Week 1 · Event inventory
We list every permission-adjacent event you already fire and mark gaps. Missing OS result codes are treated as blockers, not as “unknown.”
Flagship programme
Permission Funnel Diagnostics
A six-week desk for product, growth, and analytics partners who need a shared ledger for notification opt-in. You bring event names. You leave able to defend a prompt change without inflating the denominator.
Learning outcomes
- Define exposure, defer, accept, deny, and later recovery as separate events, including in-app channel toggles.
- Cut cohorts by first-open versus returning, UI language, paid versus organic, and OS version.
- Write a one-page decision memo a PM can take to leadership without a supporting slide deck.
- Identify recovery paths that are legally and technically ineligible, especially after a system deny.
- Keep Thai festival and payday calendars out of “creative failed” conclusions.
Modules
Week 2 · Soft-ask versus system prompt
Measure whether the in-app screen changes Allow rates or merely delays Don’t Allow. Includes bilingual CTA labelling.
Week 3 · Denominator hygiene
Remove users who never reached a prompt-capable screen. Rebuild opt-in so marketing cannot accidentally report prompt views as decisions.
Week 4 · OS split
iOS, Android 13 runtime permission, older Android defaults, and OEM skins. One blended percentage is retired.
Week 5 · Recovery eligibility
Which denies can be asked again, which must go to settings, and which should be left alone. We do not teach dark-pattern re-prompts.
Week 6 · Decision memo
Each squad writes and defends a one-page recommendation. Feedback focuses on the claim, the window, and the limitation.
Instructor
Mali Srisawat
Mali led notification analytics for a Bangkok consumer finance app before joining Dataauditbase. She still reviews every Prompt Ledger memo. Her bias, stated up front: if the SDK cannot read OS state, she will not let a team publish an opt-in rate.
FAQ
Do we need a warehouse?
It helps. We have run the desk on well-labelled CSV extracts. If your events lack OS result codes, week one will be spent designing those events rather than reading rates. That is a real delay, not a failure of the student.
Will you implement the prompt in our app?
No. Dataauditbase is a teaching studio. Engineering, copy, and vendor configuration stay with your team. We will review instrumentation plans; we will not merge code.
What is the honest limitation of this programme?
Six weeks cannot create permission state that the OS will not expose. On some OEM Android builds, “denied” and “not determined” collapse in ways no ledger can uncollapse. We document that ceiling instead of promising a workaround.
Can marketing attend without analytics?
One marketer on a four-person seat is useful. A room of only marketers tends to skip denominator work. We ask for at least one person who can write a query or defend a metric definition.
Reviews of this desk
Week 3 hurt. Our opt-in number fell once exposure was cleaned. Leadership still approved the delayed prompt because the memo named the window and the OS split. I would not sit this desk without a data partner in the room.
★★★★☆Mali is precise to the point of blunt. The bilingual CTA session saved us from treating Thai and English labels as one experiment. Homework volume in week 4 was heavier than the syllabus implied.