28 Jan 2026 · Field notes
The silent drop between install and first value
Install is a loud event. First value is often quiet. Between them sits a stretch of minutes that many Bangkok apps still do not name: splash, OS permission, OTP, empty home, first tap that is not a skip control. In a funnel chart this stretch becomes a single cliff labelled “activation.” That label is how arguments start.
Onboarding flow optimization begins by splitting the silent interval into steps a human can point at. We do not need twenty events. We need every screen or system prompt that can fail for a different reason. A slow OTP vendor is not the same failure as a location prompt before the user has a reason to share it. If both hide under one drop, engineering and design will fix the wrong thing.
A practical split
In the Activation Lab we ask for five to nine named steps from first open to value moment. Splash counts if it can hang. Consent counts if it can exit. Account create counts only if it is required before value; if guest browse is possible, guest browse is a step and account create is later.
Session replay is the referee. If replay shows people lingering on an empty state while analytics shows them “active,” your event fired too early. If replay shows a crash and analytics shows a clean skip, your skip control is swallowing errors. Neither problem is exotic. Both are common in apps that instrumented onboarding once during a launch crunch and never revisited the names.
What “silent” usually means
Silence is not a lack of user intent. It is a lack of language in the data. Users are doing something: waiting, rereading, switching to LINE to grab an OTP, dismissing a card they did not understand. Until those actions have names, the team will debate personality (“our users are impatient”) instead of sequence.
Thai-market apps often add an extra silent patch around OTP and dual-language screens. If the Thai string wraps and the English string does not, you will see a drop that looks random by device language. That is still an onboarding leak. It just does not look like a classic Western carousel problem.
Continue with consent screens in Thailand, or sit the mapping work in the First-Session Activation Lab.