DotActiv’s new cluster optimization feature is almost here and will allow you to create clusters for your business, tailored to your market environment. This software and service package was designed based on the notion that retailers were clustering their stores based on subjective information rather than data-driven results.
In this article, we take a look at the data that the cluster optimization feature requires to cluster your stores effectively. This information is used as input criteria for the cluster analysis. DotActiv provides you with cluster data that you can analyse and leverage to target each of your clusters with a customer-centric assortment plan and associated strategies.
How will data be provided?
Built into DotActiv Enterprise, the cluster optimizer feature requires full database integration with your business. This allows your business or a DotActiv representative, should you select a software and service package, to access the different types of data required to conduct a cluster analysis.
What data is used in the setup process?
If you are working with the software and service package from DotActiv, the Cluster Development Manager and space planner will first conduct a customer scoping session. We use this meeting to gain information from the category buyer or retailer representative about the category. Key roleplayers, time period selection and the current state of clusters within your business will be used to create your store clusters. At this point, we will discuss the cluster development drivers or the input variables used in the cluster analysis to ensure that the cluster analysis is tailored to the goals of your business. Finally, timelines and project expectations must be provided to ensure the effectiveness of the clustering process.
Next, we move on to working with the DotActiv software and the Cluster Optimizer. You’ll have access to both the software and Cluster Optimizer regardless of if you have purchased a software or services package.
We use the following data as part of the criteria setup process to create your store clusters for a selected product category. You can treat the below as steps when collecting the necessary data.
First, you will need to select the category you would like to cluster. DotActiv has built the Cluster Optimization feature on the principle of category-based clustering. This is because differing customer purchasing patterns and behaviour are exhibited within different categories of the same store.
Selecting the right category to cluster is important because you will need to determine whether this process meets your business needs and whether you have enough store branches and varying customer demand to make this exercise feasible.Store-level data
Next, you need to select the stores you would like to cluster. Should all your stores follow the same format, you can perform the clustering exercise on all store branches. However, if you have varying store formats such as distribution centres that you may not want to include in the clustering process, you can exclude these from the cluster analysis.Product hierarchy data
You will also need to specify which product hierarchy level you would like to use to cluster your stores. You can select one field of either subcategory, segment, brand or barcode to complete the cluster analysis.
DotActiv recommends that you work with barcode-level data because this will allow you to analyse data for the fields up the hierarchy such as brand and subcategory. If you select barcode item detail, this means that products that perform in a similar manner across certain stores will be grouped together in the same cluster.Fact and period data
You will need to decide which fact fields you would like to use to cluster your stores. Fact fields are linked to performance data such as sales, units and profit. The fields you can use are limited to the availability of data on your database.
In this case, DotActiv would recommend using one to four fact fields as the fewer fields you use, the more targeted and accurate your cluster analysis becomes.
Next, you can select the time period you would like to use. Typically, you will need between six months and one year’s worth of data to conduct an accurate cluster analysis. This is because this approach avoids working with only seasonal trends or abnormalities that may affect the performance of your selected category.Market data
If you would like to add additional input criteria to your cluster analysis, you will need access to market-level data such as store region/province or store format. You can use this data to ensure your stores are clustered according to the market data selected as well as performance (Fact) and product-level data.
What data do you get out?
In terms of data analysis, this is the part of the process where the largest difference is realised between the software and service packages.
If you purchase the software or service package, the following information is provided from the Cluster Optimizer feature.Generated clusters grid
The first grid that is generated details the clusters that have been created by the Cluster Optimizer. This grid specifies the number of clusters that have been created, the applicable store codes, names, formats, sizes, locations and number of drops associated with each. All grids created by DotActiv software can also be extracted as a CSV or XLS format to be further analysed if needed.Store Mappings
The store mapping grids specify the changes made from the current cluster formats to the new cluster formats. This grid speaks directly to DotActiv’s Cluster Maintenance feature, which you can later use for assortment planning and space planning.Criteria Data
The criteria data grid details which information on your database was used to group your stores together based on the input criteria you selected. Each store is drillable where each product’s contribution can be analysed as well.Silhouette Analysis
The Cluster Optimizer feature provides you with graphs that explain how the optimal number of clusters was selected for your business. The algorithm analyses the ‘Silhouette Coefficient’ at each number of clusters up to the specific cluster cap/upper limit. The Silhouette Analysis aims to maximise the similarities of stores within the same cluster and minimise the similarities of stores in different clusters.
The number of clusters with the highest Silhouette Coefficient is, therefore, the optimal number of clusters for the specific product category.Linkage Distance Metrics
This measure specifies the difference between two clusters in terms of various linkage distance metrics. The user can analyse the percentage of the difference between two clusters to determine if they are happy with the results of the cluster analysis. This is also the first point of reference when determining whether two clusters are similar enough to be merged together.Composition changes report
This cluster data is provided where changes can be seen between previous clusters and the newly generated store groupings. This information is analysed from a cluster-level right down to product level and you can use it to analyse differences between clusters.
The next step of the process focuses on interpreting the information received from the Cluster Optimizer and combining it with meaningful market insights to profile each cluster and facilitate stakeholder understanding.
The reporting step of the process is only available with the service package of DotActiv.Cluster plan summary
The first section of the reporting process focuses on the original Cluster Optimizer setup and the criteria used to group your stores. This refreshes the buyer or retailer representative on the number of stores and products in the plan as well as the input criteria.Cluster optimization
Cluster optimization takes you through how the software calculated the optimal number of clusters for the specific product category. Recommendations will be made at this point on whether the current number of clusters is sufficient for the specific category and retail environment.Cluster composition
Cluster composition specifies the number of clusters that have been created, the number and code of stores in each cluster as well as the geographic location and number of drops associated with each. At this point, recommendations will be made relating to floor planning or splitting clusters into two if there is a large difference between drop counts in the same cluster. This needs to be done because the goal of cluster analysis is to create one customer-centric assortment plan for each cluster. However, if there is a large difference between drop counts of stores within the same cluster, this will need to be mitigated.Market profiling
To gain a holistic understanding of each cluster from a store level down to a product level, the stores in each cluster will need to be profiled. Stores are classified as either low, middle or high-income based on the predominant customer base of the area. This information is used to assign one initial profile to each cluster as either economy, mid-tier or premium.Brand profiling
Next, we move on to profiling the brands in the product category according to the average retail selling price. This step will require working in conjunction with the space planner or buyer of the category to determine the price points that will differentiate economy, mid-tier and premium brands.Consumer profiling
Consumer profiling has been added to this process to understand the purchasing patterns and demand drivers of the consumers who purchase the category. This will help DotActiv to understand and justify trends that are seen in each cluster as well as understand the type of consumer who shops the category.Cluster subcategory performance and comparison
Next, each subcategory is analysed by creating a category analysis deck. This analysis method allows our consultants to assess each cluster in terms of sales, units and profit contribution, cumulative contribution, average retail selling price, the Pareto Principle and more. We also complete a cluster mapping exercise at this point to visualise where each store is geographically located within the cluster.
Once each cluster has been analysed in terms of subcategory contribution, the clusters are compared against one another to highlight their similarities or differences in terms of subcategory performance. This information is linked to the market, brand and consumer profiling that was conducted to provide justifications for the results that have been generated.Cluster brand and product performance and comparison
Each brand and product is then profiled according to the average retail selling price and sales contribution. This allows DotActiv to determine the predominant profile of each cluster and name each as an economy, mid-tier or premium with all profiling considered.
Profile names can also change at this point to suit your needs. For example, terms such as family-oriented or Millennials may be more applicable for certain product categoriesCluster comparison and summary
Finally, each cluster is compared according to the average contribution per store which allows DotActiv and the retailer to analyse each cluster on a store level. At this point, final decisions about whether the clustering exercise has been successful are made and provisions for the next step of the process, range optimization, are set out.
DotActiv is developing a basic, intermediate and advanced level of the cluster optimizer that will be accompanied by a service offering provided by DotActiv’s Cluster Development Manager. Each level of the optimizer will require more advanced data types than the previous to provide you with more in-depth insights per product category.