Week five’s calendar exercise saved our Songkran campaign. We had planned a two-week paywall test straight through the holiday. Nalinee made us redraw the window on the whiteboard; the control would have been unreadable.
Flagship course
Experiment Design for Mobile Releases
Eleven weeks on the decisions that happen before anyone taps “start experiment”: who is randomised, what counts as success, and which numbers may never be primary. Taught by Nalinee Prasert from the Bangkhunprom studio, with remote seats on the same call.
Learning outcomes
- Write a hypothesis that names the user action, the constraint, and the time window — not a slogan.
- Choose a randomisation unit that survives shared devices and logout/login on Thai consumer apps.
- Build a sample plan that respects App Store and Play review delay instead of pretending traffic is a tap.
- Set one primary metric and two guardrails, including at least one stability metric (crash-free or ANR).
- Deliver a one-page readout memo a release manager can accept or reject without opening a BI tool.
Modules
- Week 1–2 · Surfaces and assignment Store listing, first-open, paywall, push opt-in. Where identity actually lives on iOS 17+ and modern Android.
- Week 3–4 · Hypotheses that can fail Minimum detectable effect against thin weekly actives. Writing the “we will not ship if…” clause first.
- Week 5–6 · Sample and calendar Festival weeks, payday spikes, and why October in Thailand is a terrible default window. Holdouts that do not starve the control.
- Week 7–8 · Variant craft under review Screenshot tests versus in-app copy tests. What you may change after a store rejection without breaking assignment.
- Week 9–10 · Readout discipline Heterogeneity without fishing. Guardrail vetoes. Communicating a flat result without inventing a silver lining.
- Week 11 · Public memo You present to the room. Alumni from the reading circle sit in. No slide theatre.
Nalinee Prasert
Former growth analytics lead for a Bangkok super-app. She still reviews alumni memos on Thursday evenings and refuses to grade on enthusiasm.
Informational pricing
A Variant Atelier seat on this flagship is listed at ฿28,500 for eleven weeks, including critique and office hours. Hypothesis Desk (recordings + two office hours) is ฿14,900. Team seats sit on the pricing page. There is no payment form here; write to the studio to reserve a chair.
Questions we actually get
Do you cover server-side feature flags?
Only at a glance. This flagship is built around client-side assignment, store-listing experiments, and in-app flows. Teams running remote-config-heavy backends should pair the course with their vendor’s own documentation; we will not pretend six lectures replace that.
What if my app cannot run a live test during the cohort?
You work from the archive: a historical brief with real traffic curves. You still write the memo. We will not invent a fake live experiment for your portfolio.
Is the Bangkok room required?
No. Remote seats join the same critique. The difference is hallway time, not curriculum.
Can I expense this?
Most students do. We issue a Thai tax invoice after enrolment is confirmed by email. See refund terms before you commit.
Reviews of this course
The statistical modules assume you already live in spreadsheets. I paused after week three, caught up on power calculations over a weekend, and only then did the Android holdout exercise make sense. The pause was on me, not on the syllabus — but the jump is real.
★★★★☆
Memo format is now our internal template. I still wish there was a dedicated StoreKit lab; we improvised with screenshots.