A cohort study, week by week

1. Name the starting event

A cohort is a group of customers who share one dated event: first visit, first paid order, membership start, or another event you can find in the extract. If two departments use different starting events, the study waits until one is chosen. Mixing them produces a table that nobody can staff against.

2. Name the return event

Return is not “engagement.” It is a second visit, a second order, a renewal, or another event recorded in the same units as the start. Bundlelogicbase will not treat an email open as a return from a visit-based start unless you explicitly want a separate, labeled behavior review.

3. Freeze the files

You send exports. We log what arrived, which columns were empty, and which groups are too small to interpret. After the freeze date, new rows wait for a revision pass. This is how two people stop arguing from two different Tuesdays.

4. Build the period tables

Rows are the named groups. Columns are the later windows. Cells that cannot support a claim are marked thin or missing rather than filled with a decorative rate. Seasonal closures belong in an annotation lane under the same periods.

5. Sit with the table

The 90-minute session is for correcting operational context we cannot see in the file: a storm week, a price change, a roster that never made it into the extract. The final pack includes that context so the next quarter does not start from folklore.

See what the flagship engagement includes

Operators who keep notes locally can look at the desktop utility listing. Available in Veve