Reviews

What people said after the memo, not during the applause.

These notes refer to specific labs on this site. Voices differ on purpose. One of them is lukewarm on purpose, too.

Retention Horizon forced us to stop averaging Monday into Friday. Our 14-day usage forecast for wallet top-ups finally stopped overshooting payday week. I still think the holiday lecture could have been a handout.

Arthit W., analytics lead · Bangkok
★★★★★

The event-dictionary workshop in week one of Retention Horizon was the unglamorous part I needed. We had three names for the same home screen. After that, the model almost felt secondary.

Rated after cohort 09 · studio desk

I joined Cohort Curves for Product Leads because I do not write SQL every day. The overlays made sense. I would not try Prophet after this — the instructor was clear it was the wrong next step for our series.

Ploy · Chiang Mai

Anomaly Detection in Daily Active Series taught me to leave a spike alone when a store feature went live. We used to “correct” those days and then wonder why the forecast drifted.

Client in travel super-app

Glimpse was enough for me. I only needed the weekday lab and one critique. The recorded pacing is slower than Horizon, which suited a single contributor on night shifts.

Sam, independent Android developer
★★★☆☆

The weekday-seasonality lab assumed cleaner event names than we had. I still finished the 7-day forecast, but I spent a weekend renaming Firebase events first. Worth knowing before you enrol in Horizon.

Anonymous · insurance app, Thailand

Case study · wallet reactivation

Two colleagues reviewing documents at a table

A Bangkok fintech sent three analysts to Retention Horizon after a reactivation campaign looked “successful” in DAU and failed in 28-day return. Their series mixed a new onboarding event with an older login fire.

In week three they split the events, marked payday Fridays, and forecasted return without the campaign days. The miss against actuals shrank from roughly a fifth of daily actives to a band their PM could live with. They did not attribute the whole gain to the course — a store listing change landed in the same month — but the memo became the artefact leadership asked for in the next review.

Case study · grocery evening peak

A grocery delivery desk took Atelier rather than Horizon because security would not let event samples leave the building. We sat in their war-room for nine working days.

The useful finding was not a fancier model. Their “global” evening prior peaked at 19:00. Bangkok orders kept climbing past 21:30. Once the prior moved, a 7-day volume forecast stopped under-staffing night pickers. They still run a simple seasonal baseline; they declined Prophet on purpose after reading our journal note on when not to use it.

Fresh produce on shelves in a grocery aisle