Simple alternative to K-Means: each account is counted by backend activity signals (saves + Homeroom/account flags). GA4 has no per-user dimension, so this uses DynamoDB profiles only.
| Bucket | Users | % of Accounts | Definition |
|---|
| Segment | Daily Avg | % of Total | Definition | Boundary Rule |
|---|
User Clusters (PCA Projection)
Each point is one real user (equal 1,000-per-cluster sample of ~805K accounts), projected to 2D via Principal Component Analysis over 28 activity-only features. Colours are natural K-Means clusters. Grey lines are the cluster-separating Voronoi boundaries — the exact decision frontier between neighbouring cluster centroids in PC space.
PC1 Loadings
The linear combination of activity features that defines the x-axis.
PC2 Loadings
The linear combination of activity features that defines the y-axis.
Feature Importance (Eta-Squared)
Fraction of each feature's variance explained by the cluster assignments. Higher = stronger driver of separation.
Cluster Profiles (Normalised)
Normalised activity signature per cluster. Each cluster has a distinct behavioural shape.
Rule-Based Funnel
Static threshold segmentation for the engagement funnel. Newsletter runs as a parallel channel.
Weekly Trends
Engagement channel volumes over time (Q1 2026).
Segment Transitions
| From / To | Trigger | GA4 Signal | Intervention |
|---|---|---|---|
| Passive to Light | First community surface visit | Homeroom or T2T page view | Stronger CTAs on article pages |
| Light to Active | First card click | Discovery card click event | Improve card relevance and placement |
| Active to Contributor | Content creation | WP question/reply or upvote click | Lower friction for first contribution |
| Any to Newsletter | Opens an email | SendGrid unique_open | Cross-promote at engagement peaks |