Data Analytics: Foundations to Practice · Segmentation and Cohort Analysis
Why Aggregate Numbers Hide Important Differences
A single overall average can conceal sharply different patterns happening simultaneously within specific subgroups, and this concealment is often exactly where the most actionable insight in a dataset actually lives.
An aggregate metric, calculated across an entire population at once, averages out any real differences between meaningfully distinct subgroups within that population; a product showing flat overall engagement, for instance, might actually be growing steadily among new users while simultaneously declining among long-tenured users, two genuinely opposite trends that a single overall aggregate number would report as an unremarkable, misleadingly flat, unchanging line.
- An aggregate metric averages out real differences between subgroups; flat overall engagement might hide new-user growth offsetting long-tenured-user decline entirely.
- The aggregate view can actively misdirect a team's efforts, since the appropriate action for a hidden divergent pattern differs completely from the aggregate's own implied action.
- Segmentation breaks a population into meaningful subgroups based on a relevant shared characteristic, examining a metric separately within each one.
- Not every possible division is a meaningful segment; a segmentation scheme should relate plausibly to the metric and produce genuinely different, actionable subgroups.