CRM and Information Visualization
Gürdal Ertek, Ph.D. Tuğçe Gizem Martağan
Customer Relationship Management (CRM)
Traditional Marketing CRM Goal: Expand customer base, Goal: Establish a profitable, increase market share by long-term, one-to-one mass marketing relationship with customers; understanding their needs, preferences, expectations Product oriented view Mass marketing / mass production Standardization of customer needs Transactional relationship Customer oriented view Mass customization, one-to-one marketing Customer-supplier relationship Relational approach
What is CRM?
“The approach of identifying, establishing, maintaining, and enhancing lasting relationships with customers.”
“The formation of bonds between a company and its customers.”
Strategies in CRM for Mass Customization
• • • • Prospecting (of first-time consumers) Loyalty Cross-selling / Up-selling Win back or Save
The Marketing Perspective
RECENCY FREQUENCY MONETARY VALUE METHOD
CUSTOMER VALUE METRICS
Campaign Management: The Marketing Perspective
• • • • Developing effective campaigns Effectively predicting the future Retaining existing customers Acquiring new customers
Campaign Management: The Cap Gemini Model
Understand market and consumers’ needs and preferences Exploit customer intelligence. Use channel integration
( Offer is developed ) Define market strategies
Retain customers by: Loyalty programs Communication Service forces
Acquire customers Use sales force effectively Develop marketing programs 9
Designs/improves products/offers/services/promotions 5. Defines communication strategies 4. Defines objectives 2.. Identifies customers 3.. Tests the impacts of her decisions 6. Revises her decisions for maximum effectiveness
. 1.Campaign Management: The Marketing Perspective
The marketing manager.
Campaign Management Step 1: Define Objectives
Targeting Existing Customers Retention Strategy Creating Loyalty?
Increasing the satisfaction level?
Cross-selling or Up-selling?
Targeting New Customers Acquisition Strategy
Target customers that show characterstics similar to existing groups of customers
Campaign Management Step 2: Identify Customers
Perform SEGMENTATION • Define the right customers • Use information of past transactions as key for making predicting future ones • Define the segments and their characteristics • Develop customized marketing strategies for the different segments
Campaign Management Step 3: Communication Strategies
• Which message should be transmitted? • Which channel should be used?
marketing costs • Define the product / offer / service / promotion and its general structure • Identify effective use of sales and communication channels
. Offers.Campaign Management Step 4: Design the Products. Services and Promotions
• Analyze the price. time period. risks.
Campaign Management Step 5: Test the Impacts
• Impacts of the decisions have to be tested and and assessed on a sample
Campaign Management Step 6: Revise the Decisions
• Make revisions to the targeted offer / service / promotions • Finally apply the decisions to the whole segment or population
Monetary Value )
• Recency – When was the last customer interaction? • Frequency – How frequent was the customer in its interactions with the business? • Monetary value of the interactions
.RFM Method (Recency. Frequency.
RFM Method (Recency. Frequency.52% of the customers)
Is there any relation between the responding customers and their historical purchasing behaviours?
. Monetary Value )
Marketing Problem: A firm has sent e-mail to 30.000 of its existing customers. 458 of them responded (1. announcing a promotion of $100.
• Recency codes are assigned: Top group has code 5.RFM Method: Recency Coding
• 30.000 customers are sorted in descending order with respect to their most recent purchases • Sorted data is divided into 5 equal groups. bottom group has code 1
. each of them containing 6.
00% + 1.50% + 0.00
• According to analysis based on customer recency.38
2.5 0.10% + 2.38) / 5= 1.RFM Method: Recency Coding
4.00 0. the group having the highest recency group has also the highest response rate
1.00 1.00 5 4
• Remark: (3.1
3.52% which is the response rate • Strict Rule: Ones who have purchased recently are much more willing to buy new products than others 22 purchasing in the past
Recency code R
.62 0.62% + 0.
• Assign them to 5 groups.
– Frequency metrics: Average number of purchases made by customer in a time period t – Sort customers in descending order with respect to their purchase frequency.RFM Method: Frequency Coding
• Sort the 30. • Assign frequency codes such that the top group has code 5 and the bottom group has code 1. 23
.000 customers with respect to frequency metrics. top %20 in the first frequency group.
RFM Method: Frequency Coding
2 1.5 1 0.5 0 5 4 3 2 1 Frequency code F
1.3 0.8 0.9
• It is observed that highest response rate is from the customers having highest frequency • Frequent people respond better than less frequent ones but differences between groups are less than the ones in the recency • The lowest frequency group always contains new customers 24 • That is why it is named RFM
RFM Method: Monetary Value Coding
• The same process as recency and frequency coding • Sorting is done with respect to monetary value metric
– Monetary value metric is the average amount purchased in a time period t
• At the end of the monetary value coding, assign monetary value codes M = 1,...,5 to groups according to their groups.
RFM Method: Monetary Value Coding
1.8 1.4 1.2 1.1
2 1.5 1 0.5 0 5
• It is observed that highest response rate is from the customers having highest monetary value • Unlike the recency case, there are not big differences between groups
Monetary value code M
RFM Method: Putting the Codes Together
• At the end of the monetary coding firm obtain R F M metrics for customers. Each customer belongs to one of 125 possible combinations of the RFM values:
.. • Increase the response rate • Increase profitability
.RFM Method: STEPS
• Create 3 digits RFM codes cells • All cells having the same number of customers in them • RFM values are used to define group of customers that marketing campaign should target or should avoid • Used for identifying customers having high probability to respond to campaigns: 555’s response rate > 552’s > 543’s >541..
Customer Value Metrics
• Critical measures used to define customer worth in knowledge-driven and customerfocused marketing
Customer Value Metrics: Size of Wallet
• Size of wallet =
Sales to focal customer by firm j
• Assumption: Firms prefer customers with large size of wallet in order to retain large revenues and profits
which also considers customers with no purchase
• Individual share of wallet % =
Sales to focal customer by firm j
.Customer Value Metrics: Individual Share of Wallet (SW)
• A proportion expressed in terms of percentage. calculated among buyers • Measured at individual level • A measure of loyalty • Can be used in future predictions • Different from the “market share”.
..Customer Value Metrics
• Share of wallet and size of wallet should be analyzed together because.
Size of Wallet Customer 1 $500 Customer 2 $100 Share of Wallet 50% 50% Purchases $250 $50
than from A to C: 10%*20% = 2% Brand A Brand A Brand B Brand C 60% 10% 20% Brand B 30% 80% 15% Brand C 20% 15% 70%
.Customer Value Metrics: Transition Matrix
• Shows expected share of wallet from multiple brands • Depicts consumer’s willingness to buy over time • Transition probability from B to A.
The Engineering Perspective DATA MINING
data • Data readily resides in the company’s data warehouse • Data cleaning is almost inevitable
. and analysis of –typically huge amounts of. storage.Data Mining
inputs and outputs • Predicting future patterns
Goals of Data Mining • • • • Developing deeper understanding of the data Discovering hidden patterns Coming up with actionable insights Identifying relations between variables.
Data Mining: Steps
• • • • • Data selection Data cleaning Sampling Dimensionality reduction Data mining methods
Data Mining: Methods
• Exploratory Data Analysis • Segmentation
– Cluster Analysis – Decision Trees
• • • •
Market Basket Analysis Association rules Information Visualization Prediction
– Regression – Neural Network – Time Series Analysis
Data mining algorithms. • Can only detect certain types of patterns and insights • Are too complex for end users to understand
• A field of Computer Science which has evolved since the 1990s. computational power.
• Before 1990s: Graphical methods for data analysis to pave the way for statistical methods • After 1990s: – Computer hardware has advanced with respect to memory. ERP)
. graphics calculations – Software has advanced with respect to user interfaces – Data collection systems have advanced (barcodes. RFID.
• The analyst does not have to understand complex algorithms. • Almost no training required. • There are no limits to the types of insights that can be discovered.
Case Studies Analysis of Supermarket Sales Data
Field Name Desciption
Case Studies Analysis of Spare Parts Sales Data
....12) Year (ex: 2002) Quantity required
Price of one unit of product in YTL*
Revenue from the order line
Assumption: Each customer gives at most one order each day.
Depot ID SKU (Stock Keeping Unit) Number
DAY MONTH YEAR QUANTITY
Vendor (Customer) Number
Day of the month (1.....31) Month of the year (1.
Determining Top Products: Pivot Table for Determining REVENUE_SUM
Determining Top Products: Pivot Table for Determining COUNT (Frequency)
Determining Top Products: Scatter Plot
..Seasonality of Top Products
Seasonality of Top Customers: Pivot Table
Determining Top Customers: Pareto Curve (ABC Analysis)
100 90 80 70 60 50 40 30 20 10 0 0 10 20 30 40 50 60 70 80 90 100
Cumulative % Revenue
Cumulative % Customers
Seasonality of Top Customers: Starfield Visualization
Case Studies Analysis of ÖSS 2004 Data
Field Name HS_NAME HS_TYPE_TEXT
UNIV_NAME UNIV_DEPT RANK_SAY
Desciption High School Name High School Type
University Name University Department Rank According to Sayısal Score
Pareto Squares: Model Definitions
Pareto Squares: Optimization Model
(2001) Information Visualization. (2004) Data Mining Techniques. A. John Wiley & Sons Inc.com/articles/Art149. W.. G. V. Sabancı University. ACM Press. G.htm • Kumar.References
• Berry. Reinartz. (2006) Customer Relationship Management. Istanbul. Visual Data Mining with Pareto Squares for Customer Relationship Management (CRM) (working paper. Quick profits with RFM analysis. • Spence.
. J. A.. • Ertek. Linoff. M.. A Databased Approach. http://www. M. A framework for visualizing association mining results (accepted for LNCS) • Hughes. R. A. G. Demiriz.dbmarketing. Wiley Publishing. S. Turkey) • Ertek. J.