Permission Funnel Diagnostics
Instrument exposure, defer, deny, and later recovery. Build a ledger that product and data can argue from without inflating prompt-view counts.
Open the syllabusBangkok · Permission analytics studio
Opt-in is not a prompt. It is a funnel with a memory.
Dataauditbase trains product, growth, and data teams to read notification permission the way they already read checkout: as a sequence of exposures, hesitations, and recoveries. We work from event logs, not slideware, and we keep Thailand’s OS mix and language nuance in the model.
18
Studio programmes delivered since 2019, each rebuilt from live event schemas rather than generic growth playbooks.
7
Thai product teams in the most recent cohort, spanning wallets, commerce, and media.
2.4×
Median re-permission recovery versus firing the OS dialog on first open alone.
11 mo
Typical observation window we ask teams to keep before calling a prompt “done.”
Programmes
Three desks you can sit at this quarter
Each programme is a working method, not a survey of buzzwords. The flagship goes deep on funnel instrumentation; the others isolate quieter failure modes.
Quiet Hours & Fatigue Modelling
Separate true opt-out from mute-and-ignore. Map send density against session quality before you rewrite copy.
Ask about the next intake
Channel Mix After Opt-In
Once permission is granted, most teams still treat push as a single pipe. This lab splits transactional, lifecycle, and promo without double-counting reach.
Ask about the next intakeFrom the room
What analysts actually changed after sitting with us
We had been reporting “prompt shown” as a success metric. After Permission Funnel Diagnostics we split exposure from decision, and the growth review finally stopped celebrating a screen that 40% of users never actually saw.
The Quiet Hours module made our night-time blast look worse than we wanted to admit. I still think the fatigue model over-weights weekday office hours for our student audience, but it stopped us from raising frequency “to test reach.”
Short and specific: we delayed the iOS prompt until after the first saved address. Accept moved from the low twenties to the high forties in six weeks. No new creative, just timing from the ledger.
How a desk works
Bring the schema. Leave with a permission ledger.
01
Event inventory
We map what you already fire: screen views, soft-ask CTAs, OS dialog results, settings toggles, and uninstall-adjacent silence. Missing events are listed, not guessed.
02
Cohort cuts
First-open versus returning, Thai-language versus English UI, Android 13+ versus older, organic versus paid. Opt-in looks different in each slice.
03
Decision memo
A one-page recommendation: when to ask, what to stop sending, and which recovery path is even eligible. Tuition does not include implementation inside your app.
Why teams book us
Permission work that survives a sceptical data review
No inflated denominators
We teach you to exclude users who never reached a prompt-capable screen, and to keep pre-permission impressions out of “opt-in rate.”
OS honesty
iOS, Android 13 notification permission, and in-app channel toggles are modelled as separate gates, not one blended percentage.
Local reading
Copy tests and festival calendars in Thailand change prompt patience. We keep that in the analysis instead of importing a US SaaS template.
Journal
Recent field notes
Soft-ask before the system prompt
Why a cheap in-app screen is not “just UX,” and how to measure whether it actually trains Allow versus Don’t Allow.
Read the note
Seasonality of opt-in in Thailand
Songkran, 11.11, and payday weeks move prompt patience. Ignoring the calendar makes a good experiment look like a failed creative test.
Read the noteBring a messy event log. We will help you read it.
Tuition is listed plainly. There is no checkout on this site — write to the studio if a desk fits.