Through Creative Modeling &

Size: px
Start display at page:

Download "Through Creative Modeling &"

Transcription

1 Gaining Superior Results Through Creative Modeling & Segmentation Presented by: Marc Fanelli Experian Corporation DMA - October 28-31, 2001 What You Will Learn... Information about a proven modeling technique which will: Empower you with a total solution for your customer acquisition needs Maximize the size and performance of your prospecting universes Help you establish a life-long stream of fresh prospects for growing your business

2 What We ll Cover... ❶ Defining Analytic Services: Profiling & Modeling (The Baker s s Analogy) ❷ Profiling VS Modeling: A Practical View ❸ Nuts & Bolts: The Modeling Process ❹ Industry Applications ❺ Tips for Success ❻ Brief Case Study Analytic Services: Profiling & Modeling Scientific methods which can improve the results of your database marketing efforts Modeling is an extension of profiling* Profiling is like making a trip to the grocery store; Modeling is the act of preparing your meal *Note: Profiling is being referred to as traditional list selections based upon past knowledge as a result of market research

3 The Baker s s Analogy... Scenario #1 - The Baker Scenario #2 - The Marketer The Challenge Try to recreate fabulous cake eaten at favorite aunt s s holiday party The Challenge Leverage catalog mailing to acquire new customers profitably The Baker s s Analogy... The Baker - Phase I The Marketer - Phase I Action Plan Go to grocery store to buy ingredients Action Plan Call data provider and purchase a list Problem Don t t know exactly what is in the cake; Must purchase everything that could possibly be needed Outcome $100 grocery bill; Very angry spouse Problem Don t t know exactly who represents the best target audience; Must use broad selections (Age 18+, Income above $20k) Outcome Mailed entire USA; Cost $1 billion to mail; Lost millions of $; Very angry boss

4 The Baker s s Analogy... The Baker - Phase II The Marketer - Phase II Action Plan Call aunt and ask what is in the Cake; Purchase only those Items Action Plan Have data provider create a Profile Analysis; Use results to select list Problem Don t t know exact proportions of the ingredients; Aunt won t t divulge secret recipe; Have to wing it Outcome $20 grocery bill; Funny tasting cake Problem More selections more potential customers eliminated; Fewer selections too many potential non-customers retained Outcome Targeted 1,000,000 prospects; Mailing broke even; Must improve The Baker s s Analogy... The Baker - Phase III The Marketer - Phase III Action Plan Send leftover cake to forensic laboratory; Ask scientist to provide a detailed chemical composition breakdown of the cake Problem Solved Outcome $20 grocery bill; Perfect tasting cake Action Plan Have data provider create a customer acquisition Model; Use model to select the list Problem Solved Outcome Model rank orders entire USA from Best Prospect to Worst Prospect ; Mail 1,000,000 Best Prospects ; Profits go through the roof; Ecstatic boss

5 The Baker s s Analogy Profiles provide us with the ingredients; = Models provide us with the recipe; = Profiles determine important attributes, but treat them equally Models assign weights to each attribute to determine relative importance Models allow us to use attributes most efficiently Profiling VS Modeling Mailing Results - Intuitive Selections Income: $50K-$75K Income: $25K-$49K Age: Age: = Responder = Non-Responder

6 Profiling VS Modeling Preparing to mail again Need 4/10 (40%) response rate Can mail top left-most cell only Intuitive selections provide the ingredients, but not the recipe Pros Easy to understand and interpret Safe Relatively simple to explain Cons Limited dimensionality = limited targeting precision Tell you where you have been, not necessarily where you can go Are not predictive Profiling VS Modeling Mailing Results - Model High Scores = Responder = Non-Responder Low Scores Model effectively groups potential responders Can mail as deep as necessary to acquire the majority of the potential responders

7 The Segmentation & Modeling Process Append customer database with attributes: Demographics, activities, hobbies, interests, product purchases, etc Segment customers into sub-populations Identify niche markets within customer database The Segmentation & Modeling Process Extract market segments from prospect database Identify clones of customer market segments within prospecting universe Use models to rank prospects from Best to Worst Market segments are ordered with respect to profitability metrics

8 The Segmentation & Modeling Process Example: Bank Card - Response and Approval Models Apply market segments to prospect database to extract prospect clones of each market segment Customer database Group 1 Group 2 Group 3 Prospect database Utilize customer data, appended demographic / geo-level indicators and a bit of intuition to perform market segmentation to identify sub-populations with customer database Prospect group 1 Prospect group 2 Prospect group 3 The Segmentation & Modeling Process Targeting without Targeting with market segmentation market segmentation Income Income $40 $40 Age Age $20 $20 Miss targets completely Hit targets with precision Separate models = increased target precision Market segments open potential new niche markets Promotional offers can be custom tailored to various market segments

9 The Segmentation & Modeling Process Example: Bank Card - Response and Approval Models Apply market segments to prospect database to extract prospect clones of each market segment Customer database Group 1 Group 2 Group 3 Prospect database Perform market segmentation to identify sub- populations within customer database Prospect group 1 Prospect group 2 Prospect group 3 High Response High Response High Approval Low Approval High Response High Response High Approval Low Approval High Response High Response High Approval Low Approval Low Response Low Response High Approval Low Approval Low Response Low Response High Approval Low Approval Low Response Low Response High Approval Low Approval Rank order prospects with respect to projected net acquisition Industry Applications Financial Institutions Response Approval Activation Catalogers Response Average Order Lifetime Value Publishers Response Pay-up Retention Insurance Companies Lead Generation Conversion Travel and Entertainment Response Conversion Non-Profits Pledge Fulfilled Donor Average Gift Reactivation Telecommunications / Utilities Churn Retail Outlets Store Traffic Average Purchase

10 Tips for Success Clearly define objective Augment valuable transactional data with external demo/psychographic and geo-level data Explore - unlock the power in your database Validate models Test & refine Case Study Client Upscale fine gifts cataloger Objective Improve response rates, tap into large prospecting universe

11 Background Mail approximately 500k per month Select prospects by age, income and other demographic / lifestyle criteria Typical response between 1.00% to 1.20% Response falling off in recent campaigns The Process Identify market segments, or clusters within customer database Apply clusters to recent mail campaign Create response models for each cluster identified Segment prospect universe into tiers of projected responsiveness Launch test campaign Read results; refine system

12 Process Flow Diagram 1. Identify sub- populations within the customer database Customer Group 1 Customer database Customer Group 2 Customer Group 3 3. Score or apply models to rank order prospects with respect to projected response Recent Campaign Files / Prospecting Database 2. Apply market segments to recent campaigns to construct response models for each group Prospect group 1 Prospect group 2 Prospect group 3 High response High response High response Low response Low response Low response Analysis Results Identified four unique clusters in customer database Cluster I Swingin Singles Cluster II Refined Executives Cluster III Country Bumpkins Cluster IV American Family

13 Analysis results Cluster profiles Median Home Value in Area Exact Age of Respondent Home Owner Estimated Household Income % in Area Graduate / Professional Degree Occupation: Business Owner / Self- Employed Marital Status: Married Child(ren) Present in Household Cluster I $163K 48 35% $69K 10% 3% 20% 5% Cluster II $162K 48 80% $79K 12% 8% 85% 57% Cluster III $89K 52 44% $60K 24% 5% 59% 20% Cluster IV $117K 46 75% $75K 22% 19% 95% 80% Cluster I Swingin Singles Cluster II Refined Executives Cluster III Country Bumpkins Cluster IV American Family Analysis Results Observe prior mailing results by cluster Prior Campaign Response Rate % Customers % Prospects Development Index Cluster I 0.77% 24% 35% 69 Cluster II 0.46% 35% 18% 194 Cluster III 1.06% 21% 35% 60 Cluster IV 1.17% 20% 12% 167 High response / high development index productive, unsaturated market High response / low development index productive universe, untapped opportunity Low response / high development index unproductive, saturated market Low response / low development index unproductive, untapped market

14 Analysis Results Construct response models for each cluster and rank into ten tiers (A-J) of performance Cluster I Cluster II Cluster III Cluster IV Projected % Cum. Projected % Cum. Projected % Cum. Projected % Cum. Model Tier Response Lift Response Lift Response Lift Response Lift A 2.70% 251% 1.52% 228% 3.80% 263% 2.67% 130% B 1.54% 175% 0.88% 159% 2.11% 181% 2.23% 100% C 1.01% 129% 0.62% 117% 1.35% 130% 1.77% 85% D 0.77% 97% 0.45% 87% 1.08% 98% 1.36% 68% E 0.53% 71% 0.32% 64% 0.70% 71% 1.04% 52% F 0.43% 52% 0.25% 46% 0.52% 51% 0.83% 39% G 0.28% 35% 0.17% 30% 0.37% 34% 0.82% 29% H 0.18% 21% 0.18% 18% 0.30% 21% 0.53% 18% I 0.08% 10% 0.12% 9% 0.23% 10% 0.49% 10% J 0.04% 0% 0.09% 0% 0.09% 0% 0.14% 0% Summary 0.77% 0% 0.46% 0% 1.06% 0% 1.17% 0% Analysis Results Cluster I: Swingin Singles Model table % Attributes in the Model Effect When compared w/prospects, clients are likely to: Contribution to Model Interest in Fine Arts + Purchase these items 32% Estimated Household Income + Have a higher income 16% Mail Order: Buyer + Order through the mail 15% Outdoor Sports Enthusiast Not participant in outdoor sports 9% Gains chart Model Tier % Cumulative Lift Interest in Fine Arts Estimated Household Income Interest in Outdoor Activities Age Renter A 117% 87% $88K 5% 57 71% B 89% 83% $72K 11% 54 65% C 70% 82% $66K 13% 51 64% D 55% 80% $63K 17% 49 59% E 43% 78% $61K 21% 57 55% F 34% 76% $59K 28% 45 51% G 25% 75% $57K 35% 44 44% H 16% 71% $56K 39% 42 39% I 8% 67% $54K 42% 38 31% J 0% 57% $52K 48% 37 21% Summary 0% 76% $63K 27% 45 50%

15 Mail Campaign Mailed total of 150k prospects across all four clusters Tiers selected for mailing are shaded Cluster I Cluster II Cluster III Cluster IV Projected % Cum. Projected % Cum. Projected % Cum. Projected % Cum. Model Tier Response Lift Response Lift Response Lift Response Lift A 2.70% 251% 1.52% 228% 3.80% 263% 2.67% 130% B 1.54% 175% 0.88% 159% 2.11% 181% 2.23% 100% C 1.01% 129% 0.62% 117% 1.35% 130% 1.77% 85% D 0.77% 97% 0.45% 87% 1.08% 98% 1.36% 68% E 0.53% 71% 0.32% 64% 0.70% 71% 1.04% 52% F 0.43% 52% 0.25% 46% 0.52% 51% 0.83% 39% G 0.28% 35% 0.17% 30% 0.37% 34% 0.82% 29% H 0.18% 21% 0.18% 18% 0.30% 21% 0.53% 18% I 0.08% 10% 0.12% 9% 0.23% 10% 0.49% 10% J 0.04% 0% 0.09% 0% 0.09% 0% 0.14% 0% Summary 0.77% 0% 0.46% 0% 1.06% 0% 1.17% 0% Reading The Results Actual Response Predicted Response Cluster 1 Cluster 2 Cluster 3 Cluster 4 Control 1.13% 0.95% 1.01% 1.00% 0.79% 1.17% 1.00% 1.06% 1.17% 0.85% 30% increase in available universe (650k-500k)/500k ν 18% lift in response

16 Final Thoughts Compare predicted results to actual results Always use a control group Track results at the finest possible manageable level Refine results Gaining Superior Results Through Creative Modeling & Segmentation Presented by: Marc Fanelli Experian Corporation DMA - October 28-31, 2001