How hard can it be to compute conversion rate? Take the total number of users that converted and divide them with the total number of users. Done. Except... it's a lot more complicated when you have any sort of significant time lag.

Erik Bernhardsson

Naive conversion rate calculations are misleading when significant time lags exist between user acquisition and conversion. Using a simple total-converted/total-users formula conflates cohorts with different amounts of time to convert, making growing products look worse than they are. The correct approach is cohort analysis — measuring conversion rate at a fixed time T for each cohort. Even better is the Kaplan-Meier estimator from survival analysis, which handles 'censored' observations (users who haven't yet had enough time to convert) and extracts more signal from recent data. A Python implementation using the lifelines library is demonstrated with startup exit rate data.

Conversion rates – you are (most likely) computing them wrong

An example – exit rate for startups 2008-2015

The right way to look at conversion rates – cohort plots

The 😎 way to look at conversion rates – Kaplan-Meier