A stable overall conversion rate can conceal a sharp decline for users on one operating-system version. Aggregate metrics blend changed and unchanged experiences, sometimes allowing a healthy majority to mask the cohort that needs help.
Useful segmentation follows plausible mechanisms. Look at app version, device capability, operating system, market and the path taken through the feature. Avoid slicing until random variation looks meaningful.
Before release, write down the segments most likely to respond differently. This makes the analysis faster and protects the team from inventing a convenient explanation after seeing the result.
Need a clear view of a live release? Tell us what your team is deciding.