This document discusses fraud detection using graph learning. It notes that fraud numbers are increasing each year as fraud becomes more complex and organized. Graph learning can help by providing a unified view of disparate data sources and enabling new insights through novel data connections. For corporations, fraud detection is predictive, while for legal enforcement agencies (LEAs) it is also investigative. Graph learning helps LEAs unify data from multiple sources and identify syndicates through community detection. While unifying data is challenging due to legacy systems and information silos, graph representations allow visualizing and computing on unified data. The document demonstrates how graphs can present relevant transaction details and connections to support fraud investigations. It recommends an approach using domain expertise, latest technologies, and

1. Fraud Detection and Compliance
with Graph Learning
Sai D. Burra
Founder and Chief Innovation Officer
Sept 2020

2. Introduction
• Sai D. Burra, Founder and Chief Innovation Officer
• Abhay Solutions and ASI Consulting
• We provide solutions to P&C industry and LEAs
• Regional presence: USA, UK, Singapore, India and Philippines.
•

3. Data Challenges in Fraud detection and Analytics
• Fraud numbers and volume are increasing each year
• Fraud complexities are increasing
• Organized crime is biggest threat
• Volume and Variety of data
• Legacy systems
• Multiple application / information sources
• Legislation / Regulatory requirements
• Data Privacy

4. Fraud detection and analytics for Corporates
• Fundamentally graph brings new representation of data to existing data computation
technologies and enables new insights.
• Fraud solutions in corporate are more predictive in nature and focused on identifying Illegit
transactions.
• Fraud detection is predictive in nature and helps focus on critical cases.
• Fraud examiners also use analytics for understanding patterns to improve investigations, process
and procedures within the company.
• Graph helps AI and ML computation engines in improving detection accuracy through newly
established attribute level relationships.

5. Fraud detection and analytics for LEAs
• Legal enforcement agencies (LEA) are more investigative in nature though some sections use
predictive analytics for real-time fraud identification like Immigration, Tax and Bank regulatory
authorities
• LEAs have very big challenge of collecting and unifying data from multiple agencies and
authorities like (Banks, phone companies, tax authorities, financial institutions).
• LEA are not just bound to Identify fraudulent transactions but also build complete conspiracy and evidence.
• Use Analytics to examine existing cases to investigate and to build evidence / circumstantial
evidence.
• Unified view of data is extremely important for LEAs
• Community detection plays big role in identifying Syndicates and group frauds

6. How Unified data helps
• Graph representation allows for novel ways of connecting disparate information.
• Visualization of unified data is made simple by Graph databases.
• Though effort is required to harmonize the information to a schema.
• Data computations is made simple across disparate information by graph.
• Hash functions provide a way combine multiple graphs built from different sources.
• LEAs unification of data is more complex than corporates.
• Graph learning technologies amplifies data computation capabilities on large
data.
•

7. Why Unified View is difficult
Insurance Industry example
Billing Information
Claims Information
Life Event Triggers
Social Media
External Data
Quotes & Applications
Preferences
Policy Information .…
Product Information ……
Marketing Information .nb
Digital Information …..
Household Information …
Agent / Sales Information …
Call Center Information ….
Challenges
Disparate Systems
Information Silos
Legacy Systems
Multiple Records

8. Why Unified View is difficult
Legal Enforcement Agency example
Criminal database
Suspects info
Emails
Social Media
External Data
Onfield Reports
Preferences
Case info .…
Call Data Records ……
Business Accounts info .nb
Payment Gateway info …..
Bank statements …
Computer forensic info …
Mobile forensic info nfo
Challenges
Disparate Systems
Information Silos
Legacy Systems
Multiple Records

9. Graph learning helps Fraud investigations
• Let’s look into benefit of Graph representation in Fraud investigation:
• Demonstrate complete picture of a particular transaction (Claim in this
example)
• Demonstrate relevant information in one picture instead of 25 records/rows.
• Document the transaction
• Present any relevant connections ( to signify group participation in fraud)
• Exchanging Case information between departments (in companies) /
Authorities (in LEAs).
• Summarize pages of data and explanation in one view.
• Few pages instead of 10s and 100s of pages.

10. Graph learning in Fraud investigations

11. Graph learning in Fraud investigations

12. Graph learning in Fraud investigations

13. Recommended approach
• Domain Expertise + Latest Technology + Algorithms
• Focus on compliance
o Detect
o Document
o Demonstrate
• Pull data from current applications as required
• Automate Workflow/Orchestration for Data Movement, Process flow, and Flagging
• Co-exists and integrate with current applications (Not a Rip & Replace)
• Enable Web and Mobile interface
• Choice of Cloud and / or On-premise solutions
• Take advantage of modern applications stack

14. Q & A
• Name: Sai D. Burra
• Email id: SaiBurra@AbhaySolutionsInc.com
• Phone: +1 630 853 2257
• Website: www.AbhaySolutionsInc.com