Transcript What are predictive models

OESAI COMPREHENSIVE LIFE
INSURANCE TECHNICAL TRAINING
Predictive Underwriting
How insurers can use statistics models to make
sales process easier
OESAI COMPREHENSIVE LIFE INSURANCE
TECHNICAL TRAINING
Ezekiel Macharia
Group Actuary - Jubilee Holdings Limited
Day 1, Wednesday 11th November, 2015
AGENDA
•
•
•
•
•
•
•
Predictive Underwriting
Making a Life Insurance Sale
What are predictive models
Usage of predictive models
Sample scoring
Developing an predictive model
Conclusion
Type “statistics” on eBay and an advertisement
comes related to your search – how did they know
what you like or if you will click on the
Predictive Underwriting
• Using predictive models to give insights
into the day-to-day underwriting
processes of a life insurer
• For example, determine the profile of the
client beforehand and determine which
people are fast tracked and those that
require a medical report
Making a Life Insurance Sale
Before underwriting (High Sum Assured)
• Ten people want to buy a life insurance policy with a sum
assured of $100,000
• Each requires a medical report as per the underwriting
guidelines for the sum assured requested
• Also required to fill in 10 page questionnaire
Making a Life Insurance Sale
• Five people give up!!
• Three people are ok
• One requires premium to be adjusted with exclusions
• One is rejected
After underwriting (High Sum Assured)
Load
OK
OK
OK
Decline
Making a Life Insurance Sale
• Sale process was unsuccessful due to the following:•Process is cumbersome for client but critical for insurer
–Requires a third-party medical exam
•Broadcast approach – check everyone (we don’t now who is a
high risk and who is a low risk)
•Blame Others: Our agents made a hard sale? Was this the
right customer? The product is expensive, if the price was
lower – could they have bought the product?
What are predictive models
• Example - Models that use statistics to score the risk profiles of
potential clients and provide insights as to which clients require
further investigation, e.g medical checkup
• We can now require less people to go through the rigorous
process of underwriting & verification – improving the sale
process
• In the example below – 5 people do not need to take
medical examinations after risk scoring
What are predictive models
• The predictive models can be automated in the IT system
Possible usage of predictive models
for life insurance companies
Sales
Agent Selection
Shortlisting productive agents
Customer Segmentation
Which customers will buy life insurance
Cross-Selling
Which term assurance clients can buy endowment?
Price Optimization
Different prices for different channels
Risk Selection
Risk scoring, ordering underwriting requirements
Others
Fraud
Over-insurance and anti-selection
Pricing
Reflect risk more effectively
Reserving
Setting the right technical provisions
Sample Scoring – Underwriting
requirements
Pass
Refer to
underwriter
Medical
Test
Reject
Key requirements for predictive model
• Data, data, data….
•Historical data (preferable in suitable format)
•Data Warehouse
• Rating Factors: Age, Gender, Smoking status, Sum Assured,
Admitted family history, BMI, Negative admitted personal
medical history , current findings on blood (haemoglobin), lipids
(e.g fats), Liver test (GGTP), etc
• Configuration with experience (need regular updates)
Developing a Predictive Model
1. Data Mining - Establish Patterns
•Collect data, clean data and assign data distribution
2. Logic & Algorithm
•Develop decision trees & identify factors and predictors
3. Build Model (can be repetitive)
•Build, Test & Calibrate
4. Validate
5. Implement & Document
6. Monitor and Recalibrate
Popular Predictive Models
1.
2.
3.
4.
5.
6.
7.
8.
Decision Trees
Regression Trees
Cox Model
Generalized Linear Model
Logistic Regression
Regression Spline
Neural Networks
k-Nearest Neighbour
Disadvantage
1. The model may be wrong
–
–
–
If not checked/updated/calibrated regularly with recent data
Overfitting/wrong predictors
May not make sense (common sense)
2. Black box – nobody knows what is inside it
3. May depend on modeller (biased by perceptions)
4. Requires IT infrastructure, data (lots of it) and human
capital
1. Prediction
–
Advantages
Customers are happy if the sale process is shortened or
the sale is warmer (selling to a client already looking for a
particular product)
2. Some prediction models require minimal statistical
knowledge – neural nets
3. Various statistical methods available for prediction
models
4. Usage of already collected data to improve business
process – insurers with rich history, strong data
integrity can leverage – perfect for online business
Conclusion
• Predictive underwriting uses data analytics to give
insights into the customer
• These insights can be used to provide competitive
advantage for an insurer – this can be in sales, claims,
pricing or reserving
• Prediction models can be build but require data
• Expected to grow with more adoption of big data
and data mining techniques
• Perfect for online business
QUESTIONS
[email protected]
+254 722 540 045