Cluster analysis
Partitions respondents into mutually exclusive segments that are internally homogeneous and externally heterogeneous.
Inputs and outputs
- Survey inputs: Likert scales (1 = definitely agree to 5 = definitely disagree), volumetrics and binary questions.
- Quality metrics: Determination (1 to 100%), Calinski-Harabasz Index, Silhouette Score and Inertia.
- Cluster label renaming: click the default cluster headers ("Cluster 1", "Cluster 2") to assign custom persona titles (for example "Value Shoppers", "Tech Enthusiasts") for presentations.
Metrics for cluster quality
Metric | Purpose and interpretation |
|---|---|
Determination | Finding the optimal number of clusters that best represents the data structure. Shown on a 1 to 100% scale; higher is better . If Determination scores are similar, the data may be too similar to create more clusters; more varied data produces more cluster group recommendations. |
CH-Index (Calinski-Harabasz) | Measures separation between clusters and compactness within clusters. A higher CH-Index means better-defined, more distinct clusters, and helps identify the optimal number of clusters. |
Silhouette Score | Quantifies how similar data points are to their assigned cluster compared with other clusters. Used to determine the optimal number of clusters and evaluate separation. |
Inertia (within-cluster sum of squares) | Measures compactness. Lower inertia means points are more tightly grouped around their cluster centroids. |
Saving and viewing Cluster Analysis results
You can save multiple cluster analysis results to review them at a later date within the same report.
- After running an analysis, click the "Save Cluster Analysis" button. Provide a name and an optional description, then click Save.
- To retrieve past runs, click "View cluster analysis" to open a modal listing your saved iterations, and click Select to load it onto the canvas.
Use cases and benefits
- Identifying differentiated audience segments: split a group such as a product category into mutually exclusive groups of like-minded people, based on responses to lifestyle and attitudinal statements.
- Understanding buying behavior: find groups of buyers and analyse each group separately on favorite stores, brand loyalty, price willingness and purchase frequency.
- Identifying new product opportunities: cluster brands and products to find competitive sets. Brands in the same cluster compete more with each other, which exposes gaps in current offerings.
- Attitudinal profiling: categorize people by attitudes, beliefs and preferences (psychographics) rather than only demographics or behavior, to build more personalized campaigns.
- Media industry segmentation: segment audiences to create targeted content or advertisements, maximizing engagement and ROI.