Store Clustering

Analytical Processes and Considerations
The opportunity to pursue advanced clustering initiatives is due to the elimination of technical barriers and costs. In consulting firm Kurt Salmon’s “The Five Deadly Sins of Clustering,” authors Christina Bieniek and Ron Fleischer note that retailers can economically store transactional data, leverage the analytical horsepower of widely available software, and access detailed demographic and psychographic consumer information from a variety of sources. Retailers can use this combination of information and technology to accurately cluster stores.

The process begins with collection of the appropriate data. A well-defined and properly executed clustering strategy uses attributes that clearly delineate the uniqueness of stores and shoppers. Fundamental data categories and potential elements include: the shopper (purchase history and demographics such as age, ethnicity, lifestyle, and income); the store (revenue, profitability, and size); the local area (demographics and climate); and location (nearby competitors, co-located stores, geography, and area type).

Of these elements, shopper data is very important. “Shopper data helps retailers move beyond simple clustering and managing to the middle,” notes Hauptman. He recommends using loyalty data because it accurately aligns with actual behavior and needs to develop localized solutions.

It is also important to capture information about potential shoppers, according to Allison Jones-Farmer, C&E Smith Associate Professor of Business Analytics at Auburn University. “Using only data from existing customer transactions can lead to missed opportunities,” she states. “It can be useful to merge transactional data from existing customers with demographic information and other external data to capture the full potential of the market. This analysis may lead to different store clusters and allow an organization to attract a broader customer base.”

After the data is gathered and prepared, retailers can choose from a variety of tools. One option is to use proprietary algorithms, technologies, and analytical processes from companies like Willard Bishop and Dunnhumby to analyze consumer data for store clustering. Off-the-shelf packages are also available, notes Jones-Farmer. “Many software companies have built specialized store clustering modules such as Oracle Retail Assortment Planning and SAS Intelligent Clustering for Retail.”

Finally, the cluster analysis output creates groupings of stores with similar customer bases. These groupings can be used to develop and create product mixes, placement, and distribution that meet the needs of the cluster stores and attract more customers without the need to reinvent for individual stores.

Managers should not think store clustering analysis is a risk-free activity—there are numerous issues that can derail the initiative. First and foremost, the data must be accurate and impartial. “Clustering algorithms can be particularly sensitive to outliers or extreme observations in a data set,” says Jones-Farmer. “Thus, careful preparation and cleaning of the data is required prior to jumping head-first into an analysis. Many heartaches and poor decisions can be prevented by a careful pre-screening and visualization of the data prior to clustering.”

Bieniek and Fleischer’s article discusses additional pitfalls, including an overreliance on point-of-sale data; prejudicial assumptions that drive the use of irrelevant attributes; failing to use a group of attributes that clearly define meaningful store groups; and, belief that clustering is a one-time effort. Most importantly, retailers cannot implement a strategy without considering the implications: clustering will create complex changes in purchasing, planning, and allocation that retailers must accommodate.