November is the Month of Learning at the Association of Consumer and Retail Professionals (ARC), and we’re readying our next training course for space planners: Store Clustering Through Store Level Data and Tools. Space Planning Community members, be on the lookout for this in-depth course in the coming weeks.

To set the table for this exciting new learning, and to introduce this important topic for space planners, we present some FAQs on category specific store clustering and its applications, as well as some food for thought for members seeking to begin clustering or considering implementing new strategies.

What is Store Clustering?

In simplest terms, clustering is the process of grouping stores by shared characteristics for category optimization purposes through demographic, geographic, and other data.

Store clustering has its roots in the sound category management principles of the CatMan 3.0 process, and a shopper-tailored version of the four Ps (price, placement, product, promotion) are the outcomes of any successful clustering strategy.

Done correctly, clustering can offer the advantages of improved assortment, better pricing decisions to increase sales and profit, smarter, more-focused marketing to engaged potential buyers, and improved inventory management. Armed with this data, space planners can make optimized layout and display decisions, too.

Finally, clustering across a series of “test” stories affords the opportunity to try out new strategies to see if they will scale at a broader level.

Common clustering factors include:

Geographic Location. This is commonly used and is table stakes in category management and space planning. Nevertheless, it is a fundamental skill that must be correctly applied, lest the wrong products end up on shelf in the wrong markets, such as road salt in stores in warmer-weather climates.
Store Size. A smaller store could have a different clientele than a bigger store, and with less space, in-store assortment is more limited.
Customer Demographics, including age, income, and other factors.
Sales Performance. It stands to reason that better-performing stores will have different assortment, and that the reasons lower-performing stores struggle will come under the microscope.
Store Format. A convenience-oriented store, for example, will be heavy on grab-and-go stables.
Product Preferences. This can be viewed through both the lens of shoppers and brands. Some brands, for instance, aren’t as likely to be in smaller stores. And above all, shopper preferences on products are the drivers on what appear on shelves.

Martha Johns, director of category leadership at Mitchell Grocery, a CMA member and a wholesale distributor for nearly 200 combined independent and corporate stores in the Southeast, has implemented clustering strategies both in CPG with Kraft Heinz and in her current role. In an interview with The Space Planning Community, she shared her insights into some of the variables she’s considered in the many clustering tests she’s run throughout her career.

“When you’re looking at being able to look at pricing, you usually cluster by geography because you want to compare to the same competitors in the same marketplace,” she said. “When you’re doing clustering for the purpose of planogramming, you usually match the demographics of the shoppers — mainstream, premium, and value, along with the average household size — as close as you can. That’s what we’ve been in the process of doing, and we’ve used Spectra data to do that, to cluster the stores in multiple ways at the same time, to sort of weight that.

“When we’re doing weighting for planograms to cluster, we typically try to look at it in terms of mainstream for the majority, and then which stores are premium and which stores are value-focused, and then look at value differently depending on whether it’s a lower-income value or whether it’s a higher income with larger families that are looking for larger-size packs. So the household size makes a difference as well.”

For more on the steps of CatMan 3.0, members should view Course 102: Category Management Overview, in the ARC Training Library.

What is multivariate clustering?

In multivariate clustering, numerous factors are considered in determining how to group stories within a category.

For instance, a store may look to cluster for variables including income, store-size, and geographic factors. The appeal of multivariate clustering is that done correctly, it can provide unique, layered insights simply not possible by looking at just one variable. However, it is a complex process involving statistical analysis and software, ranging from Microsoft Excel to more complex systems to both conduct the clustering analysis and implement post-clustering strategies.

No matter your clustering tools, having a good data set over a relevant period of time — and being able to accurately assess the output from the tools — is essential, according to Kent Rodina, director of merchandise presentation at O’Reilly Auto Parts and a Space Planning Community board member.

“(Clustering tools help) you identify, based upon the demographics and sales that you have loaded into the tool, what measurables are driving differences in those stores,” said Rodina, who has experience using clustering analytics through Blue Yonder.

Clustering analytics, he noted, gives a view into “the stores that are selling the assortment differently to help determine what’s unique about those stores. What demographics are unique about those stores? So that requires some history that you have to pump into such an analysis.”

The right analysis tool will depend upon the complexity of the analysis being considered. Members with category management and space planning tools should not hesitate to reach out to their solution-provider partners for assistance to ensure the right use of tools. The CMA’s Solutions Marketplace can be a wonderful resource when seeking help. Asking other members for insights into what has worked for them can be illuminating, too. The Space Planning Community team is always happy to connect members with other members to have a professional-to-professional discussion which can provide the insights needed to move a project forward by bringing new perspectives to the table.

What are some of the key steps involved in the execution of multivariate clustering?

There are five common steps in proper execution of multivariate clustering, and it’s important to bucket them in terms of demographics and business opportunity. The steps are:

Demographics

Identify the Store-Level Demographic Profile (SLDP). This step compares the demographics of a given store against the market at-large.
Identify the Product Demographic Affinity Profile (PDAP). In this exercise, product sales and total sales by demographic are analyzed against sales to the entire population buying the product.

Business Opportunity

Calculate Product Demand Potential (Demand Index). This involves the creation of a Demand Index, which measures product demand potential against the product’s store sales, or All Commodity Volume (ACV).
Calculate Current Sales Performance: The focus here is to attach a dollar amount to a Sales Opportunity Gap (or Overperformance) with this calculation: (Demand Index * Product Actual Dollar Sales / Sales Index) – Product Actual Dollar Sales.
Create Clusters, Emphasizing Relevancy and Opportunity: This key step takes the analysis above and begins to create the clusters based on demand and performance, with an emphasis on the opportunity to begin tailoring assortment to meet obvious demand gaps.

What should drive the decision to cluster: sales date or insights data?

Both data sets have their value in decision of whether and how to cluster, and the experts interviewed by the Space Planning Community said the decision to lead with sales or insights comes down to the problem that clustering is intending to solve.

“It depends on the perspective, because sometimes when you’re doing clustering, it’s a competitive defensive measure in order to fight a competitive condition or a particular competitor that’s pulling volume away,” Johns said. “But sometimes what you’re doing is trying to mirror, to better meet the needs of a shopper. Clustering works well for both.”

In Rodina’s view, sales data is best to use the following cases:

The category or categories being analyzed are static.
In straight-forward assortment decisions, where the retailer has resolved to stop carrying the worst-selling items.
If a decision is needed quickly, sales data’s ready availability makes it tough to top.

Rodina’s preference, however, is to use insights data when beginning to make a clustering hypothesis.

“There is nothing wrong with using the sales metrics as a starting point,” Rodina said. “I have just found that when doing it that way, I am generally analyzing the sales data to try to understand why and to what level a specific bit of market/demographic data is influencing a category. I then try to define the ‘why’ through trends or market assumptions I already have about the category — ‘Oh, it’s because it’s the colder time of year’ or ‘A different portion of the community likes the best portion of the assortment more than the good portion [in a ‘good-better-best’ analysis].

“At the end of the day, I am using sales data to make educated guesses. By approaching it the opposite direction and starting with the hypothesis, many of which are already being driven by plenty of category/store insights and sales data, then testing that hypothesis, it’s just a matter of if the sales data performed better or worse during my testing period — and I am not spending my efforts trying to interpret as much of the ‘why’ behind that.

“Using the sales data as the main driver is going to base my results on historical customer decisions. Testing a hypothesis and measuring that data will show me how my customers are responding now based upon the specific change I made.”

Bringing it All Together

This article is intended to provide a general framework around clustering fundamentals. In future installments, we will explore common hurdles and opportunities around clustering, with insights from practitioners on implementing winning strategies. If you have suggestions on clustering insights and how-tos you would like to see, drop us a line.