A customer cohort groups people by when they first bought. Start with a FIXED expression that finds each customer's first order across the full dataset. Then calculate the number of calendar periods between that date and each later order. These periods become the columns of a retention table; the acquisition cohort becomes its rows.
Count distinct customers in each cell. Counting order lines would reward customers for buying several products in one order. Divide the returning count by the cohort's original customer count, which stays fixed as you move across the row.
Compare equal ages. The 2025 cohort has no observed year one in data ending in 2025, so its missing cell is unobserved, not evidence of zero retention. Calendar-year cohorts also give customers different amounts of time in year zero. Monthly cohorts reduce this difference in observation time.