Data Warehouse Design to Support Customer Relationship Management Analyses
description
Transcript of Data Warehouse Design to Support Customer Relationship Management Analyses
Data Warehouse Design to Support Customer Relationship Management Analyses
Colleen Cunningham, Il-Yeol Song and Peter Chen
DOLAP ‘04November 12, 2004
2
Agenda
Background Motivation Methodology Results Areas for future research Contributions & Conclusions Q & A
3
Agenda Background
CRM Definition Why Use CRM? Customer Lifetime Value (CLV)
Motivation Methodology Results Areas for future research Contributions & Conclusions Q & A
4
CRM Definition
Proactive strategy Utilizes organizational knowledge Utilizes technology
Support profitable long-term relationships with customers
All customers are not equal More expensive to acquire new
customers than it is to retain customers
Repeat customers can generate more than twice as much gross income as new customers
Why Use CRM?
6
Customer Lifetime Value (CLV)
CLV = Historic value + Potential Future value
Historical Value = Nj=1 (Revenuej - Costj)
j: individual products that the customer has purchased
Potential Future Value = Nj=1 (Probabilityj X Profitabilityj)
j: individual products that the customer could
potentially purchase
7
Customer Lifetime Value (CLV)
Use customers’ Lifetime Value (CLV) to classify customers
Table 1: Customer Segments
Historic Value
Low High
Future Value
High II. Re-Engineer IV. Invest
Low I. Eliminate III. Engage
8
Customer Lifetime Value (CLV)
Table 2: Corresponding Segmentation Strategies
Historic Value
Low High
FV
HighUp-sell & cross-sell activities
and add valueTreat with priority and
preferential
Low Reduce costs and increase prices
Engage customer to find new opportunities in order to sustain
loyalty
9
Agenda
Background Motivation
Overview New Metrics
Methodology Results Areas for future research Contributions & Conclusions Q & A
Motivation The DW directly impacts a company’s ability to
perform analytical CRM analyses
50% - 80% of CRM initiatives fail (Myron and Ganeshram 2002; Panker 2002)
Systematically examine CRM factors that affect design decisions for DWs in order to: Build a taxonomy of CRM analyses Develop heuristics for CRM DW design decisions Create metrics to objectively evaluate CRM DW models
New Metrics
% Success Ratio (rsuccess) = Qp / Qn
Qp: the total number of analyses that the model could successfully handle
Qn: the total number of analyses issued against the model
It measures the “robustness” of the model
New Metrics
CRM Suitability Ratio (rsuitability) = Ni=1(XiCi)
/ N N: the total number of applicable analysis
criteria C: individual score for each analysis
capability X: weight assigned to each analysis capability It measures the “appropriateness” of the
model for a specific company
13
Agenda
Background Motivation Methodology
Identify Minimum Requirements Preliminary Starter Model for CRM DW Implementation Evaluation of Model
Results Areas for future research Contributions & Conclusions Q & A
Methodology Overview Identify categories of analyses Identify specific analyses & KPIs Categorize the specific analyses & KPIs Identify specific data points Design the CRM starter model Implement the CRM starter model Continue collecting additional analyses Randomly select analyses to run Evaluate the model
Minimum Design Requirementsfor CRM DWs
Analysis Type/Data Maintenance DescriptionCampaign Analysis Ability to evaluate different campaigns and responses over
timeChannel Analysis Ability to evaluate the profitability of each channel (e.g.
stores, web, phone, etc.)Cross-selling Analysis Ability to identify additional types of products that customers
could purchase, which they currently are not purchasing.
Customer Attrition Ability to identify root causes for customer attritionCustomer Loyalty Ability to understand loyalty patterns among different
relationship groupsCustomer Profitability Ability to determine profitability of each customerCustomer Retention Ability to track customer retentionCustomer Scoring Ability to score customersCustomer Segmentation Ability to segment customers into multiple customer
segmentationsCustomer Service Analysis Ability to track and analyze customer satisfaction, the average
cost of interacting with the customer, the time it takes to resolve customer complaints, etc.
Demographic Analysis Ability to perform demographic analysis
Table 3: Minimum Design Requirements for CRM Data Warehouse
Minimum Design Requirementsfor CRM DWs
Table 3: Minimum Design Requirements for CRM Data Warehouse (Continued)
Analysis Type/Data Maintenance DescriptionHousehold Analysis Ability to associate customers with multiple extended
household accounts.Market Profitability Ability to determine profitability of each marketProduct Delivery Performance Ability to evaluate on-time, late and early product deliveriesProduct Profitability Ability to determine profitability of each productProduct Returns Ability to analyze the reasons for and the impact of products
being returnedTrend Analysis Ability to perform trend analysisUp-selling Analysis Ability to analyze opportunities for customers to buy larger
volumes of a product or a product with a higher profitability margin
Web Analysis Ability to analyze metrics for web site
Preliminary starter model for CRM DW
Profitability Fact Table
PK,FK1 ProductKeyPK,FK2 ChannelKeyPK,FK3 SupplierKeyPK,FK4 PromotionKeyPK,FK5 CustomerKeyPK,FK6 CustomerDemographicsKeyPK,FK7 TimeKeyPK,FK8 MarketKeyPK,FK9 CommentsKeyPK,FK10 ExtendedCustomerKeyPK,FK11 SalesRepresentativeKey
OrderNumber (DD)
Channel Dimension
PK ChannelKey
Customer Dimension
PK Customer Key
CustomerID (Natural Key)FK1 CommentsKeyFK2 MarketKeyFK3 SalesRepresentativeKeyFK4 ActivationDateKey (FK)FK5 AttritionDateKey (FK)FK6 CountyDemographicsKey Product Dimension
PK ProductKey
Supplier Dimension
PK SupplierKey
Time Dimension
PK TimeKey
Promotion Dimension
PK Promotion Key
Customer Behavior Dimension
PK ExtendedCustomerKey
Customer Demographics Dimension
PK CustomerDemographicsKey
Market Dimension
PK MarketKey
Extended Household
PK,FK2 HouseholdKeyPK,FK1 CustomerKey
Household Dimension
PK HouseholdKey
Sales Representative Dimension
PK SalesRepresentativeKey
CustomerExistence
PK,FK1 Customer KeyPK,FK2 StartDate
ProductExistence
PK,FK1 ProductKeyPK,FK2 StartDate
Customer Service Fact Table
PK,FK1 StartDatePK,FK2 StartTimePK,FK3 Customer Key
InteractionID (Natural Key)FK4 ChannelKey
Future Value Fact Table
PK,FK1 ScenarioKeyPK,FK2 Customer KeyPK,FK3 StartDatePK,FK4 ProductKey
Scenario Dimension
PK ScenarioKey
sTime Dimension
PK sTimeKey
18
Preliminary starter model for CRM DW
Profitability for any transaction in the fact table can be calculated as follows: Gross Profit = Gross Revenue –
Manufacturing Cost – Marketing Cost – Product Storage Cost
Net Profit = Gross Profit – Freight Cost – Special Cost – Overhead Cost
Gross Margin = Gross Profit/Gross Revenue
Profitability Fact Table
PK,FK1 ProductKeyPK,FK2 ChannelKeyPK,FK4 PromotionKeyPK,FK6 CustomerDemographicsKeyPK,FK7 TimeKeyPK,FK3 SupplierKeyPK,FK5 CustomerKeyPK,FK8 MarketKeyPK,FK9 CommentsKeyPK,FK10 ExtendedCustomerKey
OrderNumber (DD)QuantitySoldGrossRevenueManufacturingCostMarketingPromotionCostSalesDiscountAmountNetRevenueProductCostProductStorageCostGrossProfitFreightCostSpecialCostOverheadCostNetProfitItemEarlyCountItemOnTimeCountItemLateCountItemCompleteCountItemDamageFree CountTaxAmountUOMConversion FactorUOMProductReturned
19
Implementation
Operating System: Windows 2000 Server
DBMS: SQL Server 2000
Hardware: DELL 1600 database server, single processor, 2.0 MHz
Fact tables contained 1,685,809 records
Evaluation of Model
A series of randomly-selected CRM queries were executed against the proposed data warehouse schema
The metrics were computed % Success Ratio (rsuccess)
CRM Suitability Ratio (rsuitability)
Evaluation of Model
Category Analysis
Channel AnalysisWhich distribution channels contribute the greatest revenue and gross margin?
Customer Attrition What are the top 10 reasons for customer attrition?
Customer AttritionWhat is the impact of the value of the customers that have left on revenues?
Customer Profitability Analysis What are the customers' sales and margin trends?
Customer Profitability AnalysisWhich customers are most profitable based upon gross margin and revenue?
Customer RetentionHow many unique customers are purchasing this year compared to last year?
Market Profitability Analysis Which markets are most profitable overall?Market Profitability Analysis Which products in which markets are most profitable?
Order Delivery PerformanceHow do early, on time and late order shipment rates for this year compare to last year?
Order Delivery Performance & Channel Analysis
How do order shipment rates (early, on time, late) for this year compare to last year by channel?
Product Profitability Analysis What is the lifetime value of each product?Product Profitability Analysis Which products are the most profitable?
Returns AnalysisWhat are the top 10 reasons that customers return products?
Returns AnalysisWhat is the impact of the value of the returned products on revenues?
Returns AnalysisWhat is the trend for product returns by customers by product by reason?
Table 4: Sample CRM Analyses
Evaluation of Model: Sample Queries
SELECT b.CustomerKey, b.CustomerName, Sum(a.GrossRevenue) AS TotalRevenue, Sum(a.GrossProfit) AS TotalGrossProfit, TotalGrossProfit/TotalRevenue AS GrossMarginFROM tblProfitabilityFactTable a, tblCustomer bWHERE b.CustomerKey=a.CustomerKeyGROUP BY b.CustomerKey, b.CustomerNameORDER BY Sum(a.GrossRevenue) DESC;
Figure 1: Customer Profitability Analysis Query - Which customers are most profitable based upon gross margin and revenue?
SELECT c.Year, b.MarketKey, b.LocationCode, b.Location, b.Description, b.CompetitorName, d.ProductCode, d.Name, Sum(a.GrossRevenue) AS TotalRevenue, Sum(a.GrossProfit) AS TotalGrossProfit, TotalGrossProfit/TotalRevenue AS GrossMarginFROM tblProfitabilityFactTable a, tblMarket b, tblTimeDimension c, tblProductDimension dWHERE b.MarketKey=a.MarketKey And a.TimeKey=c.TimeKey And a.ProductKey=d.ProductKeyGROUP BY c.Year, b.MarketKey, b.LocationCode, b.Location, b.Description, b.CompetitorName, d.ProductKey, d.ProductCode, d.Name, b.MarketKeyORDER BY Sum(a.GrossRevenue) DESC;
Figure 2: Product Profitability Analysis Query - Which products in which markets are most profitable?
23
Agenda
Background Motivation Methodology Results
Initial Taxonomy of CRM Queries Initial Heuristics for CRM DW Design Decisions
Areas for future research Contributions & Conclusions Q & A
Initial Taxonomy of CRM Analyses
Decision Class Category Analysis Potential Use(s) KPI
S Channel Analysis
Which distribution channels contribute the greatest revenue and gross margin? Resource allocation
S & T Customer AttritionWhat are the top 10 reasons for customer attrition?
Insights for process improvements attrition rate
S & T Customer Attrition
What is the impact of the value of the customers that have left on revenues?
Insights for process improvements attrition rate
SCustomer Profitability Analysis
What are the customers' sales and margin trends? Classify customers
gross margin, revenue
SCustomer Profitability Analysis
Which customers are most profitable based upon gross margin and revenue? Classify customers
gross margin, revenue
S Customer Retention
How many unique customers are purchasing this year compared to last year?
Identify the threshold to overcome with new customers
unique customers/year
S & TMarket Profitability Analysis
Which markets are most profitable overall?
Setting performance goals, allocate marketing resources
gross margin/market
S & TMarket Profitability Analysis
Which products in which markets are most profitable?
Setting performance goals, allocate marketing resources
gross margin/ products/ market
Table 5: Initial Taxonomy of CRM Analyses (S=Strategic and T=Tactical)
Initial Taxonomy of CRM Analyses
Decision Class Category Analysis Potential Use(s) KPI
S & TOrder Delivery Performance
How do early, on time and late order shipment rates for this year compare to last year? Setting performance goals
early delivery, on-time delivery, late delivery
S
Order Delivery Performance & Channel Analysis
How do order shipment rates (early, on time, late) for this year compare to last year by channel?
Setting performance goals, monitoring trends
early delivery, on-time delivery, late delivery
S & TProduct Profitability Analysis
What is the lifetime value of each product?
Managing product cost constraints, identify products to potentially eliminate from product line
gross margin/ product
S & TProduct Profitability Analysis
Which products are the most profitable?
Managing product cost constraints, identify products to potentially eliminate from product line
gross margin/ product
S & T Returns AnalysisWhat are the top 10 reasons that customers return products?
Create pareto charts to identify problems to correct, setting performance goals count
S & T Returns Analysis
What is the impact of the value of the returned products on revenues?
Create pareto charts to identify problems to correct, setting performance goals
count, revenue, profit
S & T Returns Analysis
What is the trend for product returns by customers by product by reason?
Create pareto charts to identify problems to correct, setting performance goals, identify problematic accounts (identify customers that may leave), assess additional service fees
count, revenue, profit
Table 5: Initial Taxonomy of CRM Analyses (S=Strategic and T=Tactical) (Continued)
Initial Heuristics for DW Design Decisions
# Heuristic Benefit
1Include all attributes required to compute the profitability of each individual transaction in the fact table(s)
The ability to generate a profit & loss statement for each transaction, which can then be analyzed along any dimension
2Each dimension that will be used to analyze the Profitability fact table should be directly related to the fact table
Provides improved query performance by allowing the use of simplified queries (i.e. support browsing data)
3 Pay careful attention to the Customer dimension It forces attention to the center of CRM, the customer
4Create a relationship between the Customer dimension and the Market and Sales Representative dimensions
Provides the ability to quickly determine the current market and Sales Representative for the customer by merely browsing the Customer dimension
5Include the attrition date and reaason for attrition attributes in the Customer dimension
Provides the ability to quickly determine if a customer is no longer a a customer by browsing the Customer dimension only
6
Attributes that are likely to change at a different rate than other attributes in the same dimension should be in a separate dimension Minimize the number of updates
7Create a separate "existence" dimension for any entity that can have a discontinuous existence
Provides the ability to track the periods in which the instance of the entity is valid (needed to support some temporal queries)
Table 6: Initial Heuristics for Designing CRM DWs
Initial Heuristics for DW Design Decisions
# Heuristic Benefit
8Create a separate "existence" dimension for any attribute whose historical values must be kept
Provides the ability to track accurate historical values, even during periods of inactivity
9Create a relationship between the Time dimension and each "existence" dimension
Provides the ability to perform temporal queries efficiently using descriptive attributes of the Time dimension
10"Existence" dimensions should be in a direct relationship with their respective original dimensions
11 There should always be a CustomerExistence dimensionThe ability to track and perform analyses on customer attrition
12
If some products are either seasonal or if it is necessary to determine when products where discontinued, then create a ProductExistence dimension
The ability to perform analyses for seasonal and discontinued products
13There should be a Household dimension and an ExtendedHousehold dimension Provides the ability to perform Household analyses
14The organizational hierarchical structure can be contained in one "Market" dimension
Provides the ability to maintain a history of the organizational changes, and the ability to perform analyses according to the organizational structure
Table 6: Initial Heuristics for Designing CRM DWs (Continued)
28
Agenda
Background Motivation Methodology Results Areas for future research Contributions & Conclusions Q & A
29
Areas for Future Research
Compile & categorize additional queries and KPIs that are relevant to CRM
Develop a taxonomy for DW schemas by industry Which schemas are best suited for which types
of analyses?
Compare alternative models
30
Areas for Future Research
Develop data mining techniques that can be utilized with the starter model
Efficiently build aggregation and cube for MOLAP Construction rules
31
Areas for Future Research
Effective use of materialized views in ROLAP What types to create? How to tune? How to evolve?
Contributions
Starter model for CRM Taxonomy of CRM queries and their uses,
including KPIs Heuristics for designing a data warehouse
to support CRM Sampling Technique New Evaluation Metrics
% Success Ratio = Total Passed / Number of Queries
CRM Suitability Ratio = Total Score/Total # of criteria
Conclusions
Our starter model can be used to analyze various CRM analyses: customer profitability analysis, product profitability analysis, channel profitability analysis, market profitability analysis,…
34
Agenda
Background Motivation Methodology Results Areas for future research Contributions & Conclusions Q & A
35
Q & A
Thank You!
Contacts Colleen Cunningham:
[email protected] Dr. Il-Yeol Song: [email protected]