1. Window honestly
Draw 24-hour and 7-day curves. Separate delayed conversion on PromptPay and payday cycles from true abandonment.
Conversion drop-off diagnostics, without the mural
This page is the studio’s working method. It is what we teach in the flagship room and what we refuse to decorate with north-star posters.
Start with a sentence that can be wrong
A diagnostic claim names a step, an audience, a window, and a suspected mechanism. “Checkout feels broken” is a mood. “Android 12 WebView users fail after the Thai error string introduced in design-system v4, within 24 hours” can be falsified. We do not move to solutions until the sentence can be attacked.
If the event dictionary cannot support that sentence, instrumentation is the assignment. Skipping ahead creates confident fiction.
The leak map
2. Split before you average
Device, locale, acquisition week, experiment flag. A national average is how Bangkok health hides Isaan breakage.
3. Sample replay last
A capped sample can support or weaken the sentence. It cannot author the sentence. Heatmaps are locations, not motives.
The Monday memo
One or two pages. Suspected cause. Disconfirming test. Instrumentation still missing. Explicit non-goals — especially A/B tests you are not licensed to run yet. The memo is finished when a product manager can put a date next to the next action without calling you.
We keep aftercare short: one new week of data against the memo. If the world changed, rewrite the sentence. Do not protect last week’s pride.
What this method will not do
It will not guarantee a conversion lift. It will not implement your SDK. It will not replace your vendor’s documentation. It will not turn a political fight about OKRs into a chart that makes everyone comfortable.
If that still sounds like the work you need, sit the flagship studio or write to the desk.
See the rooms