This document discusses using AI and data modeling to improve insurance operations. It provides examples of using structured text, customer data, and other sources to build models for risk assessment, fraud detection, and other tasks. The document emphasizes the importance of data quality and having a holistic data strategy. It also stresses an iterative approach to AI development and increasing skills. An example use case describes using natural language processing models to assist underwriters by providing relevant risk expertise and intelligence from various data sources.
2. Use DATA & MODELS to PREDICT
the most likely answer to a specific question
Today’s AI in Practice
DATA
Stats
Text document
IoT, sensors
Images
Spatial
Sound
Video
Click
Metadata…
EXAMPLES
Structure text documents as
statistical records
Customer segmentation, scoring
models, sentiment analysis,
chatbots…
Risk modelling, rating, loss
damage estimates…
Loss fraud classification, process
triage…
MODEL USE
Parse & structure information
Find patterns & ‘groups of
entities’ without a-priori
Calculate best estimates based
on set of inputs
Learn “rules” for accurate
classification tasks
[ a.k.a Narrow AI ]
4. AI QUEST for...
CUSTOMER
INTIMACY
Real-time Access
Personalized Relationship
Tailored Products
Accurate Risk Assessment
Ease of Interaction
Proactive & Reliable Service
Adequate Pricing Prevention & Assistance
Using AI to gain Deeper Understanding of Customers
5. AI QUEST for...
OPERATIONAL
EXCELLENCE
Deliver Economies of Scale
Scale Information Gathering
Improve Process Intelligence
Improve Service Delivery
Automate Workflows
Tracking & Monitoring
Maximize Customer Experience Minimize Leakages & Costs
Using AI automation to Drive Effectiveness & Efficiency
6. AI QUEST for...
RISK
EXPERTISE
Ensure Quality & Extent of Risks
Sales & Distribution
Optimize Value Proposition
Enterprise Risk Management
Underwriting & Engineering
Claims
Protect Shareholders’ Value Foresee Trending & Emerging
Risks
Using AI prediction to Manage Risks at next level
8. POOR DATA > WEAK MODEL > SUPERFICIAL INSIGHT
DATA IS OFTEN THE FIRST HURDLE TO OVERCOME
Fragmented & constrained legacy systems
Missing unified data architecture across sources
Dry upkeep investments, lack of data ownership....
GO FORWARD PRINCIPLES
Adopt holistic approach to data assets across organization
Ensure data flow beyond its traditional silos
Stress importance of «cross-functional» data needs
Focus capabilities on automated data capture & processing
Invest in external data rahter than external technologies
Define clear data ownership
9. AI ROADMAP > ITERATE > UPSKILL
ITERATE
UPSKILL
Heavy lift with the enterprise-wide data strategy
Not one single AI, but a portfolio of diverse AI use cases
Be selective & balance value-add vs. complexity / AI maturity
AI performance is gained through multiple iterations
Rethink process entirely rather than «patching» AI on top
Third party vendors may be strategic especially around data acquisition
Internalize knowlegde & skills to bridge business acumen with science
Outreach AI education & data-driven culture beyond technicians
Establish AI oversight : AI ethics, privacy, trust
Leverage & partner with Open Source for tech flexibility & durability
11. Underwriting Large Commercial Insurance
Complex risks, often with global exposures across different insurance markets
Very diverse industries requiring subject matter expertise not always available
locally
Bespoke coverages with nuanced terms & conditions adding complexity to
underwriting analysis
Insurance programmes typically exposed to low frequency - high severity loss
events
Rely heavily on qualitative risk assessment with concise market information
Price rarely a lever in current market conditions & does not mitigate alone large
risks exposures
MOST IMPORTANT FACTOR IS RISK SELECTION to warrant profitability across
the whole portfolio
12. Assist Underwriting with Risk Intelligence
By delivering all relevant expertise & know-how at the local «point of underwriting»
AI technique based on Natural Language Processing – NLP
All qualitative information in a submission automatically analyzed & linked against various
internal databases (eg. Corpus of directives, guidance papers, corpus of all portfolio submissions, risk
engineering documents, expert documents, claims intel, authority checks etc...)
The AI model enables us to extract all the relevant & applicable information based on
content semantic similarities
Save endless hours of manual work, enable comprehensive and consistent underwriting,
speed up response to market requests
Unlock all accumulated collective knowledge to augment the quality of the risk assessment
KEY OUTCOMES
Expand underwriting platforms with targeted risk knowledge & assisted intelligence
AI SOLUTION
13. ALBAN TRANCHARD
ACTUARY
I hope this is an informative perspective into AI
It is an exciting time to be in the insurance industry
AI is bound to unlock so much more value and we can all be the architects
of it
Always feel free to Connect and Message