profit based acquisition strategy for credit cards
DESCRIPTION
Based on a paper by RT StewartTRANSCRIPT
Based on paper on ‘Profit- based acquisition strategy
for credit cards’ by ‘RT Stewart’
Presented by Piyush
1. Revenue Spends/Interchange Finance charges Others
2. Cost Fixed costs Acquisition costs Other operating costs
3. Loss
Credit Card Profit & Loss
◦ Potential Revenue NOT considered◦ Approve or Decline decision based solely on
risk Minimize bad rates
Current practice for customer acquisition
Decline Approve
LowCut off
Risk
High
High
Credit Score
Bad Rate / Charge-Off rate - ◦ Ratio of number of customers defaulting on credit cards
to the total number of customers.
Credit /FICO score – ◦ A score representing the creditworthiness of a person.
Few Credit Card Jargons
Primary - Develop and test a methodology to model revenue.
Use revenue models along with risk models for acquisition decisions.
Objective
1. Revenue is highly correlated with risk
2. Structural Change / Population drift
Challenges in modelling revenue / profit
Modeling problem - ◦ Predict cumulative spends during first 2 years of account’s
life
Independent variables –◦ Credit bureau data◦ Account application data
Training data –◦ A sample data set of 300,000 credit card accounts
Segmentation – ◦ Segments based on credit bureau scores. ◦ Multiple spend models.
Methodology
Log(Spend) used as response variable
Modeling equation –log(Spend) = β₀ + β1 X1 + β2 X2 + β 3 X 3 +......
where β₀ , β1, are regression parameters
Model details - I
Independent variables (X1 , X2 , ...)◦ Binning approach used
Increases model stability Easier implementation Capture non-linear relationships
◦ Correlated with spend but uncorrelated with Risk
◦ Examples – Applicant’s monthly income
[$0-$2500] , [$2500-$5000] ,.. Age of oldest revolving trade in months
[0-71] , [71-999]
Model details - II
1. Revenue is highly correlated with risk◦ Creating risk segments based on bureau score
2. Structural Change / Population drift
Challenges addressed!
1. Revenue is highly correlated with risk◦ Creating risk segments based on bureau score
2. Structural Change / Population drift◦ Leveraging binning approach for independent
variables
Challenges addressed!
Results
Models rank order spend.
Example - Model for segment FICO (720-760)◦ Spend shows a positive
slope.◦ Charge-off line is
approximately horizontal.
Higher mean spend with same bad rate.
Approval rate(%)
Bad Rate (%)
Mean Spend ($)
FICO Only 90% 1.60% 15,032FICO and Spend
score 83% 1.60% 16,617