Modeling Techniques help to bring out the correlations that are predictive in nature. Here I talk about details of modeling statements that has been used to build life cycle management strategies
2. Page 2
Typical Model Development – Bird’s Eye View
Existing Process
Study &
Documentation
Prospect Base
Segmentation
Channel
Optimization
Credit Approval &
Delinquency distribution
by band
Process Benchmarks
Steady-State
Model
Deployment
As-is execution Map
InputSolutionStepDeliverables
Models & Underwriting
Policies
Model Based
Decisioning
Target Market
Assessment
Market Segments &
Credit Needs
Suggested Modeling
Features
Basic Market Study
Competitive Landscape &
Offerings
Prospect Segment
Profiles
Prospect Segment
Model
Segment Product
Mapping
(if sample from past)
Prospect Scoring on
all models
Product-wise
Scoring Models
Model Performance
Details
Product
Propensity
Estimation
Definition of Constraints
Total Acquisition Cost
Definition
Optimal Inventory
allocation by channel
List Generation by
channel
Underwriting Cirteria &
Suppressions Overlay
Model Deployment
Diagrams
Invalid Score Codes
H/W, S/W requirements
Process Owner
Interview
Process Team
Interview
Existing Documents
Internal Reports
Statement of
Objectives
Demographic Data
on Sample
Credit Bureau
Data on Sample
Current / Past Prospect
Sample
Current credit
Distribution
(if sample from past)
Response History
Demographic & Risk
profile of prospects @
time of solicitation
Past Campaign
Samples
Channel Performance
report by product
Channel Costs
Underwriting Criteria &
Suppressions by channel
Prospect Database
Access
Models & Allocation
details(from previous
solution steps)
Steady-state process
IT Team Meetings
IT System
Architecture Diagram
DeployDefine & Measure Analyze, Design & Verify
3. Page 3
Model Development– Data Analysis
•Data Access
•Data Transfer
•Data Storage
•Data Validation
•Quality Check
•Report Out
Customer
Data
• Missing Value Treatment
• Outlier Treatment
•Data Transformation
• Derived Variables
• Creation of Master Data Set
• Validation and Report Out of
MDS
Data
Manipulation • Generating Trend Reports
• Generating Uni/Multi-variate
and Correlation Reports
•Creating Visualization Charts
• Validation of Trends and
Correlation Reports
• Descriptive Analysis Report Out
Descriptive
Analysis
Next
4. Page 4
Model Development– Intelligence Deployment
• Model Build – Creation of Candidate Models
• Model Validation – Out-of-Sample, Out-of-
Time, Bootstrapping
• Model Selection – What-if Scenarios, Lift
Charts, Customer Dimensions & Model
Complexity
• Statistical Tests – Multi-collinearity,
specification & identification condition
Statistical
Analysis
• Deciling and Segmenting
• “Actionable” Insights – Pattern
Recognition
• Population Summary Reports
• Impact Assessment Reports
• Margin of Error Estimation
Intelligence
Creation • Portal Base Deployment of
Visualization charts .
• Deployment of Scores for rank
ordering
• Creation of Scorecards to
differentiate customer behavior
Intelligence
Deployment
Overview
Next
5. CONFIDENTIAL & LEGALLY PRIVILEGED
Case Studies
Acquisition Strategy:
a. Lead Qualification using Decision Tree
b. Channel Effectiveness
c. Response Model for Manufacturer-Driven Auto Loan Program
Customer Management Strategy:
a. Automated Credit Line Increase Program (Details Provided)
b. Improving Product Holding Ratio
c. Attrition Scorecard (Details Provided)
d. Risk Based Pricing (Details Provided)
Risk Assessment and Mitigation:
a. Developing Identity Fraud Procedures as Risk Mitigation Lever
b. Risk Categorization based on Behavior Scores (Details Provided)
c. Collections Call Center Capacity Planning
6. Page 6
Automated Credit Line Increase Program
Business Objective
Who are the customers eligible for an automated credit line increase
program
Of those customers, who are eligible, what should be the optimal size of
the increase
What are the impacts on delinquencies, loss rates and Net Income due
to this program
Business Impact or Benefit
Have identified about 20% of the customers who are eligible for one-
time CL increase program
Provided a list of customers based on the decision tree who become
eligible for CL increase each month
Annualized Net Income of $2MM was estimated
Analytics Solution Methodology
The customers eligible for credit line increase program was determined
based on the past profitability of the customers using decision tree
methodology based on CHAID algorithm.
An Account Level Profitability metric was calculated for each
customer and used as the objective function
Critical drivers including behavior scores and Risk Scores were
analyzed to identify potential downside impacts
Segments accounting for at least 5% of Net Income and having
at least 1% of the number of customers have been chosen for
credit line increase.
Genetic Algorithm based linear optimization was performed where the
constraints given, including,
Utilization rates post CL increase should not exceed 75%.
loss rates not to exceed 10% of the current level.
90+ days past due rate not to exceed 15% of the current level
An Excel-based Monte Carlo simulation exercise was conducted to
analyze the potential downside with $300 and $600 CL increase .
Key Insights and Recommendations
Past Profitability was best seen in customers have Low risk scores FICO
scores of 495 – 545 range, while, the behavior scores of 100-180.
The utilization rates were also high in these buckets
The best segments which were eligible for CL increase where
Low FICO Scores
Medium Behavior Scores
Current Utilization of 45%
Number of times 30+ Past Due <=2
Days Since Last Transaction <= 3 months
7. Page 7
Risk Based Pricing
Business Objective
What kind of revenue opportunities does re-pricing of customer
portfolio offer based on Risk Based Pricing
What are the impacts on delinquencies and long term profitability due
to changes on customer profile arising out of this strategy
What is the best method of quantifying customers’ responsiveness and
the risk behavior and determine the price point at which it is still
profitable to acquire a customer though the risk is high.
Business Impact or Benefit
The client has successfully acquired new customers from segments that
were originally not targeted.
This approach has helped to penetrate deeper into the customer base,
which was, earlier out of bound for the marketing department
This initiative helped the client to provide $2MM net income towards
the annual Net Income target.
This initiative provided the roadmap for more efficient trade-off matrix
to address the burning issue of Low Risk prospects also demonstrate low
responsiveness to marketing campaigns.
Analytics Solution Methodology
The Risk scores and Response scores for a customer has been calculated
Grouped customers into 100 segments based on the risk scores
and response scores
Calculated profitability of the segment taking into
consideration – Acquisition, activation, response rates,
utilization rates, delinquencies, roll rates, charge-off rates,
operational costs, Technology enabling costs and Customer
Service costs.
Look-alikes based on credit limit, average ticket size, vintage, sourcing
were created to understand the future behavior of customers if re-
priced.
In case of new acquisitions, pilot campaigns were conducted by
lowering the minimum Risk Score Cut-off.
Finally, Customer segments that had +ve ROI were targeted and
acquired
Key Insights and Recommendations
Higher Risk customers are comparatively price inelastic. However, the
lower risk customers display much higher elasticity towards Risk Based
Pricing.
In most cases, in high risk customers, the increased margins are negated
by higher operational costs.
Acquiring new customers at higher APR is far more profitable than re-
pricing an existing customer to higher APR because of adverse selection.
Risk Bucket Re-pricing
Very Low Risk Reduce the APR in the range
of 5%-10%
Moderate Risk Unchanged
Moderate-High Risk 1%-2% increase in APR
High Risk 2%-5% increase in APRDetails
8. Page 8
Attrition Scorecard
Business Objective
Identify customers who are likely to attrite
Business Impact or Benefit
Attrition Score provided propensity to attrite, basis which Retention
Campaigns could be evolved
This exercise also provided a detailed analysis to understand the drivers of
attrition.
An Annualized Retention of over INR 50 Crores of Balances-at-Risk by
executing retention campaigns
Analytics Solution Methodology
Solution developed analyzed the 4 stages of customer attrition –
Changes in Customer Transaction Behavior
Reasons for closing the account to identify “preventable” attrition
Link the potential reasons with actionable mitigates
Rank order customers based on their likelihood of attriting in the
next 6 months
Separate solution was developed for 2 types of attrition noticed
Silent Attrition: Customers who reduce keep only min balance and
do not transact on their account
Formal Attrition: Customers who formally close their relationship
with the Bank.
Customers were rank ordered based on their Attrition Score, CNR (customer
Net Revenue) and Product Holding
Key Insights and Recommendations
Primary Reason for attrition is change of employment followed by change
of residence.
Customers who have more than 1 product tend to be less prone to
attrition.
The early warning signs of attrition are
Reduction in Average Quarterly Balance
Reduction in number of customer-initiated transactions
Customers having Investment relationship with the Bank are least prone to
attrition.
Details
9. Page 9
Risk Categorization based on Behavior Scores
Business Objective
Categorize Risk Behavior at the time of acquisition based on expected loss
rates and PDO (points to Double Odds).
Business Impact or Benefit
The Risk Scorecard was used to identify savings account customers eligible
for Cross-Sell for Asset Products.
Analytics Solution Methodology
The Solution methodology involved the following 4 steps
Development of Risk Scorecard using Credit Bureau and Internal
transaction behavior.
Converting the default propensity scores into an scorecard ranging
from 200 to 800 using the concept of scaling.
Calculating the Points to Double Odds ratio for each scorecard by
fixing the points at 20. This is done to ensure customer risk is
ascertained across the score bands.
Run an historical validation to ascertain the ability of the scorecard
to categorize risk across the spectrum.
Key Insights and Recommendations
The critical variables that have come significant are
Number of Trades (Credit Bureau Data)
Number of times Past Due in the previous 12 months (Internal
Data)
Number of trades past due (Credit Bureau Data)
Account Vintage
Channel of acquisition
Editor's Notes
Placeholder - presentation title | Date - 6 April, 2006
Placeholder - presentation title | Date - 6 April, 2006
Placeholder - presentation title | Date - 6 April, 2006
Placeholder - presentation title | Date - 6 April, 2006