Journal · 21 June 2025
Building a cohort table without a warehouse
Not every organisation has a warehouse. Plenty of analysts still live in CSV exports from a product database and a spreadsheet that groans. You can still practise user retention cohort analysis. You cannot pretend the last column is mature.
Start with two files: a list of people and their start timestamps, and a list of retained events with timestamps. Deduplicate people with a stable id. If you only have emails, say so; emails change. Bucket start dates into weeks you define in one cell at the top of the sheet so nobody “fixes” a week boundary mid-argument.
For each person, compute tenure week of each retained event. A pivot then gives the classic grid. The incomplete last period should be labelled “too young,” not coloured like a finished cell. Hide it if your audience will quote it anyway.
What you will miss without a warehouse: late-arriving events, identity stitching, and cheap mix splits. Those absences belong in the memo. A small honest grid beats a stolen dashboard screenshot from a vendor trial.
When the sheet becomes painful, that pain is the business case for the SQL lab — not a reason to skip the definitions.