Three Steps to Finding Your Model Customer: Defining, Segmenting and Scoring

Presented by: Dr. Kate Webster | Chief Statistician AccuData Integrated Marketing
Sponsored by:

Thank you for joining us--we will be starting at 2:00 PM Eastern, 11:00 AM Pacific. If you are unable to hear music at this time, please make sure that your computer speakers are turned on and that your system has not been muted. You can access additional tips for setting up your computer using the Online Help button below.

Housekeeping
Getting help Press the button to access the help files and to view system requirements for this session. Viewing the slides Todays slide presentation should advance automatically in synch with the audio presentation. If they do not change, you may try pressing F5 (Windows) or -R (Mac) to refresh your connection to the server.. Press to get a larger view of the slides. Adjusting your volume Volume can be adjusted using the volume control on your computer You can also adjust volume using the slider on the media player at the bottom-right of your screen. Asking questions Type your question into the Questions box at left. We will answer as many questions as possible during our Q&A session at the conclusion of todays event. Recording Todays session is being recorded and will be available within 24 hours on the DIRECTs web site. You will receive an e-mail notifying you when it is available.

About DIRECT

Learn more and subscribe at www.directmag.com

Introductions
Dr. Kate Webster Chief Statistician AccuData Integrated Marketing

Webinar Sponsor

AccuData Integrated Marketing Headquarters

AccuData Integrated Marketing drives client-specific results by providing data, predictive analytics, and database marketing solutions to deliver the most complete, independent, and insightful view of customers and prospects.

Marketing and Statistical Modeling

Grow Sales and Revenue

Marketing helps us to reach stated business objectives, most notably:

Increase Profitability
Gain Market Penetration

New Customer Acquisition

Modeling and analytics, enhances the efficiency & effectiveness of marketing endeavors, particularly:

Current Customer Retention Up-Sell

Demo and Behavioral Predictors

Getting the most out of data analytical efforts requires a sound Analytical method using a Learn-do-Loop Process:

Offer Segmentation Response Analysis Ongoing Customer Optimization

3 Steps to Finding Your Model Customer

Marketing Analytic Process

New Customer Acquisition Behavioral Predictors Define Segment
Source Segmentation RFMT Segmentation RFMT Segmentation Market Penetration

Current Customer Retention

Profiling

Single-Buyer Conversion
Product Profile

K-2 Cluster Analysis

Loyalty Segmentation

Buyer Value Segmentation

Current Thinking

Score and Target

Best Customer/Prospect Look-alike models Responder models

At risk models: Retention At risk models: Optimization

Loyalty analysis/scoring Repurchase/growth model

Forward Thinking

In-market experimentation

In-market experimentation Promotional baseline Response survey

In-market experimentation Promotional baseline Response survey

Response Analysis

Promotional baseline Response survey

Optimizing Current & Future Thinking

Promotion, price elasticity

Promotion, price elasticity

Custom Scores

Product Sequencing Promotion, price elasticity

Custom Algorithms
Customer categorization

Denotes methodologically-shared applications

Step 1: Define Your Marketing Objectives

What do I need to know?
Who are my customers? What is my market penetration? Who are my best Customers? How do I find prospects that look-like" my best customers? Who is most likely to respond to my campaign? Who is most likely to convert to a customer? Which customers have the highest probability to stop buying (attrite)? Which customers are most likely to buy a certain product and how do I maximize my current customer relationships?
Predictive Modeling Solution: Score your customer database or prospect list using a predictive algorithm to predict those customers or prospects with the highest statistical likelihood of migrating toward or newly becoming one of your BEST customers

Which solutions provide the answer?

Descriptive Profiling Solution: Define overall customers and compare with Universe of consumers or businesses.

Descriptive Modeling + Segmentation Solution: Segment customers and rank them to identify your BEST customers.

How can I make this an ongoing process?

Response Analysis Solution: Compare mail file(s) to responders and current Customers. Analyze ROI as it relates to offer and call to action (learn-do-loop)

Step 1: Define Your Current Customers

Behavioral Predictors

New Customer Acquisition Define

Market Penetration

Current Customer Retention

Profiling

Single-Buyer Conversion
Product Profile

Current Thinking

Defining Your Current Customers Story...

Descriptive Profiling Defining what your current customers look like across multiple indicators Market Penetration Index Profile PLUS a penetration index that compares your customers representation divided by a random sample Universe representation (*100) Index of 100 indicate no difference (36%/36%) * 100 = 100 100 Small Effects we begin to tell the story of the customer profile 150 Large Effects capturing more of the story AccuData Intelligent Profile Web based, real-time match and append service resulting in standardized market penetration index scores that rank prospects from highest to lowest in their resemblance to customers.

Step 2: Segment
Behavioral Predictors

New Customer Acquisition Segment

K-2 Cluster Analysis Source Segmentation

Current Customer Retention

Loyalty Segmentation RFMT Segmentation

Single-Buyer Conversion
Buyer Value Segmentation RFMT Segmentation

Current Thinking

Defining Your Best Customers Story...

Segmentation Models and Ranking Identify significant differences or similarities across subgroups of customer types Cluster Analysis: Custom clusters group customers based on similarities in product purchase and demographics. RFMT: (Recency, Frequency, Monetary Value and Tenure) Transactional Values
Bag or Boost the Bronze

Slice the Silver

Go for the Gold

Protect the Platinum

Customer Loyalty Analyses (Tenure): essential to understanding lifetime value and retention across customer base

Step 3: Score and Target

Behavioral Predictors

New Customer Acquisition Score and Target

Best Customer/Prospect Look-alike models Responder models

Current Customer Retention

At risk models: Retention At risk models: Optimization

Single-Buyer Conversion
Loyalty analysis/scoring Repurchase/growth model

Forward Thinking

Finding the Best Buyers of Your Story...

Focus on mathematical overlap (look-alikes) Logistic Regression or Discriminate Function Models Mathematical algorithm/equation that predicts the likelihood (probability) of some event occurring The equation is applied to every prospect or customer record Cross Validation of the model is essential to validity and reliability Scores are ranked into deciles and list selection occurs through a top down fashion or sampling for testing

Track and Evaluate Model Performance

Behavioral Predictors

New Customer Acquisition

In-market experimentation

Current Customer Retention

In-market experimentation
Promotional baseline Response survey

Single-Buyer Conversion
In-market experimentation
Promotional baseline Response survey

Response Analysis

Promotional baseline Response survey

Optimizing Current & Future Thinking

Tracking Your Recent Buyers Story...

Quantify gains attributable to specific marketing treatments Quantifying lift Rolling 24-month marketing ROI tracking Response by Campaign Code Response by Channel Response by Vender Response by Offer Capture Trends Refine Targeting/Algorithm Re-Deploy Model

The Learn-Do-Loop

Expected Key Learning's and Deliverables

Model Debrief
Collaborative exchange of process and findings Match report of customer data Definition of customer segmentation Summary statistics on transactional and product data Comparative graphical analysis on all relevant characteristics that best define the target group Recommendations for list selection, offer and creative and proceed to scoring

Validation Scoring Reports

An interpretive narrative that details the scoring and validation distributions Gains tables or charts that depict the scoring of the universe of prospects as compared to the defined customer group and validation file Counts of prospects or best customers by score and decile A scored universe file with mail information and logistic regression score and decile for all prospects appended to the output file for list selection

Expected Timeline

Customer File Receipt and Preparation

Modeling and Output Creation

Scoring Universe and Data Pulls

1-10 Business Days

(Based Upon) Number of Files Completeness of Files Level of File Hygiene Complexity of File

3-7 Business Days

(Based Upon Output)

5-7 Business Days

Offer Segmentation Response Analysis Ongoing Optimization

5-7 Business Days 3-5 Business Days Scoring 2 Business Days Data Pull

~ 20-25 Business Days

How to get started?

A collaborative and experienced data and analytic partner Clearly established goals and objectives Current customer file (or a random sample across some specified period of time) Mail file Responders to those mail files Flags for campaigns (creative and offer) Any transactional data (if available and applicable) or data elements to help derive tenure, recency, frequency, and monetary attributes of the customer if not readily present in data Data must be drawn as random samples or full files Data should be able to be sent to client in multiple formats through secure FTP sites

Model Application

Business-to-Consumer XYZ Bank Customer Acquisition

B2C Community Bank Attrition:

Descriptive and Predictive Model
XYZ Bank: Attrition and Cross-Sell Model
Problem A well established community bank (New England) was experiencing an increase in attrition. XYZ Bank was interested in understanding if their recent attrition was related to an even more recent expansion of their DTA or was it a combination of rate sensitivity, customer channel preference, geography including distance to competitors bank and nearest XYZ BANK from customers home. Solution 300 demographic and life-stage indicators were used to enhance the clients transaction/product file. Built a predictive cross-sell model using Customer transactional data and demographic/ behavioral data elements to identify those with the highest propensity to attrite and their relationship to product(s).

Method:
Define and Segment XYZ Bank Customers
Inputs:

11 Customer/Product files over last 15 Years 4.3 Million Records/Transactions 5 Global Products (70 specific)
Define: Summarizing the Current State Data Hygiene/Merge/Purge (Match and Append 300 Demographics) Identify Attrite Group ($0 across specified time) Derive Recency/Frequency/Monetary Values and Tenure Calculate Distance (lat/long) Define Original DTA and Plot Expansion Identify Share of Wallet using consumer Universe

Segment Attrite vs. Active Customer: Diving Deep

To What extent is attrition occurring? Where do they live? What is their tenure? What are there interest rates? Who is attriting? What do they look like? What products do they start with/end with? How far do they live to a competitor bank? What channel do they prefer?

XYZ Bank : Step 2 Segment

Figure 1: Comparison of Mean Scores by Group Groups on Primary Demographics

Defining Attrite Characteristics

60 Attrition Group *50.8 50 40
Percent

Young (mean/age 24) Higher education Upwardly mobile professionals

Active Group

Predominately male Living outside designated DTA areas Rely heavily on internet banking Little branch use Start product ATM Highest interest rates with high variability Shorter lengths of residence Average 2 or less products Most Attrite 12-18 months

30 20 10 2.3 0 Services by Household Interest Rate Age Distance to Bank Distance to Competition Number of Children Length of Residence *4.5 *4.1 18.7 *21.9 *21.1 10.2 7.6 1.1 7.3 0.4 0.4 *13.0

This group of young urban professionals may indicate a trend in community banking where loyalty and affinity is lowered as a result of internet banking and a clear lack of connection to branch locations. This group may be considered Jumpers based on offers from competitors. A targeted messaging campaign that connects with this young urban segment may increase retention rates.

Results XYZ Bank: Step 3 Scoring and Tracking

Recommendations: Predictive Model and Scoring Develop Loyalty score to identify current customers with the highest likelihood of attriting The higher the score, the greater the risk of attriting Develop campaigns and messaging to interrupt loss or customer and revenue Predictive algorithm modified and applied to prospects universe within the XYZ footprint

Predictive Modeling Provided the Following Solutions Predict the relationship between number of products and increased loyalty (sequencing) Continue DTA expansion and testing Predict which customers were most at risk for attrition over the next 6 to 12 months Predict which prospects within foot-print were most likely to become a customer Predict which product offer would be best as a start product offer for differing segments

Bottom Line
Significant Increased in ROMI and greater Growth Remains a Partner practicing in the Learn-Do-Loop for cross sell and acquisition

B2B Acquisition:
Automobile Case Study
Drive home prospecting B2B sales for Big Three Automaker
Problem Large ad agency needed help identifying prospects with the highest likelihood of purchasing a newly released commercial vehicle for one of the Big Three automakers.. Solution Built a predictive cross-sell/acquisition model using Customer and Firmagraphics data to identify the best B2B prospects for new purchase and up-sell opportunities.

Results B2B Model:

Define, Segment, Score
Define Profile existing B2B customers as compared to US businesses

Segment by Product Calculate group different characteristics on geography, business size, industry, by product group Identify characteristics that most highly correlate with one product but not another

Dive Deeper Custom Cluster Segmentation Based on Industry Cluster development was mutually exclusive classifications where membership in one group excludes membership in another. Predictive Model and Scoring Logistic Regression algorithm scoring Universe of business with the highest propensity to purchase a certain type of vehicle based on cluster formation of best customer

Tracking Model Performance and ROMI

Results Modeling recommendations and report was presented to the President of Internal Marketing Department in Detroit Division Using predictive targeting, AIM model outperformed long-term internal marketing strategies in several weeks Within 30 days of model deployment the MM model resulted in the sale of 23 automobiles while the control group had sold one ROMI Average Vehicles = $25,000*23 = $575,000 Initial modeling investment $25,000 Cost of the vehicles was not provided so a true ROI was not calculated. The agency is now a long-time partner of AIM

Model Application

Additional Case Studies

B2C Acquisition:
Financial Model Case Study
Increase lead generation sales for one of the Largest Loan Aggregators in U.S.
Problem Online Loan Aggregator wanted to increase response rates of click through customers (responders) and decrease lead generation costs per mortgage application Solution (Define, Segment and Score) Built a predictive acquisition model using client data and Multi-Source consumer demographic data that incorporated capacity to purchase as well as high eConsumer Score indicators to find the best prospects with the highest likelihood of completing and application on line Result (Tracking and Evaluation) Successfully accomplished a 50% reduction in lead generation costs while boosting response rates well above 60%

B2C Acquisition and Cross-Sell:

Retail Model Case Study
Increase ROMI through Response Analysis with large Retail Chain
Problem Retailer wanted to increase response rates and increase ROI of existing customers while decreasing reliance on managed list rental Solution (Define, Segment and Score) Built a predictive, current customer retention and value maximization model using client data and multi-source consumer demographics, segmented by purchasing behaviors scored a prospect and customer file using logistic regression to target those with the highest propensity to drive lapsed spenders back to the store and increase traffic of best customer look-likes based on responder profile. Identified those offers with the highest ROI

B2C Response: Analysis Case Study

Result (Tracking and Evaluation) Successfully accomplished a 50% reduction in data costs and a 16% increase in ROI

Alumni Donor Model Case Study

Increase cross-sell lead generation for potential Alumni Donors to Top Universities
Problem Prestigious universities desire improvement in existing donor solicitations while decreasing lead generation costs of new donors Solution (Define, Segment and Score) Developed a predictive giving capacity (wealth) and propensity (other giving behaviors) model identifying best givers based on wealth indicators and combination of other consumer demographics Result (Tracking and Evaluation) Successfully accomplished a 60% increase in lead generation to existing alumni for cross-sell opportunities

Terminology

Terminology
Predictive Modeling Estimation of likely outcomes based on historical data. Results in the development of an algorithm that is applied to the customer or prospect file. Correlation Analysis (It all starts here) A statistical relation between two or more variables such that systematic changes in the value of one variable are accompanied by systematic changes in the other (knowing something about tells us something about the other) Discriminant Function Analysis Predicting membership/classification into a particular categorical outcome (e.g. responder vs. non responder, buyer vs. non-buyer) by determining which continuous variables are the best discriminators between two or more naturally occurring groups (must have continuous predictors (IVs) and categorical DV) Logistic Regression Analysis Predicting the probability that an event will occur based on the predictive power of the indicators to discriminate between the groups. Must have categorical outcome (DV) and predictors (IV) can be both continuous or categorical. Due to the very nature of complied data, Logistic Regression is a popular method in marketing research. Linear or Multiple Regression Models Finds the best fit for a set of independent variables (predictors) based on a single continuous dependent variable you would like to predict Multivariate Analysis A generic term for any statistical technique used to analyze data for more than one dependent variable in the same analyses

Terminology
Cluster Analysis Data reduction that sorts people, things, events, (rows in a database) into groups, or clusters, so that the degree of association is strong between members of the same cluster and weak between members of different clusters Factor Analysis Any of several methods for reducing (columns) continuous data into a smaller number of dimensions or factors (e.g. buying patterns) Structural Equation Modeling (SEM) A complex path analysis that combines the best of factor analysis with multiple regression

Custom Solutions Product Sequencing, Time Series, Automated Analytics and other custom solutions...
Independent Variable Also know as Predictors or Grouping Variables Dependent Variable Outcomes, what we want to replicate (can be continuous or binary

Cross Validation The practice of testing a model using a portion of the universe to test while holding the other portion constant
Index A group of measures based on penetration that when combined produce a value meant to represent a more general characteristically

Follow Up Material / Support

White Paper: A Marketers Guide to Descriptive Versus Predictive Modeling

Modeling Assessment and AIP Demo

Executive Summary: Challenges and Solutions
Dr. Kate Webster Chief Statistician
3steps.webinar@accudata.com

Question and Answer Session

Thank You for Attending!