Mirror of Oracle documentation

Converted for search and offline reading. Authoritative source: Oracle. Diagrams and some complex tables are simplified — check the PDF when in doubt.

4 Advanced Clustering

This chapter describes the Advanced Clustering Cloud Service module.

Introduction

Advanced Clustering is an enterprise-specific clustering solution that uses data mining to create store groupings at different product levels using a variety of inputs. These inputs include performance data (sales dollars, sales units, and gross profit), product attributes (brand, color, and size/fit), store attributes (climate, store format, size, and servicing distribution center), third-party data such as demographics (income, ethnicity, and population density), and customer segments.

The application’s embedded science and automation helps you to identify unique patterns within your data that you can use to create the necessary customer-centric and targeted clusters. These can be used by the assortment planning, allocation and replenishment, pricing, and promotion processes.

It optimizes clusters in order to determine the minimum number of clusters that best describes the historical data used in the analysis and that best meets your business objectives, which you define during the design of your clusters.

You can use Advanced Clustering to execute localized or customer-centric assortments and for pricing. In addition, the application can help you when forecasting, for example, if you want to cluster stores based on similar seasonal patterns. You can also use the application for allocation, by clustering stores based on similar selling patterns.

Features

The key features of Advanced Clustering Cloud Service include:

  • Scenario-based cluster generation, based on store or product attributes, customer segment profiles, or performance.

  • Three-step cluster-generation process.

  • What-if capabilities that can be used to create multiple clustering scenarios and then measure them against one another. This can help ensure that the most appropriate clusters are used by the applicable planning and execution processes.

  • Automatic ranking of cluster scenarios to support what-if comparisons. Recommendations for the optimal cluster scenario and number of clusters are provided.

  • Dynamic nesting of clusters, in which nested or mixed attribute clusters are created based on multiple attributes, performance data, and customer segments.

  • Two types of algorithms are used.

    • Proprietary BaNG (Batch Neural Gas) algorithm for convergent cluster parameters

    • K-means approach for creating clusters in a hierarchical manner, which automatically determines the best attributes to split into an additional cluster.

Table 4-1 Cluster Criteria Overview Tab

FieldDescription
NameThe criteria ID and user-assigned name of the cluster.
Cluster ByA predefined group of attributes that include Consumer Profile, Product
Performance, Store Attribute, Product Attribute, and Mixed Attribute.
These criteria types are sets of attributes. For example, store attributes
are the properties of a store. These properties can include ethnicity,
store format, and store size.
Created ByThe name of the user who created the cluster.
Last Updated ByThe name of the user who most recently updated the cluster.
Last Updated OnThe date when the cluster was most recently updated.
StatusCreated, Ready for Approval, Completed with Errors, Approved,
Rejected.
Period CountThe number of calendar nodes defined for the criteria. Hover over the
count in order to see a list of the calendar keys associated with the
criteria.
Merchandise CountThe number of merchandise nodes defined for the criteria. Hover over
the count in order to see a list of the merchandise keys associated with
the criteria.
Location CountThe number of location nodes defined for the criteria. Hover over the
count in order to see a list of the location keys associated with the
criteria.

Clustering Criteria

The following clustering criteria (which are also called “Cluster by”) are the defaults:

Consumer Profile

Cluster stores based on the similarities in the customer profile mix whose members shop in the stores or trading areas. These clusters form the basis for additional analysis that can provide an understanding of which customers shop in which stores and how they shop. Information from market research firms such as the Nielsen Corporation can help retailers develop customer profiles. Such information can be provided via a data interface.

Location Attributes

Cluster stores based on how shopping behavior varies by store attribute. In combination with the profile mix, this provides an understanding of demographic details such as income level, ethnicity, education, household size, and family characteristics. Such knowledge can help the retailer to make assortment and pricing decisions. By analyzing cluster composition and studying business intelligence, the retailer can make informed decisions based on shopper demographics.

Product Attributes

Store share is generated based on product attributes. The store clusters produced can be used in an assortment. In this type of cluster, stores with a similar share of sales for one or more attributes are grouped together. For example, for the product coffee, stores can be differentiated by the sales patterns for premium, standard, and niche brands. The percentage of each store contribution is calculated using Sales Retail $ for each product attribute value to

ee

Table 4-2 (Cont.) Pop-Up Details

Field NameDescription
Merchandise TypeThe merchandise type.
Scenario CreatedThe number of scenarios created for the cluster.
Scenario ExecutedThe number of scenarios executed for the cluster.
Location TypeThe location type.
Parent Cluster LevelThe name of the ancestor cluster that has been further clustered.
All Cluster Criteria Scenario List

This displays the scenarios for the selected cluster criteria in the All Cluster Criteria tree.

Table 4-3 Scenario List

Field NameDescription
NameThe name assigned to each scenario that has been created for the
cluster.
StatusCreated, Ready for Approval, Completed with Errors, Approved,
Rejected.
User PreferredIndicates whether or not the user prefers the cluster.
System PreferredIndicates whether or not the system prefers the cluster.
# of AttributesThe number of attributes that were used in the cluster.
Max. # of ClustersA user-provided value for the maximum clusters centers that the
clustering process should consider.
Cluster Criteria

In this pop-up, you define the initial clustering parameters for the cluster criteria of a new cluster. Note that multiple hierarchies are supported in order to facilitate comparisons between clusters. For example, you can compare clusters for the market and retail location hierarchy.

Figure 4-6 illustrates how to use a simple approach to clustering by selecting attributes from a Cluster by. For example, you can select the performance Cluster by and generate clusters using store sales units or revenue.

a

Table 4-4 (Cont.) New Cluster Definition

Field NameDescription
Cluster ByA predefined group of attributes that include Consumer Profile, Product
Performance, Store Attribute, Product Attribute, and Mixed Attribute.
These criteria types are sets of attributes. For example, store attributes
are the properties of a store. These properties can include ethnicity,
store format, and store size.
MerchandiseOnce you choose the merchandise level for the cluster, you must select
the hierarchy type, the hierarchy level, and the hierarchy node. These
are specific to the merchandise level you select.
LocationOnce you choose the location level for the cluster, you must select the
hierarchy type, the hierarchy level, and the hierarchy node. These are
specific to the location level you select.
TemplateSelect by name a predefined template that can be used to create a
cluster hierarchy.
Effective Period

You can define a time interval for the cluster by either choosing a period from the list provided or by selecting a start date and an end date.

To define the Effective Period, you select either Planning Period, Fiscal Period, or Select Date:

Table 4-5 Effective Period

OptionDescription
Fiscal PeriodIf you select this option, choose the period and the subdivisions of that
period from the drop-down lists.
Planning PeriodSelect from the range of values provided for the period. Planning
periods are user-defined buying periods for a season or a season
subset.
Select DatesIf you select this option, choose the start and end dates using the
calendar pop-up.
Summarization

Data summarization is available when you select the product performance Cluster by. Select the dimensions of the hierarchy (merchandise or calendar) to summarize the data and consider the dimension position in the clustering process. For example, when you use the category/ week sales data to generate store clusters, you can select the week dimension in the calendar hierarchy summarization. The clustering process considers all weeks (week1, week2 ,… week52) as attributes and clusters stores based on their weekly sales patterns.

Table 4-7 Template Display

PropertyDescription
Template NameName of template configured during deployment.
DescriptionDescription of template.
Cluster ByA predefined group of attributes that include Consumer Profile, Product
Performance, Store Attribute, Product Attribute, and Mixed Attribute.
These criteria types are sets of attributes. For example, store attributes
are the properties of a store. These properties can include ethnicity,
store format, and store size.
Hierarchy

A dynamic Cluster by hierarchy is displayed. For example, the template PE-ST-ST has a Cluster by hierarchy of performance/store attribute/store attribute.

Cluster By Primary Scenario

You see this when you select Cluster by in the Criteria panel when you are setting the cluster parameters or when you select Cluster by in the contextual area for the hierarchy.

The attributes configured and the primary scenario properties defined for the selected cluster are displayed. The attributes listed are those that are significant for the clustering defined during deployment.

The primary scenario is the default scenario defined during deployment. The following information is displayed.

Table 4-8 Primary Scenario

PropertyDescription
NameThe name of the primary scenario.
StatusCreated, Ready for Approval, Completed with Errors, Approved,
Rejected.
Maximum # clustersThe maximum number of clusters. The default value is 100. This is
used for analyzing the clusters.
Minimum # clustersThe minimum number of clusters. The default value is 1. This is used
for analyzing the clusters.
AttributeA list of the attributes configured during clustering.
Attribute weightThe weights associated with each attribute. This is used to calculate
distance.
Planning Period

This list displays the time period you selected for the cluster definition. This information is available only for planning periods, where it provides the start and end dates of the planning period. This content changes whenever planning period is selected in Effective Period when you are setting cluster parameters.

ee

Category Variability

By analyzing store variability, you can determine if it is worth creating store clusters for the selected categories in the selected location. Three sections are displayed.

A grid is displayed for the selected categories and the sales contribution for a selected location.

Table 4-11 Category Variability

PropertyDescription
CategoriesA list of the selected categories that are used for store variability
analysis.
VariabilityThe relative standard deviation of the stores in the category. A larger
value for the standard deviation indicates greater store variability for the
category. Such a category is a possible candidate for store clustering.
Index to averageFor a selected location, an indication of how the store performs
compared to the all store base. A value close to 1 indicates that the
selected location is similar to the all store base. If the value is lower or
higher, it indicates that the sales averages for the stores in the selected
location are different from the all store base and that you should
consider creating store clusters for the selected location.
Average store retailAverage store retail $ for the category for the selected location.
Average store unitAverage store units for the category for the selected location.
Positive/negative index to
average
The difference in value for the index to average for the all store base to
selected location. For example, a value of 1-index to average < 1 or a
value of index to average -1 > 1.

A graph is displayed for the index to average. This shows how the selected location performs compared to the all store base if the average sales metric is below, above, or the same when compared to the all store base average. A red color indicates a value below the all store base average. A blue color indicates a value above the all store base average.

A graph for standard deviation is displayed. This shows the standard deviation for the selected category. If the store value is greater than two standard deviations, then store clustering should be considered for the selected merchandise because the stores sales variability is sufficient.

» Summary \ Mame Cotes Holiday Season Region ME Chettiarby Product Pedsemance Merchandise §Frodect Hierarchy (Location Location Hierarchy CLE REGION Planing Petes (Crriimas10008Cotter(ew Years nested 33-Southeant Regen,2 - NechewestR. Cluster Criteria Summaryis Available Throughout The Entire Clustering Process, for user to review the initial ustering parameters for the current run.

Table 4-13 (Cont.) Scenario Definition

Field NameDescription
Max. # of ClustersSet the maximum number for the total number of clusters that can be
generated. The application determines the optimal number of clusters
during the generation process.
Min. # of ClustersSet the minimum number for the total number of clusters that can be
generated. The application determines the optimal number of clusters
during the generation process.
Exact # of ClustersIndicates that the exact number of clusters should be generated. The
application does not determine the optimal number of clusters.
Attributes

The Attributes table is used to define which attributes are included in the cluster criteria and the weights that should be assigned to each participating attribute. You can

  • search by attribute, attribute value, and attribute weight

  • assign equal weight to the selected attribute

  • assign weight to the selected attribute or attribute value

  • reset weights to default values

The following information defines the attributes that are participating or non-participating.

Table 4-14 Attributes

Field NameDescription
ParticipatingA check in this column indicates that the attributes participate in the
cluster criteria.
GroupsIdentifies the group.
AttributesA description of the attribute.
WeightsThe weight assigned to the attribute. All participating attributes can
have the same weight or each participating attribute can have a unique
weight. The total of all the weights must add up to 100 percent.

The Attributes toolbar, shown in Figure 4-17, includes the following functionality:

Table 4-15 Attribute Toolbar

FunctionDescription
Action menuResets the weights to the default value overrides that the user provided
during configuration.
Weight statusProvides the weight validation status. If the weights do not add up to
100 percent, then a warning is displayed and the scenario cannot be
executed.
Include or exclude attributesAny attribute with a weight equal to zero is not included in the
clustering process.
NormalizeScaling attribute weights to ensure that weights are valid. User-
provided weights are normalized by applying a weighting average that
adds up to 100 percent.

a

—“C:SCSCSC‘é(

ee

ee

ee

New Stores and Stores with a Poor History

Advanced Clustering supports post-processing rules in order to allocate stores that are new or that have a poor history. These rules can be configured for each criterion and can be changed during deployment.

  • Like Stores. This rule allocates new stores or stores with a poor history to the same clusters that the like location belongs to. It requires data to be provided to Advanced Clustering that defines the mapping between the location and like locations. This mapping can be configured by merchandise, and one location can be mapped to multiple locations with different weights. For example, a like location can be used to correct a store with poor history or to allocate a new store to a valid performance cluster.

  • Largest Clusters. This rule allocates new stores or stores with a poor history to the largest cluster identified by Advanced Clustering. Stores can be allocated to a bigger group of stores. For example, a store that has not yet formed a customer base can be allocated to the largest cluster.

  • Cohesive Clusters. This rule allocates new stores or stores with a poor history to the most compact cluster identified by Advanced Clustering. Stores can be allocated to a compact group of stores. For example, stores can be assigned to a cluster that has not been affected because of outliers.

Insights Stage

Use the Insights stage to analyze a scenario, its clusters, and its hierarchy, based on performance and attribute contributions, prior to the approval of the scenario. This stage includes the following tasks:

  • Approve a cluster scenario

  • Create a new cluster within a scenario

  • Rename clusters within a scenario

  • Rank scenarios, if not completed earlier

  • Flag a cluster scenario as “system preferred”

  • Review a cluster hierarchy in a nested cluster

Select from the following views in this stage:

  • Criteria view. Displays the parent cluster, if it exists, or the scenario for root-level clusters.

  • Parent cluster. Displays any child clusters.

  • Cluster. Displays the stores under the selected cluster.

Each cluster is identified by the following information:

Table 4-22 Insights Stage

Field NameDescription
NameThe name assigned to the cluster.
Nearest Cluster NameThe name of the cluster that is most similar to the named cluster.
# of StoresThe number of stores in the cluster.
Is OutlierWhether or not the cluster is considered an outlier. If it is an outlier, you
may want to review that store.

ee

ieeeSCSSCiN

|

a


In this guide