(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgies presentation
Download as PPTX, PDF0 likes5,448 views
The document discusses strategic country clustering (SCC) using ensemble clustering methodologies to help multinational companies (MNCs) navigate complex business environments by identifying and developing homogeneous markets. It presents two approaches to SCC—top down and bottom up—with the latter addressing challenges like business alignment and data integration, yielding more transparent and robust solutions. The application of SCC is highlighted with examples in the consumer packaged goods industry, demonstrating its impact on market research efficiency and strategic marketing efforts.
(MRSI - 3/3) Strategic country clusters using ensemble clustering methodolgies presentation
- 1. © Absolutdata 2014 Proprietary and Confidential Chicago New York London Dubai New Delhi Bangalore SingaporeSan Francisco www.absolutdata.com April 29, 2014 Strategic Country Clusters using Ensemble Clustering Methodologies
- 2. © Absolutdata 2014 Proprietary and Confidential 2 Agenda Deep Dive Approaches & Benefits Business Need Impact Way Forward
- 3. © Absolutdata 2014 Proprietary and Confidential 3 Increasingly complex business environment demands MNCs to… Homogeneous markets for optimal strategy execution Holistic view for robust decision making Lessons from similar markets to shape future growth Learn Strategic Country Clustering (SCC) addresses these challenges Identify Develop
- 4. © Absolutdata 2014 Proprietary and Confidential 4 Complexity Data Variability Process Standardizati on Technology Intervention People Expertise Country Clustering Customer Lifetime Value Segmentatio n Marketing Mix Modeling Response Modeling Brand, Portfolio Optimization Retention and Churn Complexity Tactical Impact Strategic Marketing Mix Modeling Response Modeling Retention and Churn Brand, Portfolio Optimization Customer Lifetime Value Segmentation Assessment of analytics applications HighLow Country Clustering Impact Tactical Strategic More Complexity Less However, SCC is a complex analytical undertaking but delivers high strategic impact
- 5. © Absolutdata 2014 Proprietary and Confidential 5 To develop SCC solution we assessed two approaches: 2. Bottom Up: Generates country clusters by aggregating multiple axes solutions 1. Top Down: Generates country clusters by taking into account all the clustering variables at once Top Down Approach Bottom Up Approach
- 6. © Absolutdata 2014 Proprietary and Confidential 6 Top Down Clustering Approach Economy Demographics Usage and Attitude K-Means Clustering Final Clustering Solution Top Down Approach
- 7. © Absolutdata 2014 Proprietary and Confidential 7 Challenge Lack of Business Alignment Too Data Driven However, Top Down clustering approach throws up significant challenges and its corresponding complexity Description Complexity Misalignment with Business Data Integration across multiple parameters Information loss Data Structuring Lack of Transparency Black Box Solution Hard to understand for clients 1 2 3
- 8. © Absolutdata 2014 Proprietary and Confidential 8 Bottom Up clustering approach addresses afore mentioned challenges Bottom Up Approach Economy Demographics Usage and Attitude K-Means Clustering K-Means Clustering K-Means Clustering Economy Country Clusters Demographics Country Clusters Usage and Attitude Ensemble Clustering Method Final Clustering Solution Assigned Dimension Weights
- 9. © Absolutdata 2014 Proprietary and Confidential 9 Bottom up approaches are uniquely positioned to reduce the specific complexities of SCC Challenge Lack of Business Alignment Easier “buy-in” from the client Bottom Up approach benefits Data Integration across multiple parameters Logically segregated axis solutions Lack of Transparency Increased Client Confidence with “white box” solution 1 2 3 A robust, logical, and transparent country clustering solution aligned with client objectives
- 10. © Absolutdata 2014 Proprietary and Confidential 10 However, success of Bottom Up clustering is dependent on Ensemble approaches; Many approaches exist and choosing the right one is critical Take axis solutions as categorical features and run a clustering algorithm on them Feature Based Clustering Create a membership matrix Run a secondary cluster analysis on the membership values using k means Sawtooth Software Algorithm Create a similarity matrix for each partition Take an average over all similarity matrices to create similarity index matrix Run a suitable clustering algorithm to arrive at final clustering solution Cluster-based Similarity Partitioning Algorithm (CSPA)
- 11. © Absolutdata 2014 Proprietary and Confidential 11 But, all these approaches are inadequate on their own. Challenges Faced No. of observations Not suitable for lower no. of observations Axis Weights No scope for weighing the axis solutions Business Alignment Irreverence to Business objectives 1 2 3 Solution Modified version of CSPA where we can use weights to align the final solution with client’s strategic business objectives
- 12. © Absolutdata 2014 Proprietary and Confidential 12 Affinity Matrix Generation Create a similarity matrix for each axis Assign weights to each axis Take sum-product of the weights and similarity matrices to create an affinity matrix Cluster based Affinity Partitioning Algorithm (CAPA) We improved CSPA ensemble approach further Robust and Transparent output CAPA - Steps: Affinity based Grouping Clusters formed on the basis of affinities between countries. An iterative process with each stage being identified within a threshold defined by: Threshold = 100% - (Minimum of dimensions weights(Wi) Final Solution Strategic Country Clusters for Effective Decision Making
- 13. © Absolutdata 2014 Proprietary and Confidential 13 1. Argentina Axis Weights Demographics 15% Economy 20% Category 50% U&A 15% Affinity Matrix Generation Below illustrated CAPA – 1/4 Green indicates presence of both countries in the same cluster in a given axis while Red indicates otherwise Note:- The affinity scores are calculated considering only the green ones 2. Brazil
- 14. © Absolutdata 2014 Proprietary and Confidential 14 Below illustrated CAPA – 2/4 Affinity Score = Demographics – 15% + Economy – 20% + Category – 50% = 85% Similarly, affinity scores are calculated for all country combinations; generating a matrix of affinity scores On the basis of this matrix, countries with similar scores are grouped together 1. Argentina Affinity Matrix Generation 2. Brazil
- 15. © Absolutdata 2014 Proprietary and Confidential 15 Below illustrated CAPA – 3/4 Argentina Brazil 85% Greece Portugal 70% France35% 65% 50%20% Cluster 1 Cluster 2 France – Cluster 2 Affinity = 115%< New country – France is introduced and cluster affinities calculated France – Cluster 1 Affinity = 55% Affinity based Grouping
- 16. © Absolutdata 2014 Proprietary and Confidential 16 Below illustrated CAPA – 4/4 Argentina Brazil 85% Greece Portugal 70% France 65% 50% Cluster 2 Affinity based Grouping Based on cluster affinities France becomes part of Cluster 2 Cluster 1 Following the same process, we arrive at the final SCC solution
- 17. © Absolutdata 2014 Proprietary and Confidential 17 MDI is a measure which captures the relative maturity of markets Specific strategies can be customised based on the development stage of the market Market Development Index We have even leveraged SCC results in variety of applications It is a model used to enhance the cluster solution by adding other countries into the clustering solution This helps in understanding the characteristics of countries that were not even present in the original solution Cluster Predictive Model Used to estimate growth patterns for a country based on several controllable and uncontrollable factors Enables the decision makers to devise effective long term strategies Cluster Progression Simulator
- 18. © Absolutdata 2014 Proprietary and Confidential 18 Tactical Strategic Illustrative Business Impact in CPG Industry Market Research Efficiency Reduced market research expenditure by fielding research in India and projecting results to all countries in same cluster New Product Development Focused development on sugarized kid- friendly flavors and the small sizes preferred by Cluster 1 consumers Advertisement Planning Aired new “Bring on the Fun” TV advertising campaign, targeting kids/young adults Shifted funds from print/radio to digital channels Spends Reallocation Shifted $27 Million in Marketing expenditure across clustersTrade Spending Eliminated multi-pack trade promotions Trial Generation Started sampling and in- store display programs SCC is significantly impacting global planning process
- 19. © Absolutdata 2014 Proprietary and Confidential 19 Further, we are exploring ways to expand the horizon of country clustering Identify trends in countries ahead in progression curve and adapt strategies to develop a category in a given country Discover evolutionary trends of product categories and attributes affecting their usage Market Evolution Study Classify regions within a country based on lessons from global clustering Predict the grouping of a new region based on a given clustering Regional Clustering SCC being a scalable concept, can be leveraged by MNC’s across industries (e.g. Banking, Auto, Manufacturing etc.) in their planning process Extension across Industries