AXA Hong Kong and Macau's Green Data Drives Go-Green Home Insurance Campaign
1. 1
Public
AXA Hong Kong and Macau:
Green Data drives our Go-Green Home insurance Campaign
Public
2. 2
Public
Our Century’s
Biggest Challenge
AXA committed to
take Go-Green
Action
Led by data-driven decision,
we launched our Home insurance
campaign with GREEN features
Due to the climate change, global insurers were suffered from
greater claims losses caused by extreme weather events. AXA
committed to drive the transformation and carried out ESG-related
initiatives to protect the environment. Therefore, AXA Hong Kong
and Macau GI have launched a home insurance campaign with
“green” features to raise the customers awareness on climate
change and importance of environment protection.
Public
3. 3
Public
Our innovation: A Thousand Miles Begins With A Single Step - Data
Finding pattern in the
existing Home Claims Data
Develop Machine Learning to derive
the below factors that correlates
with claims risk:
1) Formulated an integrated Green Score
2) Weather/ non weather
Launch Energy Saving
Campaign
Data
Science
Pricin
g
Claims
Underwritin
g
1 4
3
2
Utilize the Open Source Data:
1) Energy Consumption
2) Weather
The ideation of GO GREEN home insurance campaign started with the exploration of the value of
open data such as Energy consumption and Weather - how can we leverage it for finding any
correlation with claims risk.
Throughout the processes from ideation, design, prototype and analysis, Data Science team had
worked closely with different teams including Pricing, Underwriting, Claims to get their collective
inputs, and then use open data to be applied with data science techniques - Machine Learning to
carefully analyze the significant factors then successfully concluded the discovery of correlation
between claims risk and our formulated green score
Public
4. 4
Public
21%
20%
10%
2%
1%
Keywords from Text Mining Results
Step 1:
Analysis & Insight - Extract Useful Information from Text Description in Home Claims Data
Firstly, we used text mining approach to analyze the home claims description keywords to understand the nature of home
claims. Within the keywords, property damage and water-related incidents occupied over half of home claims records, the
rest are distributed diversely in different types of claims.
To overcome the data disparity, we applied generalization for the keywords labeling/ grouping so as to produce a catalogue
of representable claims categories for later machine learning usage.
WordCloud from Text Mining on
AXA Home Claims Description
Public
5. 5
Public
000000000000000000000000000000000000
Number
of
claims
2017
Claims Frequency Distribution in Past Years
Figure 1. All claims
Figure 2. Removing weather-related claims
Outliers (weather-related)
What are the outliers?
We observed the outliers of claims counts mostly occurred
when extreme weather happened. After removing the
outliers, then we can see a clearer seasonal pattern.
Therefore, it is reasonable to separate the claims data by
non-weather-related vs. weather-related to ensure fair
analysis to the trend pattern baseline
Removing weather-related claims, an increased trend in
2020 is observed
Starting from 2020, COVID situation led more people stay
at home/ work from home resulting an increase of home
energy consumption. Coupled with the increasing trend in
claims frequency after 2020 is observed, we start to
explore the proof of hypothesis about correlation between
home energy consumption and claims risk in home
insurance, would it be caused by more human interactions/
wear and tear of fixtures, pipes, devices at home?
2018 2019 2020
Step 2:
Analysis & Insight - To See The Forest and The Tree, We Analyze the Claims Frequency
Public
6. 6
Public
𝐼𝑛𝑡𝑒𝑟𝑐𝑒𝑝𝑡 + 𝑇𝑒𝑚𝑝𝑎𝑡𝑢𝑟𝑒 + 𝑅𝑎𝑖𝑛 + 𝑆𝑢𝑛𝑠ℎ𝑖𝑛𝑒 𝐻𝑜𝑢𝑟 + 𝐺𝑎𝑠 + 𝐸𝑙𝑒𝑐𝑡𝑟𝑖𝑐𝑖𝑡𝑦 + 𝑅𝑎𝑖𝑛2
+ 𝐺𝑎𝑠2
= 𝐶𝑙𝑎𝑖𝑚 𝐹𝑟𝑒𝑞𝑢𝑒𝑛𝑐𝑦
* ** ** * **
*
+ 𝑅2
= 0.63 (ℎ𝑖𝑔ℎ𝑙𝑦 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑒𝑑)
AXA
Home
claims
Data
Temperature
Rain
Sunshine
Hour
Local
Consumption
of Gas
Local
Consumption
of Electricity
We tested the hypothesis if the longer stay-at-home due
to COVID led to the higher home energy consumption
then claims risk, plus if there would be any weather
factors also affecting home energy consumption level at
the same time
Open Data we used:
HKSAR Government Open Database
- Temperature / Rain/ Sunshine / Gas Consumption/
Electricity Consumption
Test Result:
A positive correlation among home claims, energy
consumption and a few weather-factors is confirmed
Implication:
Based on this correlation discovery, we further formulate
a predictive claims frequency formula as the basis for our
next step of calculating an integrated Green Score – to be
explained in later slides
HKSAR
Government
Open
Database
** highly statistically significant (P <0.01)
* statistically significant (P <0.05)
+
Parameterized calculation:
Step 3:
Analysis & Insight - Longer Stay-at-Home Time vs. Increased Risk Exposure?
Public
7. 7
Public
With the correlation quantitative support, GI Product team decided to introduce a new campaign in early Dec 2021,
ENERGY SAVING CAMPAIGN - By submitting the electronic bill with declined usage, customers would enjoy a discount
coupon offered. The campaign helps to promote customers’ mindset on energy saving, and the respondents’ data can
benefit the future planning of Dynamic Home Pricing
Less than 3 months, >500 customers participated this campaign showing they have lightened their carbon footprint.
Business Action: The Hypothesis Testing Result Contributed to the Launch of New ESG Campaign
Launched a new
campaign
Customers uploaded electronic
bill with declined usage &
received discount coupon
The respondents’ data can
benefit the future planning
of Dynamic Home Pricing
Public
8. 8
Public
Based on the correlation results we had, we further combined it with external
real estate database to calculate a new pricing factor model score for each
household.
How we do it?
Among 300+ property features from external real estate database, we used
machine learning model to carefully select the significant ones to formulate
the final integrated Green Score, it is a ready-to-use single new pricing factor
which can be used for future dynamic home pricing.
Business Impact #1:
A New Pricing Factor Derived From Machine Learning Model Can Be Used For Future Dynamic
Home Pricing
Public
9. 9
Public
Home claims can be categorized into weather-related claims and non-weather-related claims. Our study brings new perspective to Home
Insurance – “Weather matters”, insurers can differentiate strategies for weather vs. non-weather related claims. For non-weather related
claims, we can have the potential new pricing granularity down to individual’s green lifestyle.
Business Impact #2:
New perspective to Home Insurance: Smart Pricing Based on Different Claims Type Scenarios
Extreme-Weather-Related claims Non-Weather-Related claims
“Green
Swans”
Nature
• Mostly black rain, typhoon
• With small proportion in claims
frequency (6%)
• While with significant claims loss
once happened
Possible Strategies
Reducing known uncertainties by
labelling with higher risk
exposure, raising higher sum
insured, or rejecting offers to
avoid those known uncertainties
Nature
• Mostly property related
damage, ceiling, etc.
• With huge proportion in claims
frequency (94%)
• While with tiny claims loss
individually
Possible Strategies
Providing certain incentive like
more elastic home pricing and
nurturing customers to encourage
them having a low carbon
lifestyle
Public
10. 10
Public
A “Win-Win” Situation Between Customer And Insurer…
…A Pioneer For Future Carbon Exchange Between Insurers And Customers
Customer
Insurer
• More “Green” customers
• More precise pricing model
• Less claims losses
• Data from diversified angles for
future analytics
• Contribute to global “go-green”
action
• Precise pricing for home
product or premium
Discount
• Awareness of environment
protection and low carbon
lifestyle
Public