This thesis presents an implementation of cluster analysis on retail store clustering so that better cluster-based stock allocation plans could be effectively devised and efficiently executed. For the case study company, stock allocation is one of its key strategic decisions as improper stock allocation, especially during special occurrences with high sale volumes, may lead to loss of sales and so overstocking at some stores/clusters. Based on our initial investigations, we find that the current clustering technique is somewhat inefficient as it simply divides the stores into four groups with equal members based on store's sales performance. Besides, the coefficient of variation of allocated stocks in each cluster is comparatively high, around 30.1% 51.1%
To better improve the efficiency of current clustering operation, two more systematic clustering techniques have been therefore introduced and compared with the current technique, namely K-Means and Agglomerative clustering techniques. We find that both K-Means and Agglomerative clustering techniques provide clusters with much less coefficients of variations, about 9.5% and 9.3% respectively. Besides, the total differences between allocated stock target by store cluster and actual stock target by store are also improved from 17,818,056 units to 15,672,717 units and from 17,818,056 units to 15,830,644 units by these two techniques, respectively. When compared among these two new approaches, it can be seen that K-Means clustering technique outperforms Agglomerative clustering technique in terms of both coefficient of variation and total difference between allocated stock target by store cluster and actual stock target by store.