1. Dictionary
List the events that define open, return, and the feature you care about. Collapse duplicates. Note what is missing. This is unglamorous and usually where the forecast was already lost.
Studio brief
Usage forecasting is a product decision wearing numbers.
App analytics can describe yesterday with great confidence. Forecasting asks what people are likely to do next week with the event stream you already have — and which product lever would change that path.
What we mean by usage
Usage is not a single DAU line. It is return, session depth, feature adoption, and the quiet users who open the app without firing the event you named “success.” We forecast the series a team will actually staff or budget against.
If your leadership argues about “engagement,” we ask them to pick one measurable horizon before any model is fit. Seven days for staffing. Twenty-eight for a campaign. Ninety only as a scenario.
A working sequence
2. Calendar
Mark weekdays, payday, Thai holidays, and known store releases. A model that cannot see Songkran will invent a story about churn.
3. Horizon + memo
Fit a seasonal baseline first. Only then consider heavier libraries. Write what would falsify the forecast. Take that memo to the roadmap, not the notebook.
Where this shows up in the curriculum
Retention Horizon is the full eight-week path through this sequence. Shorter desks isolate session depth, cohort curves, or anomaly handling. The journal collects field notes — including when Prophet is the wrong instrument.
If you already know the series you need to forecast, write to the studio. If you are still choosing a desk, start with the curriculum list.
Bring the ugly export.
We would rather see missing days than a polished dashboard. The curriculum is built for the former.