Customer Segmentation and Clustering Using SAS Enterprise Miner, Third Edition

Understanding your customers is the key to your companyâ€TMs success! Segmentation is one of the first and most basic machine learning methods. It can be used by companies to understand their customers better, boost relevance of marketing messaging, and increase efficacy of predictive models. In Customer Segmentation and Clustering Using SAS Enterprise Miner, Third Edition, Randy Collica explains, in step-by-step fashion, the most commonly available techniques for segmentation using the powerful data mining software SAS Enterprise Miner. A working guide that uses real-world data, this new edition will show you how to segment customers more intelligently and achieve the one-to-one customer relationship that your business needs. Step-by-step examples and exercises, using a number of machine learning and data mining techniques, clearly illustrate the concepts of segmentation and clustering in the context of customer relationship management. The book includes four parts, each of which increases in complexity. Part 1 reviews the basics of segmentation and clustering at an introductory level, providing examples from a variety of industries. Part 2 offers an in-depth treatment of segmentation with practical topics, such as when and how to update your models. Part 3 goes beyond traditional segmentation practices to introduce recommended strategies for clustering product affinities, handling missing data, and incorporating textual records into your predictive model with SAS Text Miner. Finally, part 4 takes segmentation to a new level with advanced techniques, such as clustering of product associations, developing segmentation-scoring models from customer survey data, combining segmentations using ensemble segmentation, and segmentation of customer transactions. New to the third edition is a chapter that focuses on predictive models within microsegments and combined segments, and a new parallel process technique is introduced using SAS Factory Miner. In addition, all examples have been updated to the latest version of SAS Enterprise Miner. Dedication......Page 3 Contents......Page 5 Foreward to the Second Edition......Page 11 Foreword to the First Edition......Page 13 What’s New in This Edition......Page 15 Part 2: Segmentation Galore......Page 16 Part 4: Advanced Segmentation Applications......Page 17 Additional Help......Page 18 Publish with SAS......Page 19 About The Author......Page 21 Acknowledgments......Page 23 The Basics......Page 25 1.1 What Is Segmentation in the Context of CRM?......Page 27 1.2.1 Customer Profiling......Page 28 Table 1.3 Example of Customer Profiling in NW U.S. Region (Profile by Major Metro/3-digit Postal Code—Only Top 8 Rows Shown)......Page 29 1.2.2 Customer Likeness Clustering......Page 30 1.2.4 Purchase Affinity Clustering......Page 31 Figure 1.3 Hertzsprung-Russell Diagram: Star Clusters by Temperature and Luminosity......Page 32 1.4 Segmentation as a CRM Tool......Page 33 Code Used to Generate Output in Figure 1.5......Page 34 Figure 1.5 Simple Segmentation of Risk Index by Revenue Class......Page 35 1.5 References......Page 36 2.1 Mass Customization Instead of Mass Marketing......Page 37 Table 2.1 World Internet Usage and Population Statistics (latest data 2015)......Page 38 2.2 Specialized Promotions or Communications by Segment Groups......Page 39 Table 2.2 Three Segments of Customers for Marketing......Page 40 Process Flow Table: Data Assay Project......Page 41 Figure 2.1 Starting SAS Enterprise Miner......Page 42 Figure 2.2a Folder Location for Projects......Page 43 Figure 2.4 Example 2.1 Data Assay—BUYTEST Input Data Source......Page 44 Figure 2.5 Data Assay—Variables in the BUYTEST Data Set Node......Page 45 Figure 2.7 Sample Distribution of PURCHTOT in the BUYTEST Data Set......Page 46 Figure 2.8 INCOME Distribution in the BUYTEST Data Set......Page 47 Figure 2.11 MultiPlot Node Results of AGE versus RESPOND Variables......Page 48 Figure 2.12 Categorical Variable Stats for the BUYTEST Data Set......Page 49 Figure 2.13 Numeric Variable Stats for the BUYTEST Data Set......Page 50 Code Used to Generate Output in Figure 2.15......Page 51 Code Used to Generate Output in Figure 2.16......Page 52 Figure 2.16 Crosstabulation Results of Residence by Married......Page 53 Figure 2.17 Segment Profile Node Results......Page 54 Figure 2.18 Variable Worth Plot of CLIMATE in Segment Profile Results Window......Page 55 2.3.3 Additional Exercise......Page 56 2.4 References......Page 57 3.1 What Is Similar and What Is Not......Page 59 Table 3.1 Database Field Descriptions with Differing Attributes......Page 60 Table 3.3 Simple Distance Matrix for Age Variable......Page 61 Table 3.4 Distance Matrix Computed Using the SAS DISTANCE Procedure......Page 62 Figure 3.2a Distance Plot of Data in Table 3.5......Page 63 Figure 3.3 Illustration of Distance Measurement from Inner Product......Page 64 Table 3.6 Distance Metrics Defined......Page 65 Table 3.8 State Variable Dummy Transformations......Page 66 3.3 What Is Clustering? The k-Means Algorithm and Variations......Page 67 3.3.2 The Agglomerative Algorithm......Page 69 Figure 3.5 Hypothetical Example of Dendrogram Forming Hierarchical Clusters......Page 70 Figure 3.6 Hierarchical Clusters in Simple Customer IDs from Table 3.7......Page 71 Figure 3.7 Three Common Methods for Measuring Cluster Distances......Page 72 3.4 References......Page 73 Segmentation Galore......Page 75 4.1 Introduction to Cell-Based Segmentation......Page 77 Recency......Page 78 Table 4.2 Customers A and B Purchases in One Year......Page 79 Figure 4.1 RFM Cell Pictorial Description......Page 80 Process Flow Table: RFM Cell Development......Page 81 Figure 4.3 SAS Code Node Variables List......Page 83 Figure 4.4 SAS Code Explore Variable PCTL_VALUE24......Page 84 Figure 4.6 RFM Cell Develop Project—SAS Code Node Entry (Continued)......Page 85 4.4 Tree-Based Segmentation Using RFM......Page 86 Figure 4.7a Metadata Node Train Variables Settings......Page 87 Figure 4.8 Advanced Property Sheet Tree Settings for RFM Segmentation......Page 88 Figure 4.9 Completed RFM Segmentation Process Flow Diagram......Page 89 Figure 4.10 RFM Cell—Respond Segmentation Decision Tree Viewer......Page 90 Figure 4.11 Score Node Bar Chart Results......Page 91 4.5 Using RFM and CRM—Customer Distinction......Page 92 4.6 Additional Exercise......Page 93 4.8 Additional Reading......Page 94 5.1 Motivation for Clustering of Customer Attributes: Beginning CRM......Page 95 Process Flow Table 1: B-B Segmentation......Page 96 Figure 5.1 Input Data Source Wizard Window......Page 97 Figure 5.2 Approximate Distribution of Estimated Spend Dollars in Distribution Explorer Node......Page 98 Figure 5.3 EST_SPEND Transformed to Approximate Normal Distribution......Page 99 Figure 5.4 All Transformed Variables in the Node......Page 100 Figure 5.5 Filtering Out Outlier Values......Page 101 Figure 5.7 Cluster Node Initial Properties Settings and Flow Diagram......Page 102 Figure 5.8 Cluster Node Default Results Window......Page 103 Figure 5.10 Cluster Node Distance Plot for the Five Clusters......Page 104 Figure 5.11 Segment Profile Results Window......Page 105 Figure 5.12 Segment Profile Expanded Window View......Page 106 5.3 Using a Decision Tree to Create Cluster Segments......Page 107 Figure 5.14 Illustration of a Binary Split and Concept of Purity......Page 108 Figure 5.15 StatExplore Results Window of Target RFM Variable......Page 109 Figure 5.16 SAS Code Node Property Dialog Box Window......Page 110 Figure 5.17 SAS Code Node to Reduce RFM Levels......Page 111 Figures 5.19 Decision Tree Node Property Sheet Window Settings......Page 112 Figure 5.20 Decision Tree Node Results Window......Page 113 Figure 5.22 Model Comparison Node’s Results......Page 114 5.5 Additional Reading......Page 115 6.2 Representing Many Attributes in Multi-dimensions......Page 117 Figure 6.1 SAS Macro to Compute the Softmax Function......Page 118 6.3 How Can I Better Understand My Customers of Many Attributes?......Page 121 Process Flow Table: NY Towns Clustering......Page 122 6.4 Data Assay and Profiling......Page 123 Figure 6.5 Correlation Plots of PENETRATION Target Variable......Page 124 Figure 6.7 Inset of Figure 6.6 Showing Variable Importance of HouseMultiFamily......Page 125 Figure 6.8 Property Sheet Settings for the Variable Selection Node......Page 126 Figure 6.9b R-Square Bar Chart of Effects in the Variable Selection (continued)......Page 127 Figure 6.10 Initial Clustering Segment Size Plot......Page 128 Figures 6.12a and 6.12b Second Pass Cluster Mean Statistics and Variable Importance......Page 129 Table 6.1 Eigenvalue of the Covariance Matrix......Page 130 Figure 6.13 NY Towns Cluster Distances Plot......Page 131 6.6 Planning for Customer Attentiveness with Each Segment......Page 132 Figure 6.15a Score Node Imported Data from the Cluster Node......Page 133 Figure 6.16 Results Window of the Score Node: Viewing SAS Code......Page 134 6.8 Additional Exercise......Page 135 6.9 References......Page 136 7.1 What Is the Shelf Life of a Model, and How Can It Affect Your Results?......Page 137 Process Flow Table: Distance Metrics......Page 138 Figure 7.1 Variables to Use in the Cluster Node......Page 139 Figure 7.3 Completed Process Flow Diagram of the Cluster Analysis......Page 140 Figure 7.4 Scored Data Set in the Results Window......Page 141 Figure 7.6 Training Data Distances by Cluster......Page 142 Figure 7.7 Scored Data Distances by Cluster......Page 143 Figure 7.9 New Menu Selections in the SAS Code Results Window......Page 144 Figure 7.10b Comparison Histograms of Scored and Training Data Sets in the SAS Code Node Results Window: Scoring Data Histograms......Page 145 7.3 Testing New Observations and Score Results......Page 146 Figure 7.11 Revised SAS Code Node 2 to Reflect the New Data Set......Page 147 Figure 7.12b Results of SAS Code 2 Node Histogram Plot of Score Data Set CUST_NEWSCORE Scoring: Data Set Histograms......Page 148 7.5 Additional Reading......Page 149 8.1 The Basis of Breaking Up the Data Space......Page 151 8.2 Predicting a Segment Level......Page 152 Process Flow Table 1: Predicting Segments Project......Page 153 Figure 8.3 Clustering Segment Results Window......Page 154 Figure 8.5 Variable Importance Table from Cluster Results......Page 155 Figure 8.6 SAS Code to Generate the Target Variable......Page 156 Figure 8.7 Changing the Role of the Target Variable......Page 157 Figure 8.8 Cluster Segmentation and Predictive Model with Regression......Page 158 Figure 8.9a Interaction Variables to Be Added in the User Term Editor......Page 159 Figure 8.10 Process Flow Diagram for Cluster Segmentation and Initial Regression to Predict Segment Number 4......Page 160 Figure 8.11 Initial Regression Results Window......Page 161 Figure 8.13 Cutoff Node ROC and Precision-Recall Charts......Page 162 8.4 Creating Customer Value Segments......Page 163 Process Flow Table 2: Most Valuable Customers (MVCs)......Page 164 Figure 8.14 Customer Value Data Set Variables in the MVC Example......Page 165 Figure 8.16 Type 3 Statistics from Regression Model (First Pass)......Page 167 Figure 8.18 SAS Code to Compute LTV and NPV from Predicted Revenues......Page 168 8.5 Additional Exercises......Page 169 8.6 References......Page 170 Clustering and the Issue of Missing Data......Page 171 9.1 Missing Data and How It Can Affect Clustering......Page 173 Figure 9.1 Example of Monotonic Missing Data Pattern . = missing, X = nonmissing data......Page 174 Figure 9.2 SAS Code Node to Run PROC MI......Page 175 Process Flow Table 2: Clustering with Missing Data......Page 176 Figure 9.6 Cluster Node Property Sheet Settings......Page 178 Figure 9.7 Clustering Output Results: Distance Plot of Five Clusters......Page 179 Figure 9.9 Second Cluster Distance Plot: Missing Entries with Mean/Median Settings......Page 180 9.4 Methods of Missing Data Imputation......Page 181 Figure 9.12 Cluster Distance Plot of County Clusters......Page 183 Figure 9.14 SAS Code for PROC MI......Page 184 Figure 9.16 Comparison Histograms of Original versus Mean Imputation Cluster Distances......Page 188 Figure 9.17 Comparison Histograms of Original versus Multiple Imputation Cluster......Page 189 Figure 9.18 Comparison Tables of Original Clusters and Multiply-Imputed Clusters (Mean and Standard Errors)......Page 190 9.5 Obtaining Confidence Interval Estimates on Imputed Values......Page 191 Figure 9.21 Bootstrap Macro Output Results......Page 192 9.7 References......Page 193 10.1 Motivation of Estimating Product Affinity by Segment......Page 195 Figure 10.1 Venn Diagram of Product Association Metrics......Page 196 Process Flow Table 1: Binary Product Affinity......Page 197 Figure 10.2 Distribution of Product C......Page 198 Figure 10.3 CUSTOMERS Data Set View Table......Page 199 10.3 Combining Product Affinities by Cluster Segments......Page 200 Figure 10.5 Product Affinities Process Flow with Segment Scoring and Data Merging......Page 201 Figure 10.7 SAS Code for Affinity by Segment Computations......Page 202 Figure 10.9 Product Binary Affinity by Segment......Page 203 10.4 Pros and Cons of Segment Affinity Scores......Page 204 Figure 10.11 Product A, B, and C Quantity Cluster Statistics......Page 205 Figure 10.12 Product A, B, and C Quantity Cluster Distance Plot......Page 206 Figure 10.13 Transformed Product A, B, and C Quantities Clustering......Page 207 Figure 10.15 Cluster Plot of Normalized Products A through C with Range Standardization and Average and Fixed Number of Clusters Set to 9......Page 208 Figure 10.16 SAS Code for Softmax Transformations......Page 209 Figure 10.17 SAS Code for Softmax Transformations (Continued)......Page 210 10.6 Approximating a Graph-Theoretic Approach Using a Decision Tree......Page 211 Figure 10.19 Example of Graph Theory Finding Data Clusters......Page 212 Figure 10.20 Decision Tree Property Settings to Approximate an MST Algorithm......Page 213 Figure 10.21 Decision Tree Model Results (Tree Diagram)......Page 214 Figure 10.23 Decision Tree Model (Tree Rules and Nodes) Partial Output......Page 215 Figure 10.24 SAS Code Node Output of Product Affinity Mean Scores by Decision Tree Node......Page 216 10.7 Using the Product Affinities for Cross-Sell Programs......Page 217 Figure 10.27 Partial Results of Segment Profile Node Output......Page 218 10.8 Additional Exercises......Page 219 10.9 References......Page 220 11.2 What Is a Self-Organizing Map?......Page 221 Figure 11.2 Computer Rendition of a Neuron......Page 222 11.3 Computing and Applying SOM Network Cluster Segments......Page 223 Process Flow Table 1: SOM Segmentation......Page 224 Figure 11.6 Softmax Scoring of the Customer Data Set......Page 226 Figure 11.8 SOM/Kohonen Property Sheet Settings (1st Pass), Changes Marked with Arrows......Page 227 Figure 11.9 SOM Results Output......Page 228 Figure 11.10 SOM Segment Profile Results Output......Page 229 11.4 Comparing Clustering with SOM Segmentation......Page 230 Figure 11.13 SAS Code for Segmentation Comparisons......Page 231 Figure 11.14 Original versus SOM and VQ Segmentation Comparison Results......Page 232 Process Flow Table 2: SOM Segmentation......Page 233 Figure 11.15 Exclude RFM Values in Filter Node......Page 234 Figure 11.16 SOM Cluster Map from Attribute and Demographic Variables......Page 235 Figure 11.17 Variable Selection for Segment Profile Node......Page 236 Figure 11.18 Customer Distinction Process Flow Diagram......Page 237 11.7 References......Page 238 12.1 Background of Textual Data in the Context of CRM......Page 239 Table 12.1 Document-Term Matrix in an Example Seven-Document Set......Page 240 Table 12.3 Example Term Vectors That Indicate Relative Term Weights in Documents in Table 12.2......Page 242 Process Flow Table 1: Text Segmentation—News Stories......Page 243 Figure 12.1 Classification Chart for the Binary HOCKEY Variable......Page 244 Figure 12.2 SAS Code Statements for Precision versus Recall Chart and Table......Page 245 Figure 12.4 ROC Chart from SAS Code Node Macro Statements......Page 246 12.4 Text Document Clustering......Page 247 Figure 12.5 Text Mining Results Window from News Stories......Page 248 Figure 12.6a Completed Process Flow Diagram of News Stories......Page 249 Process Flow Table 2: Text Segmentation—Text Clustering......Page 250 Figure 12.7a and 12.7b Text Mining Clustering and Topic Node Property Sheet Settings Respectively......Page 251 Figure 12.9 Text Topic Node Results Window.......Page 252 Figure 12.10 Concept Link Plot of the Term Buy......Page 253 Figure 12.11 Expanded Concept Link Plot of the Term Sample.......Page 254 Figure 12.13 Completed Text Analytic Process Flow Diagram......Page 255 Figure 12.14 Text Mining Model Account Classification Using MBR......Page 256 12.6 References......Page 257 Clustering of Product Associations......Page 259 Process Flow Table 1: Association Analysis Process Flow......Page 261 Figure 13.1 Association Node Results Window from Data Set Assocs......Page 262 Figure 13.2 Rules Table from Association Node Results Window......Page 263 Figure 13.4 Link Graph from Figure 13.3 Subset to >= 80% Confidence......Page 264 Process Flow Table 2: Market Basket Analysis Process Flow......Page 265 Figure 13.5 Market Basket Node Property Sheet Settings......Page 266 Figure 13.6 Market Basket Node Mapping Hierarchy Settings......Page 267 Figure 13.8 Highlighted Rules in Market Basket Node Results Window......Page 268 Process Flow Table 3: Clustering Association Rules......Page 269 Figure 13.10b Association Node Link Cluster in Closer Detail......Page 271 Figure 13.11 Cluster Node Variable Setting......Page 272 Figure 13.12 Cluster Node Property Sheet Settings......Page 273 Figure 13.14 Cluster Plot Distances......Page 274 13.4 The Business and Technical Side of Clustering Associations......Page 275 13.6 References......Page 276 14.1 Typical Market Research Surveys......Page 277 14.2 Match-back of Survey Responses......Page 278 14.3 Analysis of Survey Responses: An Overview......Page 279 Process Flow Table 1......Page 280 Figure 14.2 Customer Survey Segment Data Set for Predicting Attitudinal Segments......Page 282 Figure 14.3 Transform Node and Three Variables Transformed......Page 283 Figure 14.4 Settings for the Decisions Node......Page 284 Figure 14.5 Logistic Regression Node Output Results Window......Page 285 Figure 14.6 Classification Charts of Adjusted and Non-Adjusted Prior Probability Values in Regression Node Results Window......Page 286 Figure 14.7 Output of the Comparison Node for Logistic Regression Model......Page 287 14.5 Issues with Scoring a Predictive Segmentation on Customer or Prospect Data......Page 288 14.6 Assessing the Confidence of Predicted Segments......Page 289 Figure 14.10 Bootstrap Sampling Methodology......Page 290 Figure 14.11 SAS Enterprise Guide—Assigning a Project Library......Page 291 Figure 14.13a Filter & Sort Node Variables from Score_Score Data Set......Page 293 Figure 14.13b Filter & Sort Node Filter Selection for Score_Score Data Set......Page 294 Figure 14.14 Partial Output from Histogram of SURVEY SEGMENTS 3 Predicted Probabilities and Normal Curve......Page 295 Figure 14.16 Jackboot Macro Partial Output for %Boot and %Bootci......Page 296 Figure 14.18 Specific SAS Code Partial Output for the %Bootsamp Macro......Page 297 14.7 Business Implications for Using Attitudinal Segmentation......Page 298 14.9 References......Page 299 15.1 Survey of Methods of Ensemble Segmentations......Page 301 Figure 15.1 Three Cluster/Segmentation Data Representations......Page 303 Figure 15.2 Flow Diagram of Ensemble Segmentation with Naïve Bayes (Collica, 2015)......Page 304 Figure 15.3 Initial and SOM Ensemble Segmentation Completed Diagrams (from Process Flow Table 1)......Page 305 Figure 15.4 Variable Transforms Using Logs......Page 307 Figure 15.5a Variables Used for Behavioral Cluster Segmentation......Page 308 Figure 15.5b Cluster Node Property Sheet Settings......Page 309 Figure 15.6 Ensemble Cluster Node Property Sheet Settings......Page 310 Figure 15.8 Cluster Map from Behavioral Cluster Segmentation......Page 311 Figure 15.9 Worth Chart by Segment of Segment Profile Node......Page 312 Figure 15.10b Segment Profile Node Worth Chart for Cluster 2......Page 313 Figure 15.11 SOM Node Property Settings for SOM Ensemble Clusters......Page 314 Figure 15.12 SOM Ensemble Clustering Results (Showing Frequency of Clusters)......Page 315 Figure 15.13b GLM Coding of Ensemble Clustering Segments......Page 316 Process Flow Table 2: Ensemble Clustering Method......Page 317 Figure 15.14 HP BN Classier Node Property and Variable Setting......Page 318 Figure 15.15 Naïve Bayes Network Ensemble Cluster Flow Diagram......Page 319 Figure 15.16 HP BN Classifier Node Results Window.......Page 320 Figure 15.17 Metadata Node Settings......Page 321 Figure 15.19 Actual versus Predict SOM Ensemble Segments (partial output shown)......Page 322 Figure 15.20 Data Options in Graph Explore node for 3D bar chart plot.......Page 323 Figure 15.22 Completed SOM Ensemble Cluster Algorithm Process Flow Diagram......Page 324 15.3 Presenting the Business Case Simply from a Complex Analysis......Page 325 15.5 References......Page 326 Figure 16.1 A Typical Time Series......Page 327 Figure 16.2 First Example Time Series Sequence......Page 328 Figure 16.3 Fourier Transform for a Time Series......Page 329 Figure 16.4 Plotting Target versus Input Time Sequences......Page 330 Process Flow Table: Transaction Segmentation......Page 331 Figure 16.5 TS Data Preparation Node Property Settings......Page 332 Figure 16.6 TS Data Preparation Node Results Window......Page 333 Figure 16.8a Directions of Paths for Metric Measures......Page 334 Figure 16.8b Warping Limits for Distance Metrics......Page 335 Figure 16.10 Cluster Dendogram from the Similarity node output results.......Page 336 Figure 16.12 Completed Process Flow Diagram of Similarity Analysis.......Page 337 16.4 Additional Reading:......Page 338 17.2 Automating Segment Models......Page 339 Figure 17.1 Log On to SAS Factory Miner......Page 340 Figure 17.2 Home Screen Expanded for SAS Web Applications......Page 341 Figure 17.3 SAS Factory Miner Project Definition......Page 342 Figure 17.4 Partial window of Data Definition in Factory Miner Project.......Page 343 Figure 17.6 SAS Factory Miner Model Selection......Page 344 Figure 17.7 SAS Factory Miner Results Window (Partial Output Shown)......Page 345 17.3 Other Methods for Combining Segmentations......Page 346 17.5 References and Additional Reading......Page 347 C......Page 349 K......Page 350 P......Page 351 S......Page 352 W......Page 353 Résumé : A working guide that uses real-world data, this step-by-step resource will show you how to segment customers more intelligently and achieve the one-to-one customer relationship that your business needs. -- Edited summary from book