EY + Neo4j: Why graph technology makes sense for fraud detection and customer 360 projects
Why graph technology makes sense for
fraud detection and customer 360
projects in Insurance
March 2023
Introduction
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
► 13+ years of experience in data science diversely spread across consulting,
industry, and start-up
► 3+ years in graph technology building data products
► Currently leading the tech for wavespace AI Labs and Data Science for Utilities at
EY Ireland
► A data scientists & a professional accountant who bridges the gap between
technology and business value
► Active contribution to Data Science community and a vocal supporter of DE&I in
Data Science
EY has a large and growing graph
practice, with over 200 consultants
globally.
We see a wide range of graph use cases
across all sectors and have delivered
several compelling graph solutions to
help our clients drive greater insight,
efficiency and value.
EY and Graph Technology
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Plasma
Donor 360
Retail
Customer
360
Customer
Identity
Enterprise
Org
Design
FinServ
Know Your
Customer
Regulatory
Reporting
Data
Lineage
Anti-Money
Laundering
GCN
Cruiseline
Activity
NBA
Batch
Geneaology
B2B Event
NBA
Capital
Projects
Cost
Visibility
COVID-19
Risk
Tracking
Fuels
Tradiing
Forecasting
Global
Compliance
Monitoring
Active
Directory
Access
Controls
Financial
Ledger
Transaction
Lineage
FINANCIAL
SERVICES
SALES &
MARKETING
ENERGY
ASSET
MANAGEMENT
LIFE
SCIENCES
RISK
EY SOLUTIONS
Inability to recommend
Next Best Action
(NBA)
Non-optimized fraud
identification and actioning
capabilities
Lack of full view of
customers and agents
Insurers today are struggling with identity resolution which impacts
growth
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Silo-ed legacy systems
Fast changing
Customers needs
Broker / Aggregator
mediated market
New fraud trends like
Deepfakes
Reactive & rule-based
policies
Operations at Scale
Leads to:
Many companies today utilize Customer Graphs:
To support the demands of the digital
business, enterprise architects must
consider how best to link large volumes of
complex, siloed data... Graph
databases are a powerful
optimized technology that link billions
of pieces of connected data to help create
new sources of value for
customers and increase
operational agility for customer
service.
– Forrester
Zurich
Large online
shopping site
These challenges have been successfully solved using graph databases
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Customer 360° View in an Insurance Company
Market-ing
Sales Policy
Claims
Contact
Centre
Broker
External
Data
Demographi
cs
UNIFIED VIEW
OF THE CUSTOMER
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
A unified Customer 360° view enables
• Data-driven, customer-centric experiences
• Efficient and automated sales & marketing
• Improved compliance and better underwriting
through fraud detection
• Consistent view of operational metrics across
business segments
• Improved decision-making based on more
reliable reporting
Why graph database is better than SQL
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Graph can add value in any environment where:
Data is interconnected and
relationships matter
Data needs to be read and queried
with optimal performance
Data is evolving and data model is
not always fixed and pre-defined
Before
Name Address Policy Claims Broker
Phone
Customer Golden Record
LOB - System 1
LOB - System 2
LOB - System 3
Agent Quote
Source Systems EDW Schema Slow Execution
Why graph database is better than SQL
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Graph can add value in any environment where:
Data is interconnected and
relationships matter
Data needs to be read and queried
with optimal performance
Data is evolving and data model is
not always fixed and pre-defined
After
LOB - System 1
LOB - System 2
LOB - System 3
Agent Quote
Source Systems Graph Schema Faster execution
Customer Golden Profile
Customer Golden Profile will create cross-LOB data assets to help
answer key strategic questions
Integrating data into the graph and building analytics and BI-layers on top to enable rapid extraction of cross-LOB
features on a customer for context-based decision making
Rapidly test and operationalize
new analytical capabilities
• Who are our
customers?
• What drives a
customer to make a
buy decision?
• How to understand
different customer
behaviours?
• How to get right
and up-to-date
information about
every customer?
• How to create
effective risk
policies?
We want to
understand…
Key components of the Customer Golden Profile
Example Capabilities
• Predict Churn
• Personalise product bundling
• Optimise discount via agent
effectiveness
• Predict conversion in sales
cycles
• Predict effectiveness of cross-
sell & up-sell schemes
• Predicting fraud triangles
• Effective Chatbot for Contact
Centre activities
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Identify & ingest
multiple data
sources
Link and
maintain graph
database
Create new data
signals and products
Sales / Marketing
• Customers are not always “price sensitive” but “value sensitive”
• Referral programs are effective along with product bundling
• Agent is the “influencer” but customers validates the information online
• Discount optimisation based on “influence capability” of the agents
Risk & Compliance
• Increased risk exposure due ignoring past performance (linking historical
policies for artificial persons)
• Common elements between claims - like garage, doctor, 3rd party in car &
liability insurance, etc.
• Loss of opportunities from traditional rule-based risk policies – E.g., a young
driver is not always the riskiest driver; add more dimensions for risk validation
Some of the interesting insights were
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Proving the value of Graph technology
Increase Cross-
Sell and Upsell
Increase
Retention
Increase Customer
Satisfaction
Reduce Cost to
Acquire and
Service
Reduce Fraud
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
And how to measure them?
Value(€) and Volume(#)
of policies sold to
existing customers in a
year
Measure what matters . . .
Annual customer churn
rate across and within
LOB Straight-Through-
Processing policies
Direct and Indirect
expenses by Customer
Journey milestones like
Quote, Policy,
Operations, etc
Average time-to-resolve
at Contact Centre
Average of CSAT score
and Annual NPS score
Loss and combined ratio
Production Build
Cloud Pilot
Localhost POC
Graphy Problem
Business need, Data sources Data modeling, API, example
queries
Data snapshot, reference
architecture, API suite
Hardening, scheduled & stream
ETL, Live UX
Stakeholder Input
Problem / Scope
What will the graph
solve?
Graph Design
Data Work
APIs / Data Services
Integration
Scale
Validate
What questions can
now be answered?
Connect
Does the data support
the graph model and
semantics?
Mobilize
What data does the
new experience need?
Use Cases
What is the feedback
from the business on
how well the graph
solves the use case?
Deploy
What monitoring,
testing, process needs
to be put in place to
achieve a robust SLA?
Start small and scale
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Asking better questions
Making Graphs work is not a sprint but a marathon
o Once the data integration phase is complete, the environment is ready for iterating through several 12-16 week use case sprints
o At the end of each sprint, an assessment of the results, in terms of revenue and cost benefits, will guide the decision for additional
Use Cases
o In parallel to each sprint,
o Inform the senior stakeholders on current decision processes to develop more Use Cases for the backlog
o Identify “evangelist” business users for early adoption and acting as voice of influence amongst end-users
Continuous Use Case Development – Sustain and Scale
User
Interview
Sprint Backlog and
Scheduling
Business Use
Case
Business Use
Case
Business Use
Case
Business Use
Case
Model
Development
Industrialisatio
n
BAU
Operations
Strategic
Reporting
Self-service
reporting
Code Config BI / MI
Monitoring Controls Automation
One-off outputs
that cannot be
sustained are
retired after use
Outputs
decommissioned if
not deemed feasible
Delivery Pod
Delivery Pod
Delivery Pod
Successful use cases, Pod move
into build
Experiment failed, Pod spun down
Delivery Pod
Delivery Pod
Pod creates a single use model
Pod output planned for BAU run
Model
libraries
Maintenance
Sprint Case 1
Sprint Case 2
Sprint Case 3
Sprint Case 4
Sprint Case 5
Sprint Case 6
…
-- W1
-- W4
-- W7
-- W10
-- W13
-- W16
-- …
Allocation
to
Delivery
Pods
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Thank you! Questions?
Why graph technology makes sense for fraud detection and customer 360 projects in insurance
Internal database
Customer demographics (master)
Quotations generated
Policy
Claims
Payments data
Broker
Call centre interactions (enquiry, complaints, requests)
Web portal interactions (online transactions – policy renewals, FNOL, claims, etc.)
Social media interactions (enquiry, complaints, requests, feedback)
External database
Companies database (non-individual customers)
Address database – entity resolution
Point of interest – OSM
GIS data
Weather data
Property registration, Marine vessels, Car registration data
A unified view of the customer is foundational to a successful digital transformation. The customer 360° view is derived from customer, product, sales, marketing, support and web data. Data is ingested & cleaned in a data lake, unified & analyzed in a Knowledge Graph, and mobilized via API microservices.
Using the same license plate, three different quotes were made with different names, addresses and birth dates. Therefore, the last quote was simply accepted
Key learnings:
1. Envision end product at data modelling phase
2. Fail Fast. Learn Faster. (Agile)
3. Shipping beats perfection (Product)
4. Do not underestimate the dependency on admin, IT Infrastructure & Security
5. Do not underestimate Data Quality Improvement
6. Early business adopters becomes product evangelist