Input Features (normalised 0–1)

• TP01 Satisfaction — overall satisfaction percentage
• Complaints — CH01_1 (Stage 1 complaints count)
• Arrears Rate — current_arrears / rents × 100
• Cost per Unit — management / units

Step 1 — Normalisation

Each feature is min-max normalised across all organisations: (value − min) / (max − min). This creates a 4-dimensional vector per org.

Step 2 — K-Means Clustering (k=4)

The algorithm randomly initialises 4 cluster centroids, then iterates:

• Assign each org to the nearest centroid (Euclidean distance in 4D space)
• Recalculate centroids as the mean of assigned members
• Repeat until assignments stabilise (max 50 iterations)

This runs 10 times with different random seeds. The run with the lowest total within-cluster distance is selected.

Step 3 — Auto-Labelling

Each cluster is labelled based on its average TP01 and cost per unit relative to sector averages:

• Efficient & Stable — above-avg satisfaction, below-avg cost
• High Satisfaction — above-avg satisfaction, above-avg cost
• High Cost / Low Satisfaction — below-avg satisfaction, above-avg cost
• Under Pressure — below-avg satisfaction, below-avg cost

Data Requirements

Organisations must have valid TP01, unit count, and rents. Minimum 4 organisations required.