How to compare stores and their performance if underlying store preconditions differ
Every country in today’s world has a plethora of retailers and stores. While some of these establishments are in bustling shopping districts in major cities, others can be found in small towns and villages.
Let us consider the following scenario: Let us assume you are a retailer with hundreds of stores offering the same products, scattered across a country’s major cities, towns, and villages. Keep in mind that each of these stores’ different geographic location has its own unique set of characteristics: such as differences in customer purchasing behaviour, population density, transit patterns, and even co-tenancy, to mention a few. The begging question however, is: Do you think comparing the performance of all these stores would be a reasonable, if not pointless, thing to do, given the above disparities? Let’s find out together!
In this article we will answer this question by exploring how data science techniques can help retailers or managers with the following tasks:
- Build strategies to boost their store sales.
- Compare shops in a meaningful way to identify which ones outperform the others under same store preconditions.
For driving sales growth: A personalisation strategy should be applied to stores
To begin with, personalisation is the latest solution for retailers today. In an effort to get to know customers better, retailers collect and process huge data sets. However, the personalisation should not be only limited for customers but also be adopted as a strategy towards growing store sales. One should have a look at open street maps, zip code, public transportation around the store and many more data. To achieve strong store-level growth, a ‘one-size-fits-all’ approach no longer works, and it is important to customize a strategy for each store or group of certain stores.
Introduction of store clustering
One way of comparing stores is to use clustering methods. In particular, store clustering is the grouping of stores, based on common store and demographic characteristics. There are two types of store clusters: performance and non-performance based. Performance based store clusters are grouped according to how they perform. For example, store locations with similar sales performance would be placed in the same store group. Non-performance based clusters consider store characteristics such as store size and/or store type and they even can consider customer demographics such as ethnicity, income level, age group, fashion preference etc.
Applying clustering methods in a meaningful way
One approach to our scenario is a two-level store clustering.
At the first level, a non-performance clustering is applied to group stores together with similar preconditions. This process includes variables which do not change or barely change over time. Examples are store locations, population of the corresponding regions, purchasing power data, parking availability at stores, number of shops in the surrounding area and many more. Once the shops are grouped in clusters corresponding to similar preconditions, we can compare shops in a meaningful way.
At the level two clustering, we apply the sub-clustering (meaning the grouping of shops in each cluster separately) whereby we can build performance-based clusters. Consequently, we can analyse why some shops perform better than others by adding variables where specific strategies can have an impact. Examples of such variables are customer frequencies, google ratings, number of employees and many more.
Collaboration of data science and business validation will result in successful clustering
To be able to achieve a successful clustering we should combine data science techniques and business wisdom. This can involve four key steps:
- Variable Collection
The implementation process starts with creating a list of possible variables that can be used to cluster the stores. This is typically based on the availability of accurate data.
- Variable Selection
The selection is based on the business importance of variables. As for our scenario, selection is based on which variables should be used for the level 1 and which for the level 2 clustering.
- Clustering Modelling
Once the variables for clustering are selected, the next step is to build and execute a clustering model, or even different clustering models. There are many different
For an exhaustive list, see https://link.springer.com/article/10.1007/s40745-015-0040- 1.
Generally, 5-7 clusters are great as a result where each cluster can also be justified with a business definition.
- Business Validation
Evaluation of identified clusters is subjective. Nevertheless, it is the most important step to establish that the clusters really differ in their performance and thus may require a domain expert.
What is there to take away?
For any large retailer, insights into the drivers of a store success provide a huge competitive edge. However, retailers can no longer adopt a uniform strategy for all their stores spread across the country. To continue their success narrative, stores in different regions with varying preconditions will need unique methods (personalization strategy) that can be repeated at other stores with similar features. Keep in mind, however, that when the topic of store clustering comes up, a lot of people are usually concerned about two key issues: which store clustering approach to adopt and how to successfully implement it.
Having said that, do you have any reservations about implementing a personalised clustering strategy for your stores?
Lave us a message and we would help you get started the right way!