Cohort Signals: Predicting Mobile Churn Before It Happens
Event taxonomies that survive a rebrand, hazard models you can explain to a PM, and a review of Thai store-review language as an early exit cue.
Read the syllabusBangkok retention studio
Day 23
The median silent-exit window we still see in Thai and SEA mobile cohorts — before store reviews turn bitter.
Leave the vanity funnel. Learn to hear a departing user while they are still opening the app.
Apiauditbase is a teaching desk for app analytics and churn prediction for mobile apps. We work with product analysts, growth leads, and a few stubborn founders who would rather rebuild a feature store than buy another dashboard skin.
Open the course catalogue
“The Survival Curves module stopped me treating uninstall as a binary. I still disagree with how aggressive their paywall timing lab is, but the Day-23 framing paid for the tuition.”— Mira K., growth lead, ride-hailing · Bangkok
Flagship
Cohort Signals
Our long programme on predicting mobile churn before finance asks why revenue sagged. Twelve live sessions, a scored feature store, and a limitation we do not hide: it will not rescue a product nobody wanted.
What sits beside it
Shorter desks cover push fatigue, paywall windows, and event naming. They share the same instructors and the same allergy to inflated “retention scores.”
If you only have one quarter, start with Cohort Signals. If your taxonomy is already a mess, begin with Event Taxonomy for Retention Teams, then return.
Compare programmesWhy this desk
Methods we actually teach
Read silence, not stars
Most churn is a fading session length, not a one-star review. We train you to mark quiet decay in session depth, widget use, and background refresh before the uninstall ping arrives.
Models a PM can argue with
If a hazard ratio cannot be told in one hallway sentence, it does not leave the notebook. We keep the math honest and the story short enough for a Tuesday standup.
SEA context, not imported playbooks
Prepaid users, family-shared devices, and Line-first onboarding change how “active” looks. The studio is in Samphanthawong for a reason: the logs we study come from this region first.
71
product teams taught since the desk opened
18 mo
of labelled cohort archives used in class
3.2×
median lift in 30-day return after the paywall lab
2019
Bangkok studio year — still one room, still no sales floor
Journal
Field notes, not recaps
Why Day-7 churn is a misleading KPI
A convenient number that hides the users who were never going to stay past onboarding.
Building a churn feature store for mobile
Which events deserve a column, and which ones only create noise for the model.
Reading silent uninstall signals
Background refresh death and widget neglect as earlier tells than a store rating.
Bring a messy export. Leave with a named risk window.
We do not sell software. Pricing on this site is informational. If your team already argues about retention every Monday, write to the desk or sit in on Cohort Signals.