Peer groups for organisational learning: Clustering with practical constraints.

Area of Science:

• Data Science
• Business Analytics
• Organizational Behavior

Background:

• Peer-grouping is vital for organizational learning, policy, and benchmarking across industries.
• Traditional statistical clustering for peer groups often overlooks crucial business constraints like size and stability.
• High-dimensional data and complex variables in clustering can hinder understanding for non-statistical stakeholders.

Purpose of the Study:

• To develop a methodology for applying business constraints to clustering solutions.
• To enable decision-makers to balance statistical quality with business requirements.
• To enhance the interpretability of complex clusters for diverse audiences.

Main Methods:

• Developed constrained clustering methodology to integrate business rules (e.g., group size).
• Utilized tools to identify distinguishing features within peer groups.
• Created visualizations to explain high-dimensional clusters to non-statistical users.
• Applied constrained clustering to a noisy, high-dimensional dataset over two years, focusing on cluster stability.

Main Results:

• Successfully applied constrained clustering, satisfying size constraints (≤100 members) on a real-world dataset.
• Demonstrated a stable clustering solution across two consecutive years.
• Observed a consistent trade-off between statistical goodness-of-fit and inter-year cluster stability.
• Effectively communicated complex cluster features to stakeholders with varying statistical expertise.

Conclusions:

• The constrained clustering approach effectively integrates business constraints into peer-group formation.
• The methodology provides a practical balance between statistical rigor and business applicability.
• Visualizations and feature identification tools improve the communication and understanding of clustering results for decision-makers.