Reading drop-off as a product decision, not a chart

Funnel charts invite a mechanical story: people leak. Plug the hole. Usage goes up. That story is tidy and often wrong. Many drop-offs are choices — a price shown too early, a permission dialog that feels like surveillance, a Thai-language screen that still mixes English legal copy.

When we forecast usage, we ask which drop-offs are load-bearing. If 40% of new users leave after the wallet KYC step, a 28-day return forecast that ignores KYC policy changes is not a forecast. It is a wish that the product stays still.

Name the decision inside the step

For each major drop, write the decision the person faced: continue, delay, or leave. If you cannot name the decision, you probably instrumented a screen view rather than a moment. Session Depth & Feature Adoption Models spends a week on this distinction because adoption curves lie when “viewed feature” is not “accepted trade-off.”

Forecasts need a lever

A memo that only says “return will be 18% plus or minus a band” is incomplete. We want one sentence about a lever: delay KYC, change the permission copy, or stop counting pre-KYC opens as activated users. If no lever exists, say so. Leadership can still plan, but they should not pretend the number will move because the chart is red.

What this is not

This is not an argument against measurement. It is an argument against treating every descending bar as an engineering defect. Some drop-off is the product working: people who will never fund a wallet should leave before they hit support. Forecast the users you actually intend to keep.

If you are building that kind of memo inside a cohort, start with Retention Horizon or the shorter adoption desk listed in the curriculum.