2216 customers were segmented into 4 groups using spectral clustering (kernlab) applied to income, wine spend, meat spend, recency, and web purchases. The highest-income segment averages $79,965 per year, making it the prime target for premium marketing investment. The largest cluster contains 623 customers. Review the cluster profile table for full behavioural signatures of each segment.

Five features were used for clustering: income, wine spend, meat spend, recency, and web purchases. Average annual income across the customer base is $52,247. Wine and meat spending show high standard deviations relative to their means, indicating strongly skewed distributions with a small segment of heavy spenders. Recency (days since last purchase) is roughly uniform, confirming it provides independent segmentation signal.

The heatmap shows pairwise Pearson correlations between all five clustering features. The strongest correlation is between Income and Meat Spend (r = 0.58), suggesting these two features share a common customer dimension. High income-spend correlations confirm a natural premium segment exists in the data. Near-zero correlations with recency show it provides independent information, helping spectral clustering separate recent vs lapsed customers.

Spectral clustering produced 4 customer segments from 2216 total customers. The largest segment is Cluster D with 623 customers (28.1% of total). Balanced cluster sizes indicate the algorithm found meaningful natural divisions in the data. Highly unequal sizes may suggest one dominant customer archetype with a long tail of niche groups.

Wine spend is one of the strongest differentiators between customer segments. Cluster A has the highest median wine spend at $602, compared to $28 for Cluster D. Wide boxes (large IQR) within a cluster indicate heterogeneous spending behaviour, while narrow boxes signal a tightly defined consumer type. Clusters with consistently high wine spend are prime candidates for wine loyalty or subscription campaigns.

Cluster A has the highest average annual income at $79,965 compared to $37,853 for Cluster D — an income gap of $42,112. Income is a strong proxy for lifetime value and willingness to pay for premium products. Targeting Cluster A with premium or high-margin offerings is likely to yield the best return. Lower-income clusters may respond better to value promotions or deal-based incentives.

Each row gives the complete demographic and behavioural profile for one spectral cluster. Cluster A — the highest-income segment — spends an average of $656 on wine, purchases via web 4.3 times, and last purchased 51 days ago. Average Recency across clusters: clusters with low recency values are more actively engaged and should receive retention campaigns, while high-recency clusters are lapsing and need win-back offers. These profiles translate directly into distinct marketing briefs for each segment.

Each point represents one of 2216 customers, plotted by annual income (x-axis) and wine spend (y-axis), coloured by their spectral cluster assignment. Spectral clustering can identify non-convex shapes in this space — curved or crescent-shaped groups that K-means would split or merge incorrectly. Clusters that overlap in this two-dimensional view may be well-separated in the full five-dimensional feature space. The 4 distinct colours correspond to the 4 discovered customer segments.