Analytic CRM 2006. 5. 11 tsshin@yonsei.ac.kr
Analytic CRM Analytic CRM Data Mining
Analytical CRM in CRM Ecosystem Operational CRM Business Operations Mgmt. Analytical CRM Business Performance Mgmt. Back Office Front Office Mobile Office ERP Customer Services SCM Marketing Automation Mobile Sales Legacy System Sales Automation Field Services Closed Loop Processing Customer Activity Data Mart Vertical App. Data Warehouse Customer Data Mart Product Data Mart Campaign Management Customer Interaction Call Center Automation Conferencing E-Mail Fax/ Letter Direct Interaction Collaborative CRM Business Collaboration Mgmt. Source : Meta Group
Analytic CRM Analytic CRM The capture, storage, extraction, processing, interpretation, and reporting of customer data to a user. Analytical CRM, also known as "back-office" or "strategic" CRM, involves understanding the customer activities that occurred in the front office. Analytical CRM requires technology (to compile and process the mountains of customer data to facilitate analysis) and new business processes (to refine customer-facing practices to increase loyalty and profitability). Under pressure from analysts and industry experts, most of today's CRM vendors are either creating analytical CRM capabilities or partnering with business intelligence (BI) vendors to incorporate analysis into their offerings.
Operational CRM: Touching the customer (Figure 1-1)
Analytical CRM: Understanding the customer (Figure 1-2) Figure 1-2 shows how the data and processes combine to refine business actions.
One company, two CRM systems (Figure 6-1)
Analytical CRM: The sum of its parts (Figure 6-4)
Analytic CRM 1: A Churn Management 2: B DB Marketing 3: DM (Microsoft) 4: Market-basket Analysis ( )
1: A Churn Management???? Business Issues /??
1: A Churn Management ( ) Churn Management ( ),
1: A Churn Management ( ) MW(Marketing Warehouse) : MAD MDSS............ MAD SAS/WA SAS/Access Cleaning Integration - ( ), MW ID... CIF MAN
1: A Churn Management ( ) : MAN (MINER) MDSS MAD MW MAN (MINER) ID... CIF + RFM + Score Sampleample Explorexplore Modifyodify Modelodel Assessssess,,..., Scaling,... Decision Tree ROI
1: A Churn Management ( ) : MAN MDSS MAD MW MAN Miner Call Center MW ID Explorer Campaign Manager TM/DM... CIF + RFM + Score DM Center Web Teller
2: B DB Marketing 1. 1) Cross-Selling 2), TM, DM 3) Cross - Selling Business Issues
2: B DB Marketing ( ) 2. : MAN MDSS MAD MW MAN ID... Sampleample Explorexplore Modifyodify Modelodel Modification Decision Tree Neural Network CIF + RFM + Score Assessssess ROI
2: B DB Marketing ( ) 3. ( ), : MAN Sampleample Explorexplore Modifyodify Modelodel Assessssess ( ) / Channel
2: B DB Marketing ( ) 4. ( ) : Step 1 Campaign Manager : MAN MAD MW MAN Campaign Manager ID... CIF + RFM + Score MW Call Center DM Center Web Teller
2: B DB Marketing ( ) 5. ( ) : Step 2 : MAN Explorexplore Modifyodify ( ), Modelodel, MDS Assessssess
2: B DB Marketing ( ) 6. Cross - Selling( ) - 2 Sampleample Explorexplore Modifyodify ( ) Modelodel Assessssess Neural Network Association Business Cut-Off Point ROI
2: B DB Marketing ( ) 7. 1) ( ) :, Call Center, DM 2) ( ) Cross-selling ( ) 40% 20%
3: DM (Microsoft)
4: Market-basket Analysis ( )
Data Mining
Data Mining Data Mining Data Mining Data Mining Data Mining Data Mining Rule Induction SONN
Data Mining 1960..,, Data
Data Mining ( ) "Data mining,,. (Statistics) (mathematical) (Neural Networks) (pattern recognition technologies )." (Gartner Group) Data mining (unknown), (actionable information). ( Aaron Zornes, The META Group )
Data Mining ( ) (DW) (DBM, CRM)
Data Mining (Descriptive Modeling) / ( ). (Unsupervised learning)., (Clustering/Segmentation).,. (Predictive Modeling) ( ). (Supervised learning). (Classification), (Value Prediction)., (Rule Induction), (,,, Probit).
Data Mining ( ) Predictive Descriptive ANN CBR Tree Induction Regression Association Rule Clustering Classification Estimation Affinity Grouping Clustering (Kohonen Network) (ANN: Artificial Neural Network, CBR: Case-based Reasoning)
Data Mining Scoring V isualization Transform ation and reduction Data mining Patters / m odel Selection and sampling Preprocessing and cleaning Cleaned data Transformed data Evaluation User Target data Performance system
Data Mining ( ) ( )., ( ) (t-, ANOVA, ) ( ) (Logistic Regression, Probit, MDA)
Data Mining??? (.)?.?
(Artificial Neural Networks)
: ".
( ) (Caudill and Butler, 1992), (Hecht-Nielsen, 1991)
1940, : 1943 (McCulloch) (Pitts).. (Hebb) (, neuron) (Weight).
( ) 1957 (Rosenblatt) (Perceptron) Widrow Adaline(Adaptive Linear). XOR(Exclusive OR). 1980 : Hopfield, Rumelhart, McClelland, (Error Backpropagation)
(ANN:Artificial Neural Network), Input Layer, Hidden Layer, Output Layer Layer Layer Node Node Weight( ), Weight Black Box INPUT PROCESS OUTPUT Neural Network 0 ~ 1 (0.45 DM )
,,.,.,.,. (Black box) -> (local minimum)
XOR(Exclusive OR) ( ). XOR OR, OR, XOR XOR
XOR(Exclusive OR) ( ) X 2 1 2 (0,1) v v (1,1) v v (0,0) (1,0) X X Y 1 1 0 1 0 1 0 1 1 0 0 0 X 1 x1 w1 + x2 w2 > Threshold
( ) Inout X 1 X 2 X 3 w 2 w 3 w 1 summation net n = i= 1 x w i i Output Y(net).... w n 1 X n Y( net) = weight 1 + e net Transfer function (Three-Layer Network)
(Transfer function) Types of Transfer Function Equation Hard Limiter x < 0, y = -1 x 0, y = 1 x < 0, y = 0 Ramping Function 0 x < 1, y = x x > 1, y = 1 Sigmoid Function I 1 y = 1 + exp x Sigmoid Function II x 0, y = 1-1 1 + x x < 0, y = 1 + 1 1 x Hyperbolic Tangent Function x x exp exp y = x exp x + exp y 1-1 Hard Limiter y 1 x 1 1 y = 1+ exp x Sigmoid Function I x y 1 1 x Rapping Function y 1-1 Sigmoid Function II or Hyperbolic tangent Function x
(Transfer function) ( ) Sigmoid Function 0 1. S.. y + 1 0.5 0 net
,
-
( ) Input: X 1 X 2 X 3 Output: Y Model: Y = f(x 1 X 2 X 3 ) X 1 =1 X 2 =-1 X 3 =2 0.2 = 0.5 * 1 0.1*(-1) 0.2 * 2 0.5 0.6-0.1 0.1-0.2 0.7 f(x) = e x / (1 + e x ) f(0.2) = e 0.2 / (1 + e 0.2 ) = 0.55 0.2 f (0.2) = 0.55 0.55 0.9 f (0.9) = 0.71 0.71 Predicted Y = 0.478 0.1-0.2-0.087 f (-0.087) = 0.478 0.478 Suppose Actual Y = 2 Then Prediction Error = (2-0.478) =1.522
. Rule of thumb. ( ½ (2 +1).). ( = ( + )) (Overfitting).
(epoch)
(Decision Tree) / (Inductive Learning or Rule Induction)
(decision rule) (classification) (prediction). (Class), (Attribute).
( ) (node) (arc) (leaf node) 1 2 3 4 5 5 1 2
22~25 22
( )
( )
CHAID CART QUEST,,,, ( ) - F,,,,,, - F (Levene ) (multiway) (binary) (binary)
( ) C5.0 (Quinlan, 1996) (multiple) C&R Tree (CART) (Breiman et al., 1984) 2 (binary) (entropy) ( ) ( ) ( ) / (minrec) - 2 + > minrec 2 > minrec
( ) (Chi-Square statistic) p P- (Gini index) (Entropy index).
(Entropy) Entropy in bits 1 H(A 1 ) = -P*log 2 P - (1-P)*log 2 (1-P) H(A 2 ) = -P*log 2 P - (1-P)*log 2 (1-P) H(A m ) = -P*log 2 P - (1-P)*log 2 (1-P) 0.5 1 Probability P
(Growing the Tree) (Splitting Criteria), (Stopping Rule) (Pruning).,. (Validation) (Gain), (Risk), (Cost) (Classification, Prediction)
(Splitting Criteria) (traget (interval) (regression tree), (categorical variable) (classification tree) (categorical target) - ( (categorical target) ) -F - (Chi-Square statistic) - (Gini index) - (Entropy index) - (Variance reduction) (mean)
,,,...
(Segmentation) ( :, ) (Classification) /. (Interaction)
, (Self-organizing Neural Networks: SONN) (Kohonen Networks)
, (cluster)..,,.
, (,, ).. (k-, SONN)
(Self-organizing Neural Networks: SONN) (Self-organizing Neural Networks: SONN) (Kohonen). (winner take all).,. (Kohonen Network)
SONN. (feedforward flow). ( ).
input vector: x = x, x, x,..., ] [ 1 2 3 x m T m weight vector: T wi = [ wi1, wi 2, wi3,..., wim ], i = 1,2,..., l. (winner take all) best matching node c = arg min same as x i x w w c i =, i = 1,2,..., l min{ i x w i }
(Winner take all) - ( ) - - / W new = Wold + α( X Wold)
2 (a) (b)
2 (a) (b) 1000
2 ( ) (a) (b) 6000 20000
2 ( )
SONN ( / ) :, /
(Association Rule)
(Association Rule) Agrawal et al. 1993 A B A=>B A B. (affinity grouping).
(Association Rule) ( )? EX) Products in Shop Cart (One trip, Together) Window clear 1)? 2)? 3)?? 4)?
(Association Rule) ( ) (transaction) (item) - Market Basket Analysis product service offering. Ex) Point-Of-Sale Transaction customer Set of products 1 2 3,,, Window Cleaner transaction item
(Association Rule) ( ) (Association Rule) Item Item. (Item set A) (Item set B ) ( if A then B : A B. ) - - EX)
( Cross Selling ) ( Inventory Display ) Catalog Design -,,,,
List ID, 1,, 2,, 3, 4, 5,,, 6 Co-occurrence of Product( ) 4 2 3 1 2 2 2 3 0 1 3 3 4 0 2 1 0 0 2 0 2 2 2 0 2
1. -, combination -,,. 2. -.?,.
(Support) - X Y? S = P( X Y ) = X Y ( N ) -
(Confidence) - X Y? C = P( Y X ) = P( X Y P( X ) ) = X X Y - -
( Lift / improvement ) - X Y Y? L = P( Y X P( Y ) ) = P( X Y ) P( X ) P( Y ) = Lift 1 > 1 < 1,
List ID 1,, 2,, 1,, 3, 4, 5,,, 6 50% 50% Transaction => 1,2,5 75 % => 1,4,5 75 % => 1,4,5 100 % * : (100%)
( ) (URL) 45%., 9 10 4 5. A B C.
1.. 2.. 1.. 2.. 3. DBMS. 4..