Dimensional modeling for CRM Applications
Learning objective CRM has emerged as a mission-critical business strategy that is essential to a company s survival The perceived business value of understanding the full spectrum of customers interactions and transactions has propelled CRM to the top of the charts. We examine the implications of CRM on the world of data warehousing 2
CRM Overview Need migrate from a product-centric orientation to one that is driven by customer needs Motivation the better you know your customers, the better you can maintain long-lasting, valuable relationships Goal maximize relationships with your customers over their lifetime Requirement develop a single, integrated view of each customer 3
CRM Overview CRM involves all aspects of the business: marketing, sales, operations, and service What does CRM achieve? attracts new customers, doesn t let the profitable ones leave, and converts unprofitable customers into profitable ones CRM leads to: increased sales effectiveness and closure rates, revenue growth, enhanced sales productivity at reduced cost, improved customer profitability margins, higher customer satisfaction, and increased customer retention 4
Operational CRM Synchronization of customer-facing processes across sales, marketing, operations, and service initial prospect contact, quote generation, purchase transaction, fulfillment, payment transaction, and ongoing customer service Each touch point in the customer contact cycle represents an opportunity to: collect more customer metrics and characteristics, and leverage existing customer data to extract more value from the relationship 5
Analytical CRM Data is created by operational CRM Need to store and analyze the historical metrics resulting from customer interaction and transaction systems Sounds familiar? 6
CRM and DW Data warehouses are at the core of CRM Serve as repository to collect and integrate customer information in operational systems (or external sources) Foundation that supports panoramic (360- degree) view of our customers, including customer data from various sources: transactional data, interaction data (solicitations, call center), demographic and behavioral data (typically augmented by third parties), and self-provided profiles 7
CRM and DW Analytic CRM is enabled via accurate, integrated, and accessible customer data in the warehouse Measure the effectiveness of decisions made in the past to optimize future interactions Identify up-sell and cross-sell opportunities, pinpoint inefficiencies, generate demand, and improve retention. Integrated data generate scores that close the loop back to the operational CRM 8
CRM and DW Model output translates into specific tactics recommended for the next point of customer contact, appropriate next product offer or antiattrition response 9
CRM and DW As the organization becomes more centered on the customer, so must the data warehouse Data warehouses grow as we collect more information about customers Data staging processes grow more complicated as we match and integrate data from multiple sources Need for a conformed customer dimension becomes paramount 10
Customer Dimension A well-maintained, well-deployed conforming customer dimension is the cornerstone of a CRM The customer dimension can be extremely deep (with millions of rows), extremely wide (with dozens or even hundreds/thousands of attributes), and sometimes subject to rather rapid change 11
Customer Dimension Name and Address Parsing Regardless of whether individual human beings or commercial entities, we typically capture customers name and address attributes Common (but wrong) approach: generalpurpose columns for names and addresses Name-1 through Name-3 and Address-1 through Address-6 12
Customer Dimension Name and Address Parsing Common data quality problems No consistent mechanism for handling salutations, titles, or suffixes What is the first name or how to be addressed in a personalized greeting? Multiple customers listed in a single name field No guaranteed conformance with postal authority regulations or support address matching or latitude/ longitude identification 13
Customer Dimension Name and Address Parsing Solutions: break name and location attributes into many elements Standardize ( Rd for Road ) Validate (ZIP code and state) name and address data cleansing and scrubbing tools 14
Customer Dimension International Names and Addresses Universal representation Design consistent from country to country Similar data appear in predictable places Cultural correctness Appropriate salutation and personalization for a letter, e-mail, or telephone greeting Differences in addresses Different addresses may be required foreign mailings from the country of origin to the destination country (presenting the destination city and country in capital letters) 15
Customer Dimension International Names and Addresses Include an address block attribute with a complete valid postal address including line breaks rendered in the proper order according to regulations of the destination country Telephone numbers presented differently depending on where the phone call originated attributes representing the complete foreign dialing sequence, complete domestic dialing sequence, and local dialing sequence 16
Customer Dimension Dates It s common to have dates in the customer dimension date of first purchase, date of last purchase, and date of birth Summarize these dates by the special calendar attributes of our enterprise such as seasons, quarters, and fiscal periods Dates changed to foreign key references to the date dimension Date dimension copies (role-playing dimensions) declared as semantically distinct views Example: First Purchase Date 17
A Case for Snowflaking Assume external data with 150 demographic and socioeconomic attributes of counties Rather than repeating this large block of data for every customer, we snowflake Save significant space since the customer dimension is large The data is administered and loaded at different times 18
Customer Dimension Segmentation Attributes and Scores Customer segmentation classifications or scores: Demographic Gender, Ethnicity, Age or other life-stage classifications Income or other lifestyle classifications Status new, active, inactive, closed Recency date of last purchase Frequency total purchase transaction count Intensity total net purchase amount Scores characterizing the customer purchase behavior, payment behavior, product preferences, propensity to churn, and probability of default 19
Large Changing Customer Dimension Snowflaking with minidimension 20
Large Changing Customer Dimension Snowflaking with several minidimensions 5 demographic attributes, each with 10 possible values, then 100,000 rows Cases where we need to support more Multiple minidimensions cluster similar attributes together demographic income and lifestyle purchase and credit behavioral scores Don t overdo it! No separate minidimension for each demographic attribute, e.g., age, gender, or income 21
Variable-Width Attribute Set The longer the relationship, the more we know about customers 10 million initial prospects described by few attributes 1 million customers with more attributes Not possible to store prospects and customers together in a single dimension 22
Variable-Width Attribute Set Break the dimension into: base dimension consisting of attributes common to both prospects and customers Minidimension with attributes known only for customers Many fact table rows join to an empty customer row in the minidimension 23
Customer Hierarchies Commercial customers often have a nested hierarchy of entities ranging from individual locations up through regional offices, business unit headquarters, or parent companies These hierarchical relationships may change frequently as customers reorganize themselves internally or are involved in acquisitions 24
Customer Hierarchies Fixed-Depth Hierarchies A customer dimension that is highly predictable with a fixed number of levels Example with 3 levels: corporate parent, business unit headquarters, and regional offices (from top to bottom) We add distinct attributes in the customer dimension corresponding to these levels 25
Customer Hierarchies Variable-Depth Hierarchies Set of commercial customers with relationships given by an organizational tree Need to summarize sales at any level in the tree Add a bridge table between customer dimension and fact table 26
Customer Hierarchies Variable-Depth Hierarchies The bridge table contains one row for each path from a node to each node below it: the number of levels between them bottom flag for leaves top flag for root An additional row for the zero-length path from node to itself 27
Customer Hierarchies Variable-Depth Hierarchies 28
Customer Hierarchies Variable-Depth Hierarchies We can constrain the customer table to a particular parent customer and request any aggregate measure of all the subsidiaries at or below that customer We can use the # of Levels from Parent to control the depth of the analysis We can use the Bottom Flag to jump directly to all the bottom-most customer entities but omit all higher-level customer entities 29