Predictive Analytics 101
Current Trends in
Predictive Modeling and Analysis
Frank A. Alerte, Esq.*
Agenda
Overview of Predictive Analytics
Insurance Applications for Predictive Analytics
Compliance Considerations
OVERVIEW OF
Forecasting?
What is Predictive Analytics?
Business analysis that produces a predictive score
for each customer, prospect, claim, etc…
Strategic planning?
…to guide how to treat each of them individually
Actuarial science?
65
11
53
Subject Matter
What is the Basis for PA?
DATA+
Machine
Learning
Predictive
Analytics
350 BC: Aristotle
Classification & logic
300 BC: Euclid’s
Elements
Arithmetic & geometry
How Did PA Come About?
1930s: Ronald Fisher
Putting math into statistics
1936: Alan Turing
Universal computation
1950-60s: Artificial Intelligence
Making computers intelligent
1940s: Digital Computers
Automating computation
1967: DIALOG -
Retrieving
information from anywhere
1970s: Relational Databases
Making relations computable
1980s: Neural Networks
Emulating the brain
1989: The Web
Collecting the world’s information
1994: Yahoo!
Hierarchical directory of the web
2004: Facebook
Capturing the social network
1998: Google
An engine to search the web
3,500 BC:
Written Language
Math & Statistics
Computer Science
̶
Cheaper storage
̶
Faster retrieval
̶
More computing power
Data Types and Sources
Data Types
Demographic
Behavioral
Data Sources
Internal databases
Third party vendors
General Framework
DATA
Business
Understanding
Data
Understanding
& Model
Constraints
Data
Preparation
Testing /
Modeling
Evaluation
Deployment
INSURANCE
APPLICATIONS FOR
Insurance & Predictive Analytics
Pricing / Underwriting
Claims Function
Pricing / Underwriting
NEED:
How can we better
identify and price risk?
DATA:
Claim and underwriting
information.
What Model Might Be
Used to do This?
Commonly Used Statistical Model is
called a Generalized Linear Model
(GLM)
What is a
GLM Doing?
Historical pricing methods can run into
problems in insurance!
GLMs are versatile in function and
scope.
Testing of Model Results
Hold-out Data Set Testing
Subsequent Period Testing
Impact / Dislocation Testing
50 100 150 200 250 300 350 400
Policyholder Rate Impacts from Model Implementation
What Else Can We Do With
Predictive Analytics?
Claim Fraud Detection
NEED:
Better process for
determining which claims
are potentially fraudulent.
DATA:
Claims data
Underwriting data
Reserve Setting
NEED:
Better process for establishing case
reserves and estimates of the ultimate
settlement value of claim.
DATA:
Claims data,
Strategic Planning
DATA:
Claims
Underwriting
Cat Modeling
Pricing
Expense
NEED:
Use predictive analytics, Cat models
and other information to set strategy
Marketing and Insurance
NEED:
Identify better targets for sales and
increase effectiveness of marketing.
DATA:
Quote/bind data
Underwriting data
Target
Marketing
Sell to Buyers
Fill the Holes
Increasing Retention
Better Understanding Business
Effects of Pricing Changes
Develop Retention Marketing
Adverse Selection
COMPLIANCE
Compliance Considerations
Compliance Considerations
Marketing
Underwriting /
Pricing
Claims
Unfair trade practices
Discrimination
Privacy
Unfair Trade Practices
Unfair trade practices in insurance have existed as long as
the industry of insurance itself.
Unfair trade practices laws and regulations are consumer
protection mechanisms that traditionally focus on two
aspects:
Unfair claims tactics
Unfair marketing/advertising tactics
The use of social media is subject to state insurance laws that
govern unfair trade practices.
Discrimination Issues
What information is being used?
Is it protected class information?
Is the net effect discriminatory even if not
Privacy Issues & Policies
Insurers are subject to Gramm-Leach-Bliley, the Fair Credit
Reporting Act, and the Fair and Accurate Credit Transactions Act.
Privacy policies set forth the terms by which the company will
handle the personal information collected from consumers.
Key compliance questions include:
̶
What personal information is collected?
̶
How is it being used?
̶
Are appropriate safeguards applied to protect it?
Record Retention Requirements for
Advertising Compliance
Development of protocol and retention of
specific factors used to establish marketing.
Sliding Issues
Trying to improperly move a customer to
purchase another product.
Parameters on what is appropriate and what may
cross the line.
Computer models and results may be important
Pricing / Underwriting:
Key Compliance Questions
The 2013 Florida Statutes
627.062 Rate standards.
—(1) The rates
for all classes of insurance to which the
provisions of this part are applicable may
not be excessive, inadequate, or
unfairly
discriminatory
.
Pricing / Underwriting:
Key Compliance Questions
How was the book of business actively composed?
What information does the “black box” model obtain?
How does the model use this information?
Is the info consistent with underwriting guidelines?
Does the computer acquire new info over time?
Can you verify that the computer obtained and used
Pricing / Underwriting:
Transparency for Regulators
Can the regulator see how the model works?
Can the regulator understand the information
obtained and how the information is used?
Can the regulator be assured the information
Pricing / Underwriting:
Use of Social Media
There is little to no specific regulation regarding
use of information obtained through social
media.
Does social media accurately predict behavior?
Claims Compliance Issues:
Legitimacy of Factors Used
Market Conduct concerns
Litigation exposure – bad faith and
class action
Claims Compliance Issues:
Public Info / Social Networks for Fraud
Insurance companies collect information
to determine if claims are legitimate
Can use Facebook instead of private investigator to see
physical health (i.e. “day in the life”)
Example:
̶
A woman was on medical leave for depression.
̶
Her disability benefits stopped after an insurance
employee found photos on her Facebook page of her at
the beach and hanging out at a local pub.
Claims Compliance Issues:
Other Considerations
Review by human vs. machine
Misinterpretation of photos or status updates
Fake social media accounts
Human error
Regulatory settlement related to injury claims from automobile
accidents
The issue was the carrier’s use of a software program that was
intended to standardize the claims process by providing consistent
valuation of bodily injury claims for settlement offers. Specific
issues included:
̶
Inconsistencies in the carrier’s management and oversight of
the claims software across its different claims handling regions
̶
Claims Compliance Issues:
Validation of Settlement Amount
Under the settlement:
̶
The carrier must ensure that claims are handled consistently across all
of its claim handling regions
̶
The carrier paid $10MM to 45 states to train state examiners in the
use of the claims-adjusting software
̶
Claimants will be better informed as to how the carrier arrives at a
claim offer
Claims Compliance Issues:
Validation of Settlement Amount
What’s on the Horizon?
Telematics for the Masses
Focus on Human Behavior
New Data Sources
Expansion of Social Media Mining
Incremental vs. Disruptive Innovation
Chief Analytics Officers
Data Element
Frequency and Severity of Risks
Regulatory Constraints