The Five Deadly Sins of Clustering

How to Avoid the Common Mistakes of Store Grouping

Best Practices from Kurt Salmon

Clustering is in fashion among retailers—and for good reason. This not-so-new concept, which groups stores based on shared demographics and customer purchasing patterns, can take on new value as retailers look to localization to differentiate themselves and improve performance.

The Robin ReportFor years, retailers have been limited in their ability to accurately cluster stores by three overarching hurdles that are now disappearing. The first hurdle, data storage limitations, can be easily overcome as the price of data storage continues to drop and retailers find themselves with the detailed transactional data required for clustering analysis. The second hurdle, analytical horsepower, is also no longer an obstacle, as most PCs are now capable of the iterative, detailed analysis required to analyze the data. Finally, lack of detailed consumer data required for developing truly descriptive store clusters is no longer a problem, since detailed demographic and psychographic information is now readily available from a variety of sources.

But even with the elimination of these roadblocks, some retailers are not maximizing the value of a solid clustering strategy. We have found that these retailers have access to all the necessary data, both internal and external, and possess the analytical tools to perform the analysis, but consistently limit the effectiveness of clustering by making five critical mistakes.

1. Failure to keep the end in sight

Many retailers do not consider the output of a clustering plan or how the outcome will work within their organization’s limitations when they begin the process. From the start, retailers must consider a number of factors, including whether their existing systems will be able to handle the changes recommended by the clustering analysis. As an example, will their planning and allocation tools be able to handle the addition of six to eight new store clusters? Additionally, will a retailer’s buying and product development teams be able to provide additional unique products to address the needs of the resulting clusters? Retailers also need to determine if the new clusters will fit within their existing processes and, if not, which changes need to occur to accommodate them. Finally, retailers need to establish financial goals for clustering, in terms of marked improvement in sales, inventory performance and margin. Without establishing what the expected levels of improvement will be, it is impossible to understand whether the investment in clustering is worth the effort.

2. Mistaking grouping for clustering

Many retailers think they have a clustering strategy when in fact all they are actually doing is grouping stores across a set of very subjective attributes. Most commonly, retailers will select attributes of temperature, volume or lifestyle and then assign stores to each based on their “gut” or some other subjective criteria. Clustering, on the other hand, usually incorporates more attributes than grouping, but the key is not quantity, but rather descriptiveness and process. A well-executed clustering strategy makes use of attributes that can clearly define the differences between individual stores, arming a retailer with the appropriate strategy for each store without their having to constantly examine stores at the individual level. Because a sound clustering strategy is based on an iterative analytical process supported by data and regression analysis, the resulting clusters are far more accurate and actionable. This strategy can include insights on which brands and products to feature in a given store as well as pricing, staffing, advertising and store design.

3. Starting with the wrong baseline assumptions

Although a given attribute may seem important, it may not actually define store differences. Starting with an incorrect assumption can prejudice the end result and create clusters based on attributes that are not relevant. Just because your target customer possesses a specific attribute doesn’t mean that attribute should be part of your clustering strategy. A regression analysis across all attributes reveals which are actually significant. For example, a retailer who focuses its product on the career woman may find that including career vs. casual in their clustering analysis results in weak store clusters, as all stores may be equally skewed toward the career customer.

4. Becoming overly dependent on transactional data

Transactional data describes who the current customer is and how they shop while leaving out potential or lapsed customers. For example, consider the case of a 500-store chain that has been experiencing consistently negative year-on-year sales growth. Using just current transactional data provides information on only the retailer’s remaining customers. This in turn could drive additional poor buying and product development decisions geared to address an incomplete customer picture. By also including both lapsed customer data and external demographic information, the retailer is able to see which customers they’ve lost and, more importantly, which ones they’ve never had. This will allow the retailer to develop a strategy to increase their customer base through the development of different products, implementation of new staffing plans and, potentially, the creation of new, unique store designs.

5. Assuming clustering needs to be done only once

Even after a clustering strategy has been put into place, a test mechanism is needed to measure the results and show areas for additional improvement. Every 12 to 18 months, a retailer should evaluate sell-throughs and margin to ensure growth aligns with the financial goals set at the beginning of the clustering process. If it doesn’t, the retailer should reexamine and adjust the clustering strategy. Over time, the insights produced by these results, new strategies and demographic shifts will all drive the need for reclustering. For example, it’s important to consider the changing age of a retailer’s customers: a retailer either has to change its products to fit an aging demographic; go after new, younger customers to remain relevant; or run the risk of a diminishing client base over time.

The case of the 500-store apparel chain shows that clustering, when done correctly, can produce a significant return on investment. The retailer had previously grouped stores based on two attributes: climate and ethnicity. Analysis found that over 30% of stores had been misattributed for climate because stores were assigned to groups based on opinion instead of historical weather data. The analysis also found that over 20% of stores had been misattributed for ethnicity. The resulting clusters, based on eight attributes, were 40% to 50% better correlated than the previous grouping effort. The clustering is expected to result in a 60 to 100 basis point improvement in full-price sell-throughs.

Clustering, when executed correctly, can help accurately plan everything from products and pricing to staffing and store design, without the added complexity of planning at the individual store level. To achieve a successful clustering strategy, keep the end product in mind throughout the process, use analytical tools to drive decisions, incorporate external data, and know that future reclustering will help further strengthen return on investment.

Christina Bieniek and Ron Fleischer co-authored this article. Together, they have more than 25 years of experience in retail and advising retail leaders. They can be reached at christina.bieniek@kurtsalmon.com and ron.fleischer@kurtsalmon.com. To learn more, visit www.kurtsalmon.com.