Telecom Churn--Demonstrating kmeans++ effectiveness

The example shows effectiveness of kmeans++ towards creating good clusters. When kmeans++ is used in selecting initial seed Centroids (through Weka's SimpleKMeans widget) the max Silhoutte Coeff for a data pt is 0.659 and min is 0.301. But when we use k-means widget with Random initialization of Centroids, the Silhoutte Coeff vary from 0.386 to -0.0.39.