Begin with a mechanism
Before opening the data, state why a group might behave differently: device constraints, route eligibility, prior experience or a product variation. This limits accidental storytelling after the numbers appear.
Protect against tiny samples
Report volume with rate, use consistent time windows and avoid ranking unstable slices. If the evidence is directional, label it that way and gather more data before committing a broad change.
Connect comparison to action
The useful question is not whether two groups differ. It is whether the difference points to a route condition the team can understand, test or improve without disadvantaging users.