Your Enemies Use GenAI Too! Staying Ahead of Fraud with Neo4j Knowledge Graphs_Gartner Symposium Gold Coast, Australia 2024, Emil Pastor, Head of Solution Architecture A/NZ, Neo4j

Fraudsters can (and are) adapting new technologies faster than we can, giving them an advantage in adapting their techniques to remain undetected. Off-the-shelf applications cannot adapt quickly enough to protect us. See how you can build applications on a flexible native graph database and data science algorithms to uncover and investigate complex fraud quickly.

1. INTERNAL ONLY Title: Your Enemies use GenAI Too: Staying Ahead of Fraud with Neo4j Knowledge Graphs Abstract: Fraudsters can (and are) adapting new technologies faster than we can, giving them an advantage in adapting their techniques to remain undetected. Off-the-shelf applications cannot adapt quickly enough to protect us. See how you can build applications on a flexible native graph database and data science algorithms to uncover and investigate complex fraud quickly.

2. Your Enemies use GenAI Too! Staying Ahead of Fraud with Neo4j Knowledge Graphs Emil Pastor | Head of Solution Architecture, Neo4j ANZ

3. 3 Neo4j Inc. All rights reserved 2024

4. 4 Neo4j Inc. All rights reserved 2024

5. 5 Neo4j Inc. All rights reserved 2024

6. 6 Fraud Losses are Skyrocketing Fraud is expected to surge 20% annually in coming years 1 $42B Lost to fraud globally in 2022 2 Neo4j Inc. All rights reserved 2024 1. Experian Research: Identity Theft Is on the Rise, Both in Incidents and Losses 2. PwC’s Global Economic Crime and Fraud Survey 3. BioCatch: 2023 APAC Digital Banking Fraud Trends Report 4. WalletHub: Credit Card Fraud Statistics Additional costs for each dollar of credit card fraud losses 4 $3.75 200% Surge in voice scams from 2022 to 2023 3

7. 7 A.I. is Primed to Boost Financial Fraud and make it much harder to spot Beware Of FraudGPT, The Rogue AI Chatbot Deepfake Imposter Scams Are Driving a New Wave of Fraud I Cloned Myself With AI. She Fooled My Bank and My Family Neo4j Inc. All rights reserved 2024 How criminals are getting help from AI hacking tools Generative AI contributes to increase in cybercrimes

8. Neo4j Inc. All rights reserved 2024 8 Fraud is Everywhere Push Payment Fraud Insurance Claims Fraud Digital Wallet or Retail Fraud Supply Chain Fraud, Procurement Fraud Many Others Credit Fraud Tax Fraud/ Criminal Investigation Internal Fraud Checks Pandemic Fraud/Medical Insurance Fraud

9. 9 Neo4j Inc. All rights reserved 2024 Why Fraudsters Have the Upper Hand Criminals rapidly adopt AI to exploit vulnerabilities in real-time Enterprises struggle with legacy and data silos, taking weeks or months to detect new fraud schemes

10. Neo4j Inc. All rights reserved 2023 10 Existing Traditional Systems Fall Short Inability to Uncover Complex Fraud Patterns Can’t Scale to Deliver Rapid Analysis Difficult to Adapt to Evolving Threats Analysing data without relationships fails , as entity resolution, and complex fraud detection requires specific algorithms Rigid database schema requires refactoring whole database and involving cumbersome, lengthy SQL queries Recursive fraud patterns evade detection due to slow, unscalable queries

11. Identify suspicious activity quickly and accurately with pattern matching Scale with any cloud and speed up fraud detection and investigation 11 Improve and evolve fraud applications with a flexible schema Neo4j Inc. All rights reserved 2024 How does Neo4j Knowledge Graphs help build Powerful Fraud Applications to Uncover Hidden Patterns

12. Neo4j Inc. All rights reserved 2024 12 TODO1 Increases Fraud Detection by 200% with Neo4j Impact Using Neo4j, TODO1's iuviPROFILER increased fraud detection by 200% while maintaining the same false positive rate. Challenge Latin America's rapidly growing digital banks must instantly assess transaction risks to maintain seamless user experiences, but traditional databases are inadequate for real-time detection. Solution TODO1 created iuviPROFILER, using Neo4j, to efficiently process data and assess risk, enabling banks to securely conduct transactions with minimal friction, even during peak demand times. iuviPROFILER’s Graph by the Numbers – 250 million nodes – 2.2 billion relationships Graph Performance by the Numbers – >500 transactions per second – 100 milliseconds per query For the same false positive rate, we’re able to achieve twice the detection rate.’ he said. ‘And that in the end is less friction for the customer, less losses for the bank, and a better feeling in terms of protection for their customers. Edgar Osuna Chief Data and Analytics Officer “ “ Read the full customer story on Neo4j.com

13. Identify suspicious activity quickly and accurately with pattern matching Scale with any cloud and speed up fraud detection and investigation 13 Improve and evolve fraud applications with a flexible schema Neo4j Inc. All rights reserved 2024 How does Neo4j Knowledge Graphs help build Powerful Fraud Applications to Uncover Hidden Patterns

14. Identify suspicious activity quickly and accurately with pattern matching Scale with any cloud and speed up fraud detection and investigation 14 Improve and evolve fraud applications with a flexible schema Neo4j Inc. All rights reserved 2024 How does Neo4j Knowledge Graphs help build Powerful Fraud Applications to Uncover Hidden Patterns

15. 15 Identify Suspicious Activity Quickly and Accurately Discover and investigate hard-to-find fraud quickly with pattern matching Find fraud patterns 1000x faster than relational databases Expose intermediaries and find fake profiles through pathfinding and entity resolution Neo4j Inc. All rights reserved 2024

16. Quickly Uncover Hard-to-Find Fraud Schemes Neo4j Inc. All rights reserved 2024 16 Users sharing the same IP and Card Users sharing the same IP and Card Connection of Unique Users sharing the same device, IP or card User IP Card Device Legend

17. Quickly Uncover Hard-to-Find Fraud Schemes Neo4j Inc. All rights reserved 2024 17 Create an explicit “SHARED_IDS” relationship Explicit relationship between users User IP Card Device Legend

18. Quickly Uncover Hard-to-Find Fraud Schemes Neo4j Inc. All rights reserved 2024 18 Execute the Weakly Connected Component (WCC) Algorithm to identify potential first party fraudster Graph ML’s results are visually explainable and interpretable Group 2 Group 1 Group 3 User IP Card Device Legend

19. 1000x Faster Than Relational Databases Neo4j Inc. All rights reserved 2024 19 Connectedness and Size of Data Set Response Time Relational joins and other NoSQL key-value retrievals Native Graph Database 1000x Faster Minutes to milliseconds Many hops > 3 degrees Thousands of connections 0 to 2 hops 0 to 3 degrees Few connections

20. Expose Intermediaries and Find Fake Profiles Resolve entities using GQL (Cypher) queries and graph algorithms like Node Similarity and Weakly Connected Components Uncover connections between fraudulent actors and intermediaries using pathfinding algorithms like Yen’s and Dijkstra Source-Target Make link predictions between identities in the graph, create unified views of individual identities, and group communities of nodes using community detection algorithms Neo4j Inc. All rights reserved 2024 20 Known Fraudster 1 Registered Address Registered Phone Number Registered Business Known Fraudster 2 Name: Johan Nordberg Age: 32 Occupation: Pilot Employer: SAS Address 1: Arlanda, Sweden Address 2: Uppsala, Sweden Bank Acct: XXXXX934 Name: Johan Nordberg Age: 32 Occupation: Pilot Employer: SAS Address: Uppsala, Sweden Bank Acct: XXXXX934 Name: Johan Nordberg Age: 32 Occupation: Pilot Employer: SAS Address: Arlanda, Sweden Bank Acct: XXXXX934

21. Identify suspicious activity quickly and accurately with pattern matching Scale with any cloud and speed up fraud detection and investigation 21 Improve and evolve fraud applications with a flexible schema Neo4j Inc. All rights reserved 2024 How does Neo4j Knowledge Graphs help build Powerful Fraud Applications to Uncover Hidden Patterns

22. Identify suspicious activity quickly and accurately with pattern matching Scale with any cloud and speed up fraud detection and investigation 22 Improve and evolve fraud applications with a flexible schema Neo4j Inc. All rights reserved 2024 How does Neo4j Knowledge Graphs help build Powerful Fraud Applications to Uncover Hidden Patterns

23. 23 Improve and Evolve Fraud Applications Reduce exposure to fraud by linking disparate and disconnected data sets Stay ahead of fraudsters by detecting and modelling new or emerging fraud patterns quickly Enable wider and easier use of data by connecting siloed data into an intuitive data model Neo4j Inc. All rights reserved 2024

24. Reduce Exposure to Fraud Create new relationships without having to add new foreign keys, intersection tables, or joins Neo4j Inc. All rights reserved 2024 24 Enrich your Knowledge Graph with a flexible schema that enables additional data and semantic layers without refactoring the database or rewriting application code Enable wider and easier use of data by connecting siloed data into an intuitive data model Linking Disparate and Disconnected data Sets Faster

25. Build an Agile and Intuitive Data Model Neo4j Inc. All rights reserved 2024 25

26. Business Understanding = Physical Data Model Neo4j Inc. All rights reserved 2024 26

27. Graph Data Model Grow and Evolve Neo4j Inc. All rights reserved 2024 27 User :HAS_CC IP User Device IP Device IP Card Card :HAS_CC :HAS_IP :HAS_IP :USED :USED :USED :HAS_CC :HAS_IP :HAS_IP Sharing the same identity User IP Card Device Legend

28. Build New Relationship as Patterns Emerge Neo4j Inc. All rights reserved 2024 28 :SHARE_IDENTITY User :HAS_CC IP User Device IP Device IP Card Card :HAS_CC :HAS_IP :HAS_IP :USED :USED :USED :HAS_IP :HAS_CC :HAS_IP User IP Card Device Legend

29. … In Real-Time Neo4j Inc. All rights reserved 2024 29 :SHARE_IDENTITY User :HAS_CC IP User Device IP Device IP Card Card :HAS_CC :HAS_IP :HAS_IP :USED :USED :USED :HAS_IP User :HAS_IP :USED :SHARE_IDENTITY User :HAS_CC :HAS_IP User IP Card Device Legend

30. Pattern Matching Neo4j Inc. All rights reserved 2024 30 User IP Card Device Legend :SHARE_IDENTITY User :HAS_CC IP User Device IP Device IP Card Card :HAS_CC :HAS_IP :HAS_IP :USED :USED :USED :HAS_IP User :HAS_IP :USED :SHARE_IDENTITY User User User :HAS_CC :HAS_IP

31. …Within the Same Knowledge Graph Neo4j Inc. All rights reserved 2024 31 :SHARE_IDENTITY User :HAS_CC IP User Device IP Device IP Card Card :HAS_CC :HAS_IP :HAS_IP :USED :USED :USED :HAS_IP User :HAS_IP :USED :SHARE_IDENTITY :HAS_CC :HAS_IP :SHARE_IDENTITY User User User :SHARE_IDENTITY User IP Card Device Legend

32. Don’t Spend Time Writing Long SQL Neo4j Inc. All rights reserved 2024 32 --Find Fraud Ring 3-10 txns select a.account_id , a.first_name , a.last_name , a.city , a.state , a.zip , t1. * , a2.account_id, a2.first_name, a2.last_name,

33. Neo4j Inc. All rights reserved 2024 33 Don’t Spend Time Writing Long SQL

34. Use GQL (Cypher) to Stay Ahead of Fraudsters Find larger fraud patterns without changing queries using GQL’s (Cypher) Quantified Path Patterns and recursive pattern matching Incorporate new fraud patterns into the code quickly using GQL (Cypher) queries that are shorter than the equivalent SQL queries or programmatic code Neo4j Inc. All rights reserved 2024 34 --Find Fraud Ring 3-10 txns MATCH ring=(a: Account ) (()-[: PERFORMS ]->()- [: BENEFITS_TO ]-()){3,10}(a) RETURN ring;

35. Identify suspicious activity quickly and accurately with pattern matching Scale with any cloud and speed up fraud detection and investigation 35 Improve and evolve fraud applications with a flexible schema Neo4j Inc. All rights reserved 2024 How does Neo4j Knowledge Graphs help build Powerful Fraud Applications to Uncover Hidden Patterns

36. Identify suspicious activity quickly and accurately with pattern matching Scale with any cloud and speed up fraud detection and investigation 36 Improve and evolve fraud applications with a flexible schema Neo4j Inc. All rights reserved 2024 How does Neo4j Knowledge Graphs help build Powerful Fraud Applications to Uncover Hidden Patterns

37. 37 Scale with any Cloud and Speed up Fraud Detection and Investigation Deploy your applications anywhere Find fraud in ever-larger data sets and scale up number of users and concurrent transactions Integrate with your existing enterprise ecosystem Neo4j Inc. All rights reserved 202 4

38. Focus On Your App, Not Infrastructure Neo4j Inc. All rights reserved 2024 38 Automated upgrades, zero maintenance Scalable and elastic, on-demand Enterprise-grade security High availability

39. Flexible Cloud Deployment Models Fully-managed SaaS Consumption- based pricing For private and hybrid cloud, or on-prem White-glove managed service by Neo4j experts 39 Neo4j Inc. All rights reserved 2024 Self-Hosted Cloud Managed Services Graph-as-a-Service Cloud-native Self-service deployment Fully customizable deployment model and service levels Operate In own data centers or Virtual Private Cloud Bring your own license Full control of your environment Run in any cloud, in your account

40. Build Applications for Speed and Scale Neo4j Inc. All rights reserved 2024 40 Speed up and scale fraud investigations with an adaptive query planner and clustering Get 100x faster answers with compute intensive queries that require traversal across large parts of the graph using morsel based parallel runtime Autonomous Clustering Easy, Automated Horizontal Scale-Out

41. Integrate with Your Existing Enterprise Ecosystem 41 Neo4j Inc. All rights reserved 2024

42. Preventing Payment Fraud in Real-Time for a Top Gaming Subscription Network (1/2) Neo4j Inc. All rights reserved 2024 42 Real-time detection of suspicious transactions Enrich feature-engineering process of existing ML model Objectives The client must pay a penalty to the bank for each fraudulent account. “Every percentage improvement in the model can save millions” Business Drivers Look into specific suspicious patterns (Small-Small-Big) around transactions and particular users Are there any entities, like Emails and IPs, experiencing bot attacks? Can we identify suspicious accounts based on shared entities? Can we have graph-based scores for such entities? Key Questions

43. Preventing Payment Fraud in Real-Time for a Top Gaming Subscription Network (2/2) Neo4j Inc. All rights reserved 2024 43 Suspicious Patterns flying under the radar F1 Metric( >40% uplift) 72% vs 30% Recall ( >20% uplift) 77% vs 55% > $250k / Week > $1M / Month 32% of the Top 25 features powering predictions are Neo4j Features > $13M / Year 200ms response time 600M nodes

44. Neo4j Inc. All rights reserved 2024 44 Banking & Financial Services Technology Telecommunications Energy E-Commerce Health & Life Sciences 1,700+ Organizations Use Neo4j

45. Identify suspicious activity quickly and accurately with pattern matching Scale with any cloud and speed up fraud detection and investigation 45 Improve and evolve fraud applications with a flexible schema Neo4j Inc. All rights reserved 2024 How does Neo4j Knowledge Graphs help build Powerful Fraud Applications to Uncover Hidden Patterns

46. Neo4j Inc. All rights reserved 2024 46 Proactively Protect Your Customers with Knowledge Graphs! Make Sense of Your Data with Neo4j to Outsmart Fraudsters!