Field note · March 2026
Why D7 retention misreads Thai consumer apps
D7 is a borrowed unit of time. It travelled here with slide templates. It is not a law of user behaviour, and in Thailand it often measures the wrong week.
Most consumer templates assume a weekly grocery-like cadence: you open the app, you come back inside seven days, or you are “lost.” That story fits some social products. It does not fit a lending app whose next meaningful session is payday. It does not fit a parent education tool that goes quiet during exam weeks and returns when term starts. It does not fit festival commerce, when the useful window is twelve days and then nothing.
What D7 actually captures
A seven-day retained user is someone who performed a qualifying event inside a calendar week of the cohort origin. If your origin is install, you are measuring curiosity. If your origin is first funded transaction, you are closer to a product moment — but you still have not asked whether seven days is the natural gap between funded transactions.
In Activation Signal Laboratory we treat the window as part of the activation contract, not as a default chart in the BI tool. Teams who refuse to write the rhythm down keep D7 because it is comparable to a US benchmark tweeted in 2019. Comparability to a tweet is not a measurement goal.
Thai calendars are not noise
Public holidays cluster. Songkran is not “seasonality to smooth.” School calendars move family devices between homework apps and entertainment. PromptPay and salary cycles create two or three dense days a month for anything involving money. If you average those into a smooth D7 curve, you will congratulate yourself for a dip that was always going to happen on the 28th.
We ask students to overlay the last twelve months of Thai public holidays on their retention heatmap before they pick a window. It is unglamorous. It prevents a specific self-deception: calling a holiday trough “churn.”
A better question than “what is our D7?”
Ask: what is the longest gap after first value during which we still believe the person is in the product, not gone? For one Bangkok payments team that became fourteen days plus one completed transfer, not seven days plus app open. Their D7 looked healthy because people opened the app to check a balance and did nothing. The fourteen-day funded window looked worse and was true.
Worse-and-true is the point of Product-Led Growth Measurement. If your leadership cannot bear a lower number that describes reality, the laboratory will feel uncomfortable. That discomfort is the work.
How to redraw without inventing vanity
Document the rhythm you claim. Name the qualifying event. State who is excluded (internal, shared family login if you can detect it, failed KYC). Publish the window next to the chart, not in a footnote. When the window changes, version the chart; do not silently retcon last quarter.
If you need a peer critique on a specific window, that is what Signal Circle readout days are for. The field note is free. The argument in the room is the expensive part, and worth it.