“Module 04 on Thai holidays was the first time our Day-7 spike after Songkran looked embarrassing in a useful way. I wanted more time on listing-level randomisation; we only sketched it.”
Flagship bench
Holdout Design for Product Retention
Six teaching sessions on building holdouts that survive seasonality, Thai holiday calendars, and a VP who wants a win by Friday. Informational fee: THB 18,400. No payment is taken on this page.
Ask for the next seat
What you leave able to do
- Define return without defaulting to a Day-7 vanity window
- Draw a holdout map, including known leakage sources
- Adjust windows around Songkran and payday effects
- Read a survival curve without claiming unearned causality
- Pre-register a stop rule before anyone opens a chart
- Write a one-page memo a legal or brand reviewer can parse
- Explain why an underpowered cell should not ship
- Separate novelty spikes from durable return
Modules
01 — Defining return without vanity windows
We inventory how your product currently “counts” a returning user, then rebuild the definition around the decision you need. Includes a short autopsy of Day-7 culture in Southeast Asian consumer apps.
02 — Cohort construction that survives product changes
Versioned events, renamed screens, and silent tracking-plan drift. You practice freezing a cohort definition so last quarter remains comparable after a navigation redesign.
03 — Holdout cells when volume is modest
User, store, listing, and zone randomisation. What to do when eleven thousand weekly actives are not enough for the test you sketched on a whiteboard.
04 — Guardrails: novelty, seasonality, Thai holiday calendars
Songkran, Chinese New Year spillover, monsoon delivery delays, and payday. We treat the calendar as part of the design, not an after-the-fact excuse.
05 — Reading survival curves without overclaiming
Kaplan–Meier and simpler empirical curves. Language for “still in the product” versus “opened a push.” Where uninstalls quietly bias the picture.
06 — Shipping a decision memo, not a dashboard
One page: hypothesis, unit, primary, guardrails, stop rule, leakage. You rewrite a real brief from your desk. We mark it within seventy-two hours.
Instructor
Kamonchanok “May” Prasert
Former lifecycle lead at a Bangkok super-app. May ran holdouts on wallet onboarding and rider incentives before opening Logicfeaturegrid. She marks every Experiment Bench memo herself. Visiting critiques on instrumentation are occasionally given by Daniel Ruiz.
Informational fee
THB 18,400 for the six-session Experiment Bench seat. Corporate invoicing for Thai-registered companies is available after a fit call. Field Notes (THB 4,900) is a separate self-paced archive and is not a substitute for homework review.
This site does not process payment. See other seating types or refund terms.
Who should not enrol
If you cannot query events, if your brief is paid user-acquisition incrementality, or if you need a general SQL bootcamp, this bench will under-serve you. We would rather decline than stretch the syllabus.
FAQ
Is this a statistics degree in miniature?
No. We teach enough inference to stop you from lying to yourself. We do not cover Bayesian hierarchical models or causal forests. If that is what you want, we will point you elsewhere.
Do you teach paid media incrementality?
No. This is a real limitation: the bench stays on product and messaging retention. Media-mix and ads incrementality use different leakage patterns and different politics. Enrolling here to fix a UA holdout will leave you under-served.
Can I attend remotely?
Yes. Remote seats follow the same homework clock. The Cohort Table on Chakrphong Road is a different seating type with its own fee.
What software do you require?
A warehouse you can query and a spreadsheet. We do not require a specific BI tool. Notebooks are welcome; they are not graded as artefacts — the memo is.
Reviews of this bench
★★★★☆
4/5 on our wiki after the 2024 remote bench. Memo template stuck. Live sessions ran past the advertised ninety minutes twice.
Somsak, Hat Yai: finished the course, still cannot get finance to accept a four-week primary. The limitation is our org, not the syllabus — though I wish the politics module had been longer.