CAS Predictive Modeling Seminar

1
CAS Predictive Modeling Seminar

• Predictive Modeling for a Commercial
• Insurance Company with Little or No Data
• Scott Bronstein October 5, 2004

2
Discussion points

• Data available for predictive modeling
• Models in use
• Sample results
• Summary

3
Modeling guidelines with minimal data

• Know your target population
• Tap into business data - even minimal amount of
  business performance data can be predictive
• Use consumer data as appropriate
• Segment the population
• Validate at logical intervals - can recalibrate
  if necessary

4
Composition of U.S. businesses
Publicly heldlt1
Partnerships 15
Privately held 7
Sole proprietorships 77
190 million consumers
5
Experian Business Information Solutions

• No other company houses these assets under a
  single roof

Integrated Information Solutions
Business Public Record Database
National Business Credit Database
National Business Database
National Consumer Credit Database
6
Credit Database Sources
Marketing database
Public record
Experian Business Credit Database
Firmographics
Trade payment
Banking, insurance, leasing
Collections
Standard Poors
7
Marketing Database
Business White Pages
Credit Database
National Business Database
Data Vendors
DBA/FBN (new business)
Televerified Data
National Yellow Pages
8
Scores Offer Solutions Across Customer Lifecycle
9
Intelliscore overview

• Used primarily in small business lending,
  commercial card, leasing, telecommunications, and
  business services
• Commercial Intelliscore
• Utilizes commercial credit, business
  demographics, public record and legal information
• Small Business Intelliscore
• Utilizes commercial and consumer credit, business
  demographics, public record and legal
  information

10
Intelliscore models are segmented scoring systems

• Businesses within a modeling sample are clustered
  or segmented by common characteristics such as
• Size of business
• Credit history
• Data type availability
• When predictors behave differently between
  clusters, the file should be segmented so the
  clusters can be modeled independently to capture
  subtle nuances in payment behavior

11
Model segment integration

• Scoring system integrates model segments into one
  uniform score
• Each Intelliscore model solution consists of
  several modeled segments each with its own set of
  variables and raw score

Model Segment 3 score
Model Segment 2 score
Model Segment 1 score
Transformation
Uniform 0 - 100 score
12
Commercial Intelliscore Report

• Key features

Commercial credit and business demographic
information for up to fifteen elements
Credit score and percentile
Action
Score factors
13
Commercial Intelliscore18 to 60 lift in
predictiveness
14
Small Business Intelliscore12 to 35 lift in
predictiveness
15
Summary

• Wide array of business data is available
• Familiarity with how to use the data is critical
• Small business is broadly defined and data
  availability will vary
• Segment the population
• Commercial risk scores are predictive across
  several industries - and probably others