Week 1–2 · Event dictionary and the ugly week
Rebuild names for open, return, and the feature under debate. Plot daily actives against the calendar before any smoother is allowed. Identify missing days and double-fires.
Flagship cohort
Retention Horizon: forecasting usage from event streams
An eight-week live desk. You leave with a 7- to 28-day usage forecast built from events, a one-page memo, and a clear list of what the model cannot see.
Modules
Week 3–4 · Seasonality, payday, and Thai holidays
Fit weekday effects. Mark Songkran, Chinese New Year retail, and local payday Fridays. Decide which store-listing spikes belong in the series and which belong in a footnote.
Week 5–6 · Priors, horizons, and honest uncertainty
Choose 7, 14, or 28 days. Write a prior you could explain to a designer. Compare a seasonal baseline to one heavier method — and keep the simpler one if the heavier one cannot earn the extra story.
Week 7–8 · Memo, peer critique, and a miss you can narrate
Present to someone outside your industry. Revise. Ship a one-page note: forecast, drivers, falsifiers, and the product lever you would pull if the path is wrong.
Learning outcomes
A defendable horizon
You can say why the number is 14 days rather than 90, and what would make you refuse to forecast at all.
A maintainable dictionary
Your team inherits event names that a new analyst can update without archaeology.
A memo, not a notebook tour
Leadership can read the output in a meeting. If they cannot, the desk is not complete.
Instructor
Nalinee Srisuk
Nalinee led growth analytics at a Bangkok super-app before joining the studio. She still sits in on Atelier weeks when a bank or ride-hailing team brings a series that refuses to behave. She teaches in English, with Thai examples, and will stop a lab if the event names are still fiction.
Informational pricing
Tuition for the public cohort is THB 24,900 per person. This page does not take payment. If a cycle is open, enquire and we will send dates, VAT, and the enrolment note. Private teams should look at Atelier on the pricing page instead.
FAQ
Do I need to bring my own event data?
Helpful, not required. If security blocks exports, we issue a masked Thai super-app extract with missing days and renamed screens. You will still practice the same sequence.
What tools do you assume?
A spreadsheet you can think in, plus one notebook environment you already use. We do not require a specific cloud. SQL helps; it is not a gate in week one.
Where does this fall short?
Retention Horizon will not rebuild your entire tracking plan, will not certify you on a vendor, and will not produce a reliable 90-day forecast from six weeks of noisy data. If your events are still unnamed after week two, the later models will look precise and still be wrong — that limitation is real, and we will say so rather than stretch the syllabus.
Can I join remotely from outside Bangkok?
Yes. Live labs run on Bangkok evenings. Critique is written, so time zones other than Thailand still work if you can attend two sessions a week or watch the recording within 24 hours and submit notes.
From people who finished the desk
Week one’s dictionary work felt like stalling. By week six I understood why Nalinee would not let us fit anything earlier. Our 28-day wallet return forecast finally stopped pretending campaign days were organic.
I wanted more live critique in week two than the cohort could give. The holiday calendar handout and the memo template are still what I open before a QBR.