Analytics platforms sample high-traffic data to manage processing load — meaning reports are built from a subset of sessions, not all sessions. The sampled subset may over-represent certain user types, devices, or time windows, producing averages that don't reflect what all users actually experienced.
Optimization decisions based on sampled data may improve metrics for the over-represented segment while ignoring the experience of the majority.
Sampled reports produce averages that misrepresent actual user behavior across the full population.
Analytics reports carry a sampling indicator (GA4 shows a lightning bolt icon, or the data threshold warning) but decisions are being made from those reports without adjusting for it. Segment-level numbers shift unexpectedly when date ranges are narrowed — a sign the sampled dataset is not stable across windows.