(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Customer Experience
Download as PPTX, PDF2 likes4,346 views
The document outlines a seminar from Absolutdata focusing on improving classification methodologies for better customer segmentation through a revised quiz approach. It discusses issues with the original lengthy questionnaire, introduces a shorter version with more predictive questions, and emphasizes the need for a probabilistic classification method to enhance prediction accuracy and user experience. The implications also highlight the importance of aligning questions with business objectives to significantly improve both client and customer interactions.
(MRSI- 2/3) Re-engineering Segmentation methodologies for an Enriching Customer Experience
- 1. © Absolutdata 2014 Proprietary and Confidential Chicago New York London Dubai New Delhi Bangalore SingaporeSan Francisco www.absolutdata.com April 30, 2014 XXIII ANNUAL MARKET RESEARCH SEMINAR Re-engineering Classification Methodologies for an Enriching Customer Experience
- 2. © Absolutdata 2014 Proprietary and Confidential 2 Quiz! Who am I? Am I a Bipede? (Yes/No) Do I possess opposable thumbs? (Yes/No) Am I a Herbivore? (Always/Sometimes/Never) Can I climb trees? (Yes/No) Can I fly? (Yes/No) Yes Sometimes Yes Yes Yes Are you a …CHIMP HUMAN ??
- 3. © Absolutdata 2014 Proprietary and Confidential 3 Am I a Bipede? (Yes/No) Do I possess opposable thumbs? (Yes/No) Am I a Herbivore? (Always/Sometimes/Never) Can I climb trees? (Yes/No) Can I fly? (Yes/No) Quiz! Who am I? Irrelevant / Ambiguous Questions Incorrect Classification Incorrect Scale CHIMP HUMAN Yes Sometimes Yes Yes Yes
- 4. © Absolutdata 2014 Proprietary and Confidential 4 A US based social networking site, which provides unique web- based product offerings to its customers; based on their behavior and how they perceive themselves. Our Client
- 5. © Absolutdata 2014 Proprietary and Confidential 5 A look into Client’s existing classification approach Online survey (quiz) on website sign up Identify customer segments User persona distribution in identified segments Online Quiz Administer 8 questions with statements for each of the 12 pre-defined segments. Response captured in ‘select all’ that apply format Tedious questionnaire Captures dichotomous response Classification Approach User Segments Classification based on highest tally of responses for each segment Unbreakable ties in classification Unable to sharply capture every facet of user’s personality Not a proprietary classification approach
- 6. © Absolutdata 2014 Proprietary and Confidential 6 Revise the data collection instrument Improve classification methodology Administer a quiz with a shorter set of questions Probable rescaling of quiz questions Classify the respondents accurately into their primary, secondary and tertiary segments Identify the best typing tool based on statistical evidence as well as strategic input The study objectives were, therefore, two-fold:-
- 7. © Absolutdata 2014 Proprietary and Confidential 7 Administer survey Analyze Data and Redesign quiz to be hosted on website Develop classification algorithm to accurately identify user segments A sample of 2000 collected, split equally between males and females Data captured on both ranking and select all scale Introduced attitudinal and behavioral questions to capture enriched data Algorithm to derive segments from new set of questions Identify respondent segments based on existing approach to be taken as benchmark Identify new set of questions for accurate classification and shorten the length of quiz Our approach to address these objectives:-
- 8. © Absolutdata 2014 Proprietary and Confidential 8 Shortening the quiz Revisit the quiz originally used and redesign it to a shorter version
- 9. © Absolutdata 2014 Proprietary and Confidential 9 Deriving 4 most predictive questions from the original quiz A shorter quiz (with only 72 statements from original 144 statements) Arriving at a subset of original eight questions Using descriptive statistics (cross-tabulations) Incorporating client inputs to identify priority as per business sense Identifying relevant questions to be retained
- 10. © Absolutdata 2014 Proprietary and Confidential 10 Low accuracy of segments identification from subset of 4 identified questions… Build a robust classification approach based on statistical evidence Provide a Proprietary Approach to the Client Deterministic to Probabilistic Classification Approach
- 11. © Absolutdata 2014 Proprietary and Confidential 11 Improved classification using 6 questions capturing every facet of user’s personality Reduced chances of ties as in Deterministic approach A proprietary method with set of coefficients defining user segments Build algorithm using predictive approach for segment identification Multinomial Logistic Regression (MNL) Multiple iterations using different set of variables to maximize prediction accuracy Introducing new attitudinal and behavioral questions Factoring the new questions to find the most representative statements Probabilistic classification approach
- 12. © Absolutdata 2014 Proprietary and Confidential 12 Improve prediction accuracy achieved in probabilistic approach Need to further improve prediction accuracy to match with original segment classification… Rescaling Quiz Responses and Questions
- 13. © Absolutdata 2014 Proprietary and Confidential 13 Richer insights from the same data Distribution of percentages as per classification Improved prediction accuracy Rescaling Change “Select any 3” response in original quiz to “Rank top 3” response Change data captured on “5 point agreement scale” to “Select all that apply” response Questions Responses Improved prediction accuracy User friendly experience Answering the quiz is much easier Rescaling
- 14. © Absolutdata 2014 Proprietary and Confidential 14 Build a robust and accurate user classification in line with business objectives Probabilistic model predicted segments not ALL aligned with user responses… Apply Business Rules
- 15. © Absolutdata 2014 Proprietary and Confidential 15 Business Rules Probabilistic model solutions triangulated with Deterministic approach Remove any sample bias in estimating probability coefficients Overall improved prediction accuracy Build a mixed approach for segment identification using MNL coefficients and segment tally scores Calculate tally scores for respondent MNL identified segments If a predicted segment has low tally score (below cutoff) it is replaced with next higher tally score segment
- 16. © Absolutdata 2014 Proprietary and Confidential 16 Shorter version of quiz Original quiz with 8 questions revised to 6 questions Proprietary classification tool Coefficients for each user code Classification distribution amongst top 3 segments top 3 segments with their weightings identified Final Deliverables (1/3)
- 17. © Absolutdata 2014 Proprietary and Confidential 17 Shorter version of quiz Original quiz with 8 questions revised to 6 questions Proprietary classification tool Coefficients for each user code Classification distribution amongst top 3 segments top 3 segments with their weightings identified Final Deliverables (2/3)
- 18. © Absolutdata 2014 Proprietary and Confidential 18 Shorter version of quiz Original quiz with 8 questions revised to 6 questions Proprietary classification tool Coefficients for each user code Classification distribution amongst top 3 segments top 3 segments with their weightings identified Final Deliverables (3/3)
- 19. © Absolutdata 2014 Proprietary and Confidential 19 85-95% 78-85% 70-78% 55-70% Sub-set of original Quiz Probabilistic Approach Rescaling Business Rules Selecting four of the most predictive out of eight questions Added an attitudinal and behavioral question, and moving from a tally approach to an MNL solution Changing the scales to a ranking scale and a select all scale for attitudinal & behavioral questions Combining business sense with statistical evidence Prediction Accuracy
- 20. © Absolutdata 2014 Proprietary and Confidential 20 Implications on Indian MR Industry Importance of asking relevant questions (in tandem with business objective to avoid diluting the business implications) Developing the right questions (to accurately convert business objectives into research hypothesis) Agency Client
- 21. © Absolutdata 2014 Proprietary and Confidential 21 Implications for Customers WIN WIN Enhanced experience for the customer The response time for the user is considerably reduced with fewer number of questions An appropriate scale allows the user to respond effectively