7 Customer Segmentation
This chapter provides details about the use of the Customer Segmentation application.
Overview
Customer Segmentation (CS) lets users create segments of customers based on common attributes, such as customer demographics, in order to help a retailer manage merchandise and sales strategies in a targeted way. Segments can help retailers understand the types of customers who shop in their stores and gain insight into their typical shopping patterns. This understanding can help retailers target specific customers.
The application optimizes segments in order to determine the minimum number of segments that best describes the data used in the analysis and that best meets the business objectives defined when the segments are designed. What-if scenarios and ranking can be used to compare how cohesive and well separated the segments are in each scenario as the number of segments is increased. The application uses scoring to indicate which segments fall below defined thresholds and may require manual intervention. Business Intelligence graphics illustrate the patterns in the data and the attributes that are important in each segment.
The key features available in CS are:
-
Recommendations are provided for the important attributes to use in creating segments.
-
Segments can be created on continuous attributes such as sales performance as well as discrete attributes such as customer gender.
-
Configurable segmenting criteria such as demographics and RFM (Recency and Frequency Measures) are provided.
-
Recommendations are made for the optimal number of segments and the scores for each segment. These are based on the quality of the segments: how cohesive and well separated the segments are.
Data Requirements
Customer Segmentation relies on following data, and it uses ETL to load the data.
Table 7-1 Data Requirements
Object | Granularity | Required/Optional |
---|---|---|
Hierarchy |
Product, Location, and Fiscal |
Required |
Customer Attribute |
Demographic or User Defined |
Required for demographic segmentation |
Sales Transaction |
Customer-identified, Product, Date, Transaction ID |
Required for RFM segmentation |
Alternate Hierarchy |
CM Group or Trade Area |
Optional |
Multiple Hierarchies and Support
Customer segments can be generated for the following combinations of hierarchies.
Product Hierarchy
One of the following hierarchies can be used:
-
Core merchandise hierarchy
-
Alternate hierarchy
The level at which segments should be created can be configured. The user interface is used to create the specific node identified. This allows the creation of different segments of customers for different geographic regions. For example, different segments can be defined for Canada vs. France.
Location Hierarchy
One of the following hierarchies can be used:
-
Core location hierarchy
-
Alternate hierarchy (optional)
The level at which segments should be created is configurable. The user interface is then used to create the specific node identified. This allows the creation of different segments of customers for different geographic regions. For example, different segments can be defined for Canada vs. France.
Calendar Hierarchy
This includes:
-
Core fiscal calendar hierarchy (week, month/period, quarter, half, year)
-
Gregorian calendar (week, month, quarter, half, year). Leverages a start and stop date (day level date range)
-
Planning period. Leverages alternate hierarchies, including planning periods, buy periods, and defined holiday time periods such as back to school and Fourth of July. This is optional.
Segments can be defined for any of these three calendar hierarchies (the segment effective period). Note that the source time period for historical data only uses the core fiscal calendar hierarchy. A configuration permits data aggregation at either the fiscal period or fiscal quarter, so the user can select any level that is at that level or above in the user interface.
Supported Segment Criteria
In customer segmentation, the segment criteria consist of a set of attributes that define customer segments. These attributes can be either discrete or continuous. A group of these segments is called "Segment by." For example, demographic data, such as income and gender, and store properties, such as store and formal, can be grouped into a Customer Demographic Segment by.
Here is the list of the default Segment bys that are supported.
Customer Demographics
Customers are segmented based on the similarity in the values of the various customer attributes. Examples include gender, income, educational background, age, and range. Additionally, it possible to use several user-defined numeric or discrete attributes.
RFM and Customer Behavior
Customers are segmented based on attributes that are comprised of aggregate metrics regarding their purchase behavior. Examples include the number of purchases, amount of sales, average basket size, and share of products purchased while being promoted. Retailers can analyze the segment composition and related business intelligence in order to better understand the customer shopping behavior associated with the segments.
Category Purchase Behavior
Customers are segmented in a similar manner to RFM and Customer Behavior, with the difference that these attributes are calculated for a selection of the most important product categories. The category level used here is configurable. It helps enable segmentation by product category purchase behavior.
Customer Segmentation Attributes
Segment by uses a collection of attributes, including demographics, purchase behavior, product purchase behavior, product profiles, and user defined. Each quarter, the batch processing creates new versions for each location at the configured level of the location hierarchy. During this process, attributes are summarized and their data is analyzed for usefulness for segmentation. An attribute can have a different level of usefulness for each of the different versions. For example, if the majority of customers in the Canada location provide a gender attribute, and the majority of the customers in the France location do not provide a gender attribute, then gender can be used in Canada, but not in France.
Furthermore, within the same location, it is possible for an attribute to be considered useful for the most recent quarter, while in the prior quarter it was not useful because there were insufficient values available during that time. The aggregate statistics about the attributes for a version can be seen using the Explore Data screen after selecting the segment criteria. Segments created in the previous quarter have different statistics than those that are created during a different quarter.
Demographics
Demographic segmentation relies on customer attributes that are loaded into the AIF data warehouse. The set of attributes used from the data warehouse customer dimension is fixed. If alternative attributes are needed, see User Defined. Once loaded using AIF DATA jobs in POM, they can be used by Customer Segmentation. They define details about each customer that can be used for creating customer segments using those values. During the time when a new version is created, only attributes that have 15 or fewer discrete values are used. Attributes with a higher number of discrete values are not considered for customer segmentation.
Purchase Behavior
Customer segmentation uses a fixed set of sales transaction metrics, which are obtained from sales transactions identified by a customer ID. The values include Sales Quantity, Sales Retail $, Gross Margin $, Promotional Sales Quantity, Promotional Sales Retail $, Promotional Gross Margin $, Number of Transactions, Average Transaction Count per configured period, SKU count, Transaction Basket Size, and Promoted Sales Share of Total Sales.
Product Purchase
Customer segmentation uses a selection of the top categories to portray the shopping patterns for each customer for some product categories. Each time a version is created, the categories with the highest amount of sales are picked as the categories for which Product Purchase based assessments are done. These attributes are similar in concept to the Purchase Behavior segment; however, these are specific to the top categories for the location associated with the version. The attributes include Sales Quantity, Sales Retail $, Gross Margin $, Promotional Sales Quantity, Promotional Sales Retail $, Promotional Gross Margin $, Number of Transactions, Average Number of Transactions, SKU Count, Transaction Basket Size, Promoted Sales Share of Total Sales, Average Basket Sales $, and Average Basket Gross Margin $.
Product Profile
Customer Segmentation uses a fixed set of category-based sales profile values. For the same set of top categories described in Product Purchase, the share of the customer's total purchases for the category is calculated. The share of the Promotional Sales Retail $ of the customers total sales is also calculated.
User Defined
For any attribute that is available for the customer, but is not accounted for in the default set of attributes, there are provisions for loading a set of customer or user defined attributes into the AIF data warehouse. These attributes can be either numeric values or discrete values. If the attribute value is numeric (such as a zip code), but must be treated as discrete rather than a ranged numeric value, then the attribute must be loaded to an appropriate text attribute column in RI. Any attribute that has more than 15 distinct values will not be used by the segmentation process.
Once an attribute is defined, it is possible to adjust the configuration data in the database to assign a more context-suitable name for the attribute. This enables the user interface to identify the attribute as a specific attribute, and not just as a generic Custom Text Attr or Custom Number Attr.
Configuration Process
Default configuration occurs during the installation and upgrade. The configuration process is responsible for enabling or disabling any attributes in the application. This ensures that the desired attributes are available for use during the segmentation process.
-
All attributes are enabled by default.
-
Any discrete attribute that has more than n=15 attributes values is not configured by default. Note that the value of n is a configuration and can be modified at the time of deployment.
-
The UI formatting of each attribute is identified based on the data type of the attributes and by the name of the attribute.
The following configurations may require manual overrides if the default configuration is not acceptable or data is not available.
Table 7-2 Manual Overrides
Name | Description |
---|---|
Enable or disable Segment by |
Disable Segment by. For example, if there are no customer demographic attributes, then the Customer Demographics segment by can disabled. |
Enable or disable attributes |
Enable an attribute to be considered for segmentation. For example, if there is no intention on loading user defined attributes, they can all be disabled so that when new versions are created, no processing time is spent analyzing those attributes. |
Change UI formatting |
Change formatting associated with the attributes such as label, decimals, percent, and currency. These are configurations for each attribute, and do not rely on XLIF entries. |
Location hierarchy level |
If customer segments are desired for each location at a given location level, then the configuration can be adjusted so that all processing is done at that level. This also requires an adjustment of the approval level so that it allows the segments to be approved. |
Outlier rule |
Change default outlier rules for a Segment by. By default, the distance from centroid rule is enabled. See the section below for other supported outlier rules. |
Table 7-3 Enable or Disable Segment By
Segment By | Description | Example | Enable |
---|---|---|---|
Customer Demographics |
Segment customers using demographic values |
Age range, income, gender |
Enable if consumer demographic attributes are available |
RFM and Customer Behavior |
Segment stores using location attributes |
Income, climate, size, store format |
Enable if location attributes are available for each store |
Category Purchase Behavior |
Segment stores using sales metrics |
Sales revenue, sales unit, gross margin |
Enable if retailer wants to segment stores using performance metrics |
Attribute Preprocessing
Before customer segments are created, the available customer data must be preprocessed in order to identify the sample of customers to use for segmentation and to determine which attributes are the most beneficial for use while creating segments. A few configurations can be manipulated to help improve this process. The following table defines some configurations that can be adjusted to control how attributes are used by the system. These values can be manipulated in the table CIS_TCRITERIA_ATTR.
Table 7-4 Attribute Configuration
Column | Description |
---|---|
DELETE_FLG |
Set value to Y to prevent an attribute from being used by any processing. |
SAMPLE_FLG |
Set value to Y so that the attribute to be used has a stratified sample of customers. This should help ensure that an appropriate selection of customers are represented in the sample. Up to three attributes can be set to Y. If no attributes are configured with a Y value, then a random sample of customers will be used. |
DISPLAY_FORMAT_ID |
This can be adjusted to use different display formats, as defined in RSE_DISPLAY_FORMAT. |
NUM_BUCKETS |
Allows a different number of buckets for use when showing summary metrics for a numeric attribute. |
ATTR_IMP_FLG |
When set to a Y value, the importance of this attribute is analyzed for usefulness during segmentation. If an attribute will never be used for segmentation, setting the value to N will exclude it from the attribute importance calculations. |
Another configuration that can adjusted is the CIS_BUS_OBJ_TCRITERIA_ATT_XREF.VALIDATING_ATTR_FLG. For any attribute that has a Y value for this column, the attribute is shown in the Insights portion of the UI. If it is determined that an attribute should not be displayed, then this value can be changed to a value N so that it is excluded from display in the Insights results.
Segmenting Approach
Customer segmentation uses Oracle Data Mining for the creation of its segments. The k-means approach that is used results in the creation of segments in a hierarchal manner. The process automatically determines which attribute is the best attribute to split into an additional segment. This process is continued until the desired number of clusters has been achieved.
Customer Segment Store Profile Generation
Customer Segmentation calculates the sales share of customer segments for each store. These store profiles can be generated by the user for the approved customer segments from the user interface. They can then be consumed by RI to generate business reports and Store Clustering to generate customer centric store clusters.
Preprocessing
As part of preprocessing when a version is created, Customer Segmentation first filters the customer and then samples the customer.
Filter Customer
The filtering step allows the implementer to set the following conditions in order to remove outlier customers during different phases of Customer Segmentation,
Table 7-5 Filter Step Conditions
Filter Rule | Operation |
---|---|
Fake Customers |
This rule discards customers if the number of transaction per day exceeds the value x (the max daily transaction threshold to identify a fake customer - default 10) |
Discarding Customers |
Keep customers where customer sales fall between min (sales or # of transaction) and max (sales or # of transaction).Minimum/Maximum percentile for amount of sales transactions. (default .001)Minimum/Maximum percentile for number of sales transactions. (default .001) |
New Customer Rules |
Include new customers if there is enough existing sales history for the customer. Use configurations that control the min/max values for average transaction count and average transaction amounts used to include new customers:Minimum/Maximum percentile for amount of sales transactions. (default .001)Minimum/Maximum percentile for number of sales transactions. (default .001) |
Top Categories |
A configuration that can limit the number of categories (n) that a user picks. The system picks it based on the default - Sales Revenue of the category to be used to calculate the top categories. Allowable values are limited to SLS_AMT, SLS_QTY, PROFIT_AMT, and SHARE variations of the same (i.e., SLS_AMT_SHARE). |
Sample Customer
The sampling step allows the implementer to enable and adjust sampling customers.
Table 7-6 Sample Step Conditions
Sample Rule | Operation |
---|---|
Target Sample |
Set the value to Y so that the attribute to be used has a stratified sample of customers. This should help ensure that an appropriate selection of customers is represented in the sample. Up to three attributes can be set to Y. If no attributes are configured with a Y value, then a random sample of customers will be used. This can be adjusted by changing CIS_TYPE_CRTIERIA in Data Management UI. |
Sample Size |
The sampling percentage for Customer Segmentation. This can be adjusted by changing RSE_CONFIG in the Data Management UI. |
Customer Metrics
The Retail Science process supports the following derived customer metrics, which can be used for analysis by an application such as Customer Engagement. These metrics are calculated using customer-linked transaction data. In addition to helping understand how customers have behaved in past, these metrics can also help predict future behavior.
Table 7-7 Customer Metrics
Metric | Description |
---|---|
Customer Latency/Time Between Visits |
The number of days between each of a customer's transactions sales or return. |
Customer Lifespan |
The time between a customer's first and last purchase. |
RFMThe |
RFM (recency, frequency, monetary) score determines quantitatively which customers are the best ones by examining how recently a customer has purchased (recency), how often the customer purchases (frequency), and how much the customer spends (monetary). |
Projected next purchase date |
Prediction of the next likely customer purchase date. |
Location Loyalty |
How loyal are customers to a specific location? A value of 100% indicates that they always shop at a particular location. |
Style Loyalty |
How loyal are customers to a particular style? A value of 100% indicates that they always prefer one specific style. |
Color Loyalty |
How loyal are customers to a particular color? A value of 100% indicates that they always prefer one specific color. |
Brand Loyalty |
How loyal are customers to a particular brand? A value of 100% indicates that they always prefer one specific brand. |
Price Efficiency Loyalty |
How efficient are customers in getting a promotion price? A value of 100% indicates that the customer always buys items on promotions or is very efficient in obtaining a good price. |