag. Deleting those weeks from a CAC trend makes the chart prettier and the decision worse. Mark them. Compare festival-to-festival, not festival-to-a-quiet-Tuesday.
Refunds and failed COD
A window that counts a COD order at ship and never subtracts failed delivery will understate CAC until operations explodes. Join refunds and failed-COD flags into the same cohort as the original spend. This is unglamorous work. It is also why controllers trust some teams and not others.
A policy you can print
Example from a Studio Seat workbook (anonymised): “New-to-file CAC uses spend in week W against first paid order within 21 days, minus failed COD and refunds within 45 days. We do not quote 7-day CAC in finance meetings. Media may watch 7-day operational CPA for pacing only.” That paragraph ended a year of slide wars.
The Cohort Economics Intensive exists for teams whose CRM export is noisier than their ad account. The flagship CAC Measurement Studio still starts here: if the window is fantasy, every downstream test is theatre.