Download presentation
Presentation is loading. Please wait.
Published byBartholomew Wood Modified over 6 years ago
1
PROPENSITY SCORES Learn how to efficiently identify customers most likely to respond to marketing campaigns Webinar Dec 1, 2016
2
PRESENTERS David Royal, Keaton Baughan, Client Success Manager
Product Manager
3
Anyone know who this is? Let’s test the chat functionality
We have a sweet prize for the first person to chat the correct answer! Hint 1: He’s the greatest prospector in the North Hint 2: He’s obsessed with silver and gold Hint 3: Friends with Hermey and Rudolph
4
WHAT ARE PROPENSITY SCORES?
AGENDA 1 WHAT ARE PROPENSITY SCORES? 2 HOW TO CREATE PROPENSITY SCORES 3 ANALYTICS IN ACTION
5
WHAT ARE PROPENSITY SCORES?
AGENDA 1 WHAT ARE PROPENSITY SCORES? 2 HOW TO CREATE PROPENSITY SCORES 3 ANALYTICS IN ACTION
6
PROPENSITY SCORES Visible Equity Range: 0 - 100
A number that indicates how likely a customer is to purchase a product Visible Equity Range:
7
PROPENSITY SCORE EXAMPLE
PRODUCT: Auto loan purchase in the next six months Propensity Score: John - 82 Steve - 12
8
WHY PROPENSITY SCORES? Targeting Save Resources Simple
- Identify the best customer/product combinations Filter out customers less likely to purchase - Save Resources - Time - Money - Labor Simple Easy to understand Interpretability
9
WHAT ARE PROPENSITY SCORES?
AGENDA 1 WHAT ARE PROPENSITY SCORES? 2 HOW TO CREATE PROPENSITY SCORES 3 ANALYTICS IN ACTION
10
PROPENSITY SCORE DATA
11
RESPONSE VARIABLE The response variable is the product of interest
Use predictors from previous period to predict who purchased product of interest within the period Ex. PRODUCT: Auto loan purchases in the next six months
12
PROPENSITY SCORE EXAMPLE
PRODUCT: Auto loan purchase in the next six months Propensity Score: John - 82 Steve - 12
13
MODEL SELECTION This is a classic case of classification modeling
Generalized Linear Model - Logit - Probit - Cloglog - Loglog Create training and test data sets with random assignment Visible Equity used 70% of data for training and 30% for testing
14
HOW TO CHOOSE PREDICTORS
Dealing with many possible predictors One way is to choose which variables make sense to you LASSO – known as a shrinkage method, in that it takes many variables and narrows the list down to the most predictive ones
15
PRODUCTS AND PREDICTORS
PRODUCTS (6 Months) Auto Loans Bank Cards First Mortgages HELOCS EXAMPLES OF PREDICTORS FICO Score Number of credit inquiries made in the past six months Estimated Income Number of loans with >0% utilization change in past six months % of loans never delinquent in past six months
16
MODEL SELECTION Using each model, create propensity scores for each person in test set Separate propensity scores into quintiles (sort from lowest to highest scores and group into five equal size sets) Within each quintile determine the proportion who actually purchased the product Repeat this process many times VE performed this using an expanded set of possible predictors and using a standard data pull
17
MODEL PERFORMANCE PRODUCT DATA 0% - 20% 20% - 40% 40% - 60% 60% - 80%
80% - 100% QUINTILE AUTO STANDARD 4% 7% 10% 14% 23% EXPANDED 8% 24% BANK CARD 6% 9% 11% 15% 22% 13% 27% FIRST MORTGAGE 1% 3% 5% HELOC 2% 0%
18
OTHER POSSIBLE MODELS SUPPORT VECTOR MACHINES RANDOM FORESTS
LINEAR DISCRIMINANT ANALYSIS
19
WHAT ARE PROPENSITY SCORES?
AGENDA 1 WHAT ARE PROPENSITY SCORES? 2 HOW TO CREATE PROPENSITY SCORES 3 ANALYTICS IN ACTION
20
PROPENSITY SCORES IN CUSTOMER ANALYTICS
21
FUTURE IMPROVEMENTS Obtain more representative sample of population
Use Equifax and TransUnion data Explore more predictors Include more products Incorporate time and changes in behaviors
22
SUMMARY Q&A What are propensity scores Use for marketing tool
Credit Bureau data Model development Coming soon
23
THANK YOU FOR MORE INFORMATION PLEASE CONTACT:
Similar presentations
© 2024 SlidePlayer.com Inc.
All rights reserved.