Introduction
A cluster is a grouping of stores by similar sales patterns, based on customer need. These clusters can then be used by assortment optimisation to generate customer centric ranges by cluster.
The formation of clusters is more robust with more data, so a data period of
- 1 year should be used and will provide a good base.
- It is good practice to review clusters on an annual basis, there may be slight changes due to a store’s demographics changing over time.
- Periods of change that affect retail (e.g. technological change to a sector) would also be a key time to review the clustering.
Working with clusters
The goal of the clustering module is to produce commercially viable groups of stores with distinct characteristics that the assortment module can use to build customer centric ranges. The Dashboard allows the user to then understand the difference between each cluster and explain their characteristics to stakeholders.
Producing commercially viable clusters requires the user to review the practicality of the clustering schemes. Clusters containing 1 store or only a few stores, could indicate a distinct customer profile that requires a bespoke range.
How to select the best clustering scheme?
However, balanced against this, the workload involved in creating store specific clusters needs to be considered;
- Users need to remember that each additional cluster also needs to be maintained with range updates throughout the trading season.
- Users should also consider that because each cluster may require a small, medium and large planogram, the workload could be multiplied threefold.
The user needs to assess these clusters to decide whether they are worth keeping separate or moving to join a larger cluster.
As previously described, the first cluster scheme in the list is mathematically optimal.
The user needs to assess the strength of the cluster scheme options, weighing up how distinctive different store groups are versus the increasing workload required as additional clusters are added.
Within the dashboard for a cluster scheme, the user can look at the differentials to see which attributes are positively correlated with each particular cluster and also which attributes are negatively correlated with each cluster. Geographical elements can also be highlighted.
With this information, the user could build an infographic to summarise the clusters to stakeholders.