際際滷

際際滷Share a Scribd company logo

segmentda

Steven Cosgrove, Ph.D.
Steven Cosgrove, Ph.D.

This document discusses DataActiva's approach to customer segmentation. It describes different levels and types of segmentation including mass, segmented, individual, and targeted segmentation. It also covers segmentation description techniques like a priori, cluster-based, and hybrid models. The document provides an overview of cluster analysis techniques including hierarchical, non-hierarchical, k-means, and Ward's methods. It discusses best practices for variable selection, standardization, response style effects, number of clusters, and validity checks in segmentation analysis. Finally, it notes how segmentation can be activated through customer relationship management processes and linked to other data sources.

1 of 24
Download to read offline
Statistical Note Segmentation 息 2015 DataActiva . All rights reserved. This document contains highly confidential information and is the sole property of DataActiva. No part of it may be circulated, quoted, copied or otherwise reproduced without the written approval of DataActiva. DataActivas Customer Segmentation Approach
2 Segmentation And The Fallacy Of Averages Average Offer; Average Communications Average Offer; Average Communications Just give me basic serviceJust give me basic serviceGive me all the bells and whistles Give me all the bells and whistles They are not really cutting edge. They are just after my money. I really do not want those services.
3 Segmentation Levels Mass Segmented Individual Targeted CRM No Differentiation
4 Segmentation Description Techniques A priori segmentation models--generally use as the dependent variable (the basis for segmentation) either product-specific variables (product usage, loyalty) or general customer characteristics (demographic variables) Cluster-based segmentation models--do not assume the segments are know a priori, therefore, they assign customers to a predominate cluster type based upon factor analysis, multi/nominal and cluster analysis Hybrid segmentation models--cluster-based segmentation models can be combined with some a priori bases Flexible segmentation models--differs from a priori segmentation in which segments are predetermined at the outset of the study and the cluster-based segmentation in which the selected segments are based on the clustering analysis because of its flexibility of building up segments based on consumers' response to alternative product offerings
5 Segmentation Analysis Classification procedures for determining memberships in market segments vary markedly according to the specific segmentation model used For a priori approaches it is most likely sorting and cross-tabulation When clustering segmentation is employed, some sort of cluster or factor analysis is generally used Componential and flexible segmentation involve primarily conjoint analysis (hypothetical choice models) and the computer simulation derived from these hypothetical models
6 The Best The best depends on the marketing strategy You want segments You can reach At a reasonable cost Who will respond The hard part is not in segmenting; its in developing a marketing plan for those segments and executing
7 Introduction To Clustering Response based segmentation is often done with statistical procedures called cluster analysis As much art as a science Many ad hoc procedures with little statistical justification Choice of variables and pre-processing leads to different solutions The computer will blindly create clusters even where none exist Choice of variables very important Generally, need to make sure all responses are at least within-case standardized All solutions need to be validated (cluster profiled)
8 Cluster Analysis Assumes the World is Lumpy
9 Will Try To Impose Its Lumpiness On Data Even When The Data IS Not Lumpy.
HierarchicalHierarchical Non- Hierarchical Non- Hierarchical Clustering Types 10
11 Hierarchical Cluster Techniques Single Linkage Shortest Distance Complete Linkage Longest Distance Average Linkage Average Distance
Minimize within cluster variance Very popular; generally gives good results Maximize between cluster variance Wards Methods 12
Nodal methods Selection of objects that serve as focal points or nodes for the clusters Iterative partitioning methods Nodal methods that begin with an initial partition and then update those initial points Seed points can be random, deliberately far apart, or pre- specified Non-Hierarchical Cluster Techniques 13
14 K-means Method 1 1 2 1 1 2 1 2 Initial Centers First Iteration Second Iteration and Convergence
Wards MethodWards Method K-meansK-means Final ClustersFinal Clusters Use Wards method to generate the initial clusters Input the centroids into K-means and let it clean them up Use the k-means as the final clusters *Its more work, though, and increases the analysis cost, so we often dont work this way Hybrids 15
Single most crucial decision Do what is conceptually correct Usually, you have a type of segmentation in mind, such as benefit segmentation, so you use only those variables as the active variables Clusters can then be profiled on the other variables, or illustrative variables Its hard to mix continuous and categorical variables Variables must be actionable Variable Selection 16
If scales are widely different, standardize the variables to the same mean and standard deviation However, if scales are the same, you sometimes want to use the fact that some attributes have more variance than others Using mean substitution or re-coding missing to some neutral value is widely used so that everyone has a score for each variable There are more sophisticated , albeit ,time consuming methods to impute missing values Standardization and Missing Values 17
Some people say everything is important while others say nothing is important Some use the upper end of the scale (yea-sayers) rather than the lower (nay sayers) You can eliminate these effects with within case standardization Also called respondent centered Subtract the mean for that respondent from each score Single centered data is centered around each respondents mean value This removes the elevation effects for that respondent Double centered data is also centered around the mean for all cases That is, it is standardized to the group mean Response Style Effects 18
Various technical criteria Large complex literature on measures The common market research criteria: Few enough clusters to allow complete strategy development Each large enough to warrant strategic attention and to be reachable and defensible Use what makes sense from a marketing point of view Experience is the best guide Validity checks Number Of Clusters 19
Discriminant Analysis Use discriminant analysis to show that you can predict cluster membership at very high levels (90%+) from the variables Split half reliability Split the sample in half and make certain the cluster solutions are similar Wont work on small samples Chaid (Answer Tree) Validity Checks 20
Use discriminant analysis or Chaid to predict group membership You can often reduce the number of attributes that have to be asked in a follow-up Classify the new respondents into the old clusters using the classification functions Mapping to databases Later Classification 21
Customer Relationship Management is a process that is both information- and technology-driven which aims to leverage customer behavioral data, life event triggers and marketing models to efficiently and continuously cultivate customer relationships: Maximize the potential of each customer relationship and generate incremental revenue. Migrate from a seasonal, ad-hoc campaigns to an automated, perpetual marketing process. Evaluate investment within segments, identify revenue potential and marketing opportunities. Customer Relationship Management Real-time marketing execution Data extraction, data mining and analysis Segmentation is Activated by CRM Processes 22
Linkages to other data sources is critical Competitive Forecasts Revenue Forecasts Market Size Demand Side Analysis Market Potential Purchase Behavior Economic Trends Niche Segments Competing Technologies Pricing Analysis/Forecasts Product Opportunities Supply Side Analysis Industry Structure Competitive Positioning Product Pricing Promotion Distribution Innovation Alliances, JVs, M&A Competitive Forecasts Revenue Forecasts Market Size Used for: Externalized Opportunity/ Threat Analysis Demand Side Analysis Market Potential Purchase Behavior Economic Trends Niche Segments Competing Technologies Pricing Analysis/Forecasts Product Opportunities Supply Side Analysis Industry Structure Competitive Positioning Product Pricing Promotion Distribution Innovation Alliances, JVs, M&A Used for: Internalized Strength/ Weakness Analysis 23
Market Segmentation and Data Base Development For this project, commercial & industrial and residential customers and potential consumers were surveyed to segment the market and to create a data base that could to target customers for the successful introduction of new products and services. In addition, customer valuation through identification and analysis of the factors that determine a customers current and potential lifetime value to the company were incorporated into the design. Focus groups were also employed Telecommunications Commercial Feasibility Study This study was used to predict the relative potential among local area businesses for new telecommunication services in the St. Louis area. It was used to identify and profile consumers interested in switching local exchange carriers; determine possible participation rates among consumers by services/service bundles; assess whether additional services would enhance the offering; evaluate optimal pricing strategies; and appraise customer interest by segment to map back to executable market database. Segmentation and Valuation Project in HC The primary objective was to provide a comprehensive plan including: a situation assessment, strategic direction, possible approaches, timelines, and budgets. Our team provided a thorough assessment of the current situation and a complete familiarization with and orientation around the information, techniques, and tools necessary to determine: (1) market segmentation through the use of database creation, maintenance, and analysis to target customers for the successful introduction of marketing initiatives and new products and services; and (2) customer valuation through identification and analysis of the factors that determine a customers current and potential lifetime value to the company. Case Studies 24

More Related Content

segmentda

  • 1. Statistical Note Segmentation 息 2015 DataActiva . All rights reserved. This document contains highly confidential information and is the sole property of DataActiva. No part of it may be circulated, quoted, copied or otherwise reproduced without the written approval of DataActiva. DataActivas Customer Segmentation Approach
  • 2. 2 Segmentation And The Fallacy Of Averages Average Offer; Average Communications Average Offer; Average Communications Just give me basic serviceJust give me basic serviceGive me all the bells and whistles Give me all the bells and whistles They are not really cutting edge. They are just after my money. I really do not want those services.
  • 4. 4 Segmentation Description Techniques A priori segmentation models--generally use as the dependent variable (the basis for segmentation) either product-specific variables (product usage, loyalty) or general customer characteristics (demographic variables) Cluster-based segmentation models--do not assume the segments are know a priori, therefore, they assign customers to a predominate cluster type based upon factor analysis, multi/nominal and cluster analysis Hybrid segmentation models--cluster-based segmentation models can be combined with some a priori bases Flexible segmentation models--differs from a priori segmentation in which segments are predetermined at the outset of the study and the cluster-based segmentation in which the selected segments are based on the clustering analysis because of its flexibility of building up segments based on consumers' response to alternative product offerings
  • 5. 5 Segmentation Analysis Classification procedures for determining memberships in market segments vary markedly according to the specific segmentation model used For a priori approaches it is most likely sorting and cross-tabulation When clustering segmentation is employed, some sort of cluster or factor analysis is generally used Componential and flexible segmentation involve primarily conjoint analysis (hypothetical choice models) and the computer simulation derived from these hypothetical models
  • 6. 6 The Best The best depends on the marketing strategy You want segments You can reach At a reasonable cost Who will respond The hard part is not in segmenting; its in developing a marketing plan for those segments and executing
  • 7. 7 Introduction To Clustering Response based segmentation is often done with statistical procedures called cluster analysis As much art as a science Many ad hoc procedures with little statistical justification Choice of variables and pre-processing leads to different solutions The computer will blindly create clusters even where none exist Choice of variables very important Generally, need to make sure all responses are at least within-case standardized All solutions need to be validated (cluster profiled)
  • 8. 8 Cluster Analysis Assumes the World is Lumpy
  • 9. 9 Will Try To Impose Its Lumpiness On Data Even When The Data IS Not Lumpy.
  • 11. 11 Hierarchical Cluster Techniques Single Linkage Shortest Distance Complete Linkage Longest Distance Average Linkage Average Distance
  • 12. Minimize within cluster variance Very popular; generally gives good results Maximize between cluster variance Wards Methods 12
  • 13. Nodal methods Selection of objects that serve as focal points or nodes for the clusters Iterative partitioning methods Nodal methods that begin with an initial partition and then update those initial points Seed points can be random, deliberately far apart, or pre- specified Non-Hierarchical Cluster Techniques 13
  • 14. 14 K-means Method 1 1 2 1 1 2 1 2 Initial Centers First Iteration Second Iteration and Convergence
  • 15. Wards MethodWards Method K-meansK-means Final ClustersFinal Clusters Use Wards method to generate the initial clusters Input the centroids into K-means and let it clean them up Use the k-means as the final clusters *Its more work, though, and increases the analysis cost, so we often dont work this way Hybrids 15
  • 16. Single most crucial decision Do what is conceptually correct Usually, you have a type of segmentation in mind, such as benefit segmentation, so you use only those variables as the active variables Clusters can then be profiled on the other variables, or illustrative variables Its hard to mix continuous and categorical variables Variables must be actionable Variable Selection 16
  • 17. If scales are widely different, standardize the variables to the same mean and standard deviation However, if scales are the same, you sometimes want to use the fact that some attributes have more variance than others Using mean substitution or re-coding missing to some neutral value is widely used so that everyone has a score for each variable There are more sophisticated , albeit ,time consuming methods to impute missing values Standardization and Missing Values 17
  • 18. Some people say everything is important while others say nothing is important Some use the upper end of the scale (yea-sayers) rather than the lower (nay sayers) You can eliminate these effects with within case standardization Also called respondent centered Subtract the mean for that respondent from each score Single centered data is centered around each respondents mean value This removes the elevation effects for that respondent Double centered data is also centered around the mean for all cases That is, it is standardized to the group mean Response Style Effects 18
  • 19. Various technical criteria Large complex literature on measures The common market research criteria: Few enough clusters to allow complete strategy development Each large enough to warrant strategic attention and to be reachable and defensible Use what makes sense from a marketing point of view Experience is the best guide Validity checks Number Of Clusters 19
  • 20. Discriminant Analysis Use discriminant analysis to show that you can predict cluster membership at very high levels (90%+) from the variables Split half reliability Split the sample in half and make certain the cluster solutions are similar Wont work on small samples Chaid (Answer Tree) Validity Checks 20
  • 21. Use discriminant analysis or Chaid to predict group membership You can often reduce the number of attributes that have to be asked in a follow-up Classify the new respondents into the old clusters using the classification functions Mapping to databases Later Classification 21
  • 22. Customer Relationship Management is a process that is both information- and technology-driven which aims to leverage customer behavioral data, life event triggers and marketing models to efficiently and continuously cultivate customer relationships: Maximize the potential of each customer relationship and generate incremental revenue. Migrate from a seasonal, ad-hoc campaigns to an automated, perpetual marketing process. Evaluate investment within segments, identify revenue potential and marketing opportunities. Customer Relationship Management Real-time marketing execution Data extraction, data mining and analysis Segmentation is Activated by CRM Processes 22
  • 23. Linkages to other data sources is critical Competitive Forecasts Revenue Forecasts Market Size Demand Side Analysis Market Potential Purchase Behavior Economic Trends Niche Segments Competing Technologies Pricing Analysis/Forecasts Product Opportunities Supply Side Analysis Industry Structure Competitive Positioning Product Pricing Promotion Distribution Innovation Alliances, JVs, M&A Competitive Forecasts Revenue Forecasts Market Size Used for: Externalized Opportunity/ Threat Analysis Demand Side Analysis Market Potential Purchase Behavior Economic Trends Niche Segments Competing Technologies Pricing Analysis/Forecasts Product Opportunities Supply Side Analysis Industry Structure Competitive Positioning Product Pricing Promotion Distribution Innovation Alliances, JVs, M&A Used for: Internalized Strength/ Weakness Analysis 23
  • 24. Market Segmentation and Data Base Development For this project, commercial & industrial and residential customers and potential consumers were surveyed to segment the market and to create a data base that could to target customers for the successful introduction of new products and services. In addition, customer valuation through identification and analysis of the factors that determine a customers current and potential lifetime value to the company were incorporated into the design. Focus groups were also employed Telecommunications Commercial Feasibility Study This study was used to predict the relative potential among local area businesses for new telecommunication services in the St. Louis area. It was used to identify and profile consumers interested in switching local exchange carriers; determine possible participation rates among consumers by services/service bundles; assess whether additional services would enhance the offering; evaluate optimal pricing strategies; and appraise customer interest by segment to map back to executable market database. Segmentation and Valuation Project in HC The primary objective was to provide a comprehensive plan including: a situation assessment, strategic direction, possible approaches, timelines, and budgets. Our team provided a thorough assessment of the current situation and a complete familiarization with and orientation around the information, techniques, and tools necessary to determine: (1) market segmentation through the use of database creation, maintenance, and analysis to target customers for the successful introduction of marketing initiatives and new products and services; and (2) customer valuation through identification and analysis of the factors that determine a customers current and potential lifetime value to the company. Case Studies 24