Given at the Annual NCDM Conference. An introduction to data mining and 7 Steps to Success.

1. Fundamentals ofData Mining for Marketers James R. Stafford

2. Today’s Agenda - 7 Steps to Better Models Identify the business problem Data audit - what data types are most useful and how much do I need? Exploratory data analysis Data quality - how to deal with missing data and outliers Identifying your most predictive variables Transforming your variables Choose the best modeling approach Make sure the model makes sense Model validation - the Melatonin of modeling When to re-build your model

4. Attrition/Lapse/Churn

5. Reactivation

6. Lifetime Value

7. Profitability

10. Frequency

11. Monetary

12. Products PurchasedThe most important data for modeling!

14. Recency

15. Frequency

16. Monetary

17. Products Purchased

18. Demographics

19. Age

20. Home Ownership

21. DependentsLifestyle Type of Car Hobbies Travel Preferences The most important data for customer profiling and building acquisition models!

23. Age

24. Home Ownership

25. Dependents

26. Income

27. Type of Car

28. Travel Preferences

29. SIC

30. Employees

31. Acquired from another source

32. Specific or inferred

33. Actual and reported by individual/household

34. Modeled after similar profiles

35. Pct data specific or inferred varies

37. When to sample

38. too many records

39. test campaign to get response

40. withhold some for model validation

42. C = confidence level (1.96 for 95% confidence)

43. E = acceptable error bound (0.001 = 0.1% response rate)

44. P = response rate from full file (e.g., 0.03=3%)

46. How much data do I need? --10,000 records & 3.0% RR 3.0% 3.3% 2.7% (95 times out of 100!)

47. What’s the minimum sample size I need to get 2.9 % <=> 3.1%? 3.0% 3.1% 2.9% 112,000 (95 times out of 100!)

50. Use/recode: -999, may be meaningful, e.g., lots of missing data can be important in fraud detection

51. Substitute - mean, median or mode

52. Delete records from analysis

53. Outliers - data outside of reasonable bounds

54. customer age = 170

55. customer balance = $1.5M ($10,000 = other max value)

56. identify with plots: frequency distributions/histograms

58. Effect of outliers LTV + + + + Xo + + + + + + + + + + + + + Outliers + + + + Number of purchases

59. Effect of outliers LTV + + + Xcorrect + Xoutlier + + + + + + + + + + + + + Number of purchases

61. mean

62. median

63. mode

64. ASCII string (sex, lifestyle code, …)

66. Cross-tabs

67. Scatter plots

68. Stepwise regression

70. Scatter plot LTV no relationship + + + + + + + + + + + + + AvgHH Income LTV positive relationship + + + + + + + + + + + + + + + + + + + + + + + AvgHH Income

72. 1st regression finds the single most predictive

73. 2nd regression keeps the most predictive and finds the 2nd most predictive that is significant given the presence of the 1st

76. Based on characteristics of the data

77. ZIP - SCF/1st 3 characters

78. Phone number - area code

79. Birth date - age

80. Purchase date - month purchased==>seasonality

81. Purchase date - months since last purchase

83. Variable transformations LTV + + + X + + + + + + + + + + + + + + Number of purchases

84. Variable transformations X2 LTV + + + X + + + + + + + + + + + + + + Number of purchases

85. Variable transformations LTV X + + + + + + + + + + + + + + + + + Household Income

86. Variable transformations LTV X + + ln(x)…or,ln(HHInc) + + + + + + + + + + + + + + + Household Income

88. RFM: non-statistical segmentation scheme

89. CHAID: segmentation scheme

90. Regression: statistical analysis

92. Which purchasers of product A will also purchase product B?

93. Who is most likely to lapse?

94. Predict yes/no, response/non-response

95. Output is binary, e.g., 0/1

96. Predict amount spent

97. Output is numerical, e.g., $2,563

98. What is the expected lifetime value of my customers?

100. Logistic regression

101. CHAID

102. Neural nets

103. RFM

104. CHAID

105. Linear regression

106. Logistic regression

109. Declining attendance

110. Expanding to new stadium with more seats

111. Marketing Objectives

112. Up-sell: Mini-plan to Season ticket holders

114. Appraise results - gains chart for our best model

115. Does the model make sense? Most Important Variables Cluster code Home value Age-male Home value>=100K # of HHs # of seats

116. Does the model make sense -- what do my customers look like?

117. Does the model make sense -- what do my customers look like?

118. Does the model make sense -- what do my customers look like?

119. Does the model make sense -- what do my customers look like?

120. PRIZM cluster composition for segments Modeled C1 C2 S1 S2 S3 U1 U3 Segment 1 1.6 2.4 31.2 4.0 12.8 12.0 34.4 2 2.4 16.356.5 1.6 0.8 13.7 4.0 10 5.5 28.4 11.0 5.5 18.1 1.6 4.7 19 2.8 2.8 10.1 32.1 4.6 2.8 0.0 20 5.5 0.9 18.4 22.9 2.8 0.0 0.0 TOTAL 6.0 9.5 24.0 14.9 11.0 6.1 4.9

122. Multi-lingual

123. Dense/urban

124. Home & apartment renters

125. High % of singles

126. High % of single parents

127. High unemployment

128. Lowest income group

129. Wealthy whites, Asians and Arabic

130. High spending levels

131. Highest income

132. High education

134. Use different media/marketing approaches to reach them

135. Buy prospect lists based on best segment profiles

136. Develop retention and prospecting plans with customized offers (e.g., free CD’s based on their particular tastes in music)===>> improved customer up-sell and retention and better prospecting!

137. EliteSuburbs UrbanCores PRIZM cluster composition for segments Modeled C1 C2 S1 S2 S3 U1 U3 Segment 1 1.6 2.4 31.2 4.0 12.8 12.0 34.4 2 2.4 16.356.5 1.6 0.8 13.7 4.0 10 5.5 28.4 11.0 5.5 18.1 1.6 4.7 19 2.8 2.8 10.1 32.1 4.6 2.8 0.0 20 5.5 0.9 18.4 22.9 2.8 0.0 0.0 TOTAL 6.0 9.5 24.0 14.9 11.0 6.1 4.9 Top demi-decile, i.e., those most likely to become season ticket holders

138. Potential marketing plans

140. error distribution

141. out-of-sample validation

142. cross-validation (good for small data bases)

143. In the field

145. Out-of-sample validation Training File ProbResponse =.003*HHINC+.1*NumPurch FullCustomerFile randomly extracted Training file lift chart compare Validation File Validation file lift chart

146. PC validation Training Data Validation Data

148. When significant events occur

149. new competitive product introductionsLocal ISP internet access v. cable access/DSL

151. When significant events occur

152. new competitive product introductions

153. significant price changesDigital photography

155. When significant events occur

156. new competitive product introductions

157. significant price changes

158. changes in the economyDemand for luxury products (boats) in a recession

160. When significant events occur

161. new competitive product introductions

162. significant price changes

163. changes in the economy

164. news events which affect attitudes about your product -- for better or worseMortgage bankers

165. Summary - 7 Steps to Better Models Identify the business problem Data audit - what data types are most useful and how much do I need? Exploratory data analysis (Data visualization!) Data quality - how to deal with missing data and outliers Identifying your most predictive variables Transforming your variables Choose the best modeling approach - a cookbook approach Make sure the model makes sense Model validation - the Melatonin of modeling Model maintenance

166. Fundamentals ofPredictive Customer Modeling James R. Stafford