Field note · 22 July 2026
Before explaining drop-off, inspect the denominator
A funnel can calculate perfectly and still begin with the wrong people.
Product conversations often begin at the steepest bar. Analysis should begin one step earlier: who entered the calculation at all?
Check eligibility against the product promise
An install is not necessarily a signup opportunity. Returning users, unsupported regions, internal testers and people opening a deep link may all appear before the first account screen. Write the eligibility sentence in ordinary language, then compare it with the query.
Look for repeated entry
If the entry event fires on every screen revisit, one person can contribute several apparent attempts. Decide whether the unit is a user, device, session or attempt and explain why that unit fits the decision.
Respect the outcome window
Email verification and considered purchases may finish hours or days later. A same-session funnel can mislabel delayed success as abandonment. Plot time-to-outcome before choosing the window.
Compare releases and platforms
An event renamed on Android but retained on iOS creates a segment difference with no behavioural meaning. Event volume per active user is a quick warning, not proof; follow it with implementation checks.
Preserve the definition
Record inclusion, deduplication and window rules beside the chart. Future analysts should not have to reverse-engineer the denominator from SQL during a launch review.