Funnel window
The time a person is allowed to take from the first funnel step to the last before the report counts them as a drop-off.
See it
What it is
The funnel window is the maximum elapsed time allowed from a person's first qualifying step to the final step. A 7-day window gives each entrant their own seven-day clock; it is not a calendar week. Someone who starts Friday and finishes Tuesday counts, while a finish eight days later does not.
Choose it from the real decision cycle. Minutes may fit password reset, days may fit ecommerce, and weeks may fit a considered purchase. Run the same funnel at a few plausible windows before settling, because the gap between them tells you how much conversion is merely delayed.
Gotcha: recent entrants have not had a full chance to convert. Including people who entered yesterday in a 30-day funnel depresses the result through right censoring, not bad product behavior. Exclude incomplete cohorts or label them provisional, and write down whether a new first step restarts the clock.
Ask AI for it
Implement this ordered funnel in PostgreSQL with a 7-day conversion window. For each user, select the first [step 1], then the first [step 2] after it, continuing in order until [final step], and require the final timestamp to be no later than step_1_at + INTERVAL '7 days'. Exclude entrants after analysis_end - INTERVAL '7 days' so every cohort had the full chance to convert. Return unique-user counts and step-to-step rates, then rerun the query at 1, 3, 14, and 30 days as a sensitivity table.