Graph-Based Fraud Detection with Didit and Amazon Neptune

Leverage Graph DatabasesAmazon Neptune excels at identifying complex, non-obvious relationships in data, making it ideal for uncovering sophisticated fraud patterns that traditional relational databases might miss.

Integrate Didit's Rich Identity DataDidit provides high-fidelity, structured identity verification data, including biometric, document, and liveness insights, crucial for populating and enriching your fraud graph.

Uncover Fraud Rings in Real-TimeBy connecting data points like shared devices, addresses, and even facial biometrics, businesses can detect and prevent fraud rings proactively, improving overall security posture.

Didit's AI-Native AdvantageDidit's modular architecture and AI-native approach ensure that your graph-based system is fed with the most accurate and up-to-date verification signals, enabling dynamic and adaptive fraud detection workflows.

The Rising Threat of Sophisticated Fraud

In today's digital landscape, fraudsters are constantly evolving their tactics, moving beyond simple identity theft to orchestrate complex fraud rings. These networks exploit interconnected data points, often using synthetic identities, stolen credentials, and multiple accounts to evade detection. Traditional fraud detection systems, typically relying on rule-based engines and relational databases, often struggle to identify these intricate, non-obvious relationships. This is where graph databases, combined with rich, verified identity data, offer a significant advantage.

Imagine a scenario where multiple accounts are created from the same IP address, using different names but sharing the same physical address, or even similar facial biometrics across different identity documents. A relational database might flag individual suspicious activities, but a graph database can immediately visualize and connect these seemingly disparate events into a cohesive fraud network. By understanding these connections, businesses can move from reactive fraud detection to proactive prevention.

Why Graph Databases for Fraud Detection?

Graph databases are purpose-built to store and navigate relationships between data points, known as nodes and edges. This structure is inherently suited for fraud detection because fraud often manifests as a pattern of connections. For instance, an account (node) might be connected to a device (node), an IP address (node), an email (node), and a physical address (node). When multiple accounts share these connections, a graph database can quickly reveal these shared links, indicating potential collusion or a fraud ring.

Amazon Neptune, a fully managed graph database service, provides the scalability, performance, and security needed for real-time fraud detection. Its ability to perform fast traversals and pattern matching across billions of relationships makes it an invaluable tool. Instead of complex SQL joins across many tables, a single Gremlin or openCypher query can expose an entire network of suspicious activity, significantly reducing the time to detect and respond to fraud.

Integrating Didit Data into Your Fraud Graph

The effectiveness of any fraud detection system hinges on the quality and richness of its input data. This is where Didit, the AI-native identity platform, plays a pivotal role. Didit provides a comprehensive suite of identity verification primitives that generate high-fidelity, structured data essential for populating your Amazon Neptune graph.

Consider the data points Didit can provide:

• ID Verification: Didit's OCR, MRZ, and barcode scanning extract critical information from identity documents, such as names, dates of birth, document numbers, and issuing authorities. This data becomes foundational nodes in your graph.
• Passive & Active Liveness: Detecting deepfakes and presentation attacks, Didit's Liveness Detection ensures the person presenting the ID is physically present and real. A 'Liveness Status' (Approved, Declined, In Review) and a 'score' can be added as properties to a 'Verification' node, with a warning if a 'LIVENESS_FACE_ATTACK' is detected.
• 1:1 Face Match & Face Search: The similarity percentage from a 1:1 Face Match between a selfie and an ID document photo can be an edge property. If a 'FACE_IN_BLOCKLIST' warning is triggered by Face Search, this critical information can immediately flag a user in the graph.
• Proof of Address: Verifying residency adds another layer of connected data, linking users to physical locations.
• Phone & Email Verification: These data points are crucial for linking users to communication channels, often revealing shared resources among fraudsters.

Didit's API-first approach makes it seamless to feed this data into Neptune. As users onboard or undergo verification, Didit's responses, such as the liveness object with its status, score, age_estimation, and warnings, can be directly translated into nodes and edges within your graph. For example, a user node could be connected to a document node, a liveness_session node, an ip_address node, and a device node, with edges representing relationships like HAS_VERIFIED_DOCUMENT, PERFORMED_LIVENESS, USED_IP, or USED_DEVICE. Warnings like LOW_LIVENESS_SCORE or POSSIBLE_DUPLICATED_FACE can be attached as properties to the liveness_session or user nodes, triggering alerts or review processes.

Building Your Graph-Based Fraud Detection System

Here’s a simplified approach to building your system with Didit and Amazon Neptune:

1. Data Ingestion: Integrate Didit's APIs into your user onboarding and transaction flows. Capture all relevant identity verification data (ID details, liveness scores, facial similarity, warnings, etc.).
2. Graph Modeling: Design your graph schema. Define nodes for entities like Person, Document, Device, IP_Address, Email, Phone_Number, and Address. Define edges for relationships like VERIFIED_BY, USED_DEVICE, SHARED_IP, HAS_EMAIL, HAS_PHONE, LIVES_AT, HAS_LIVENESS_SESSION, and FACE_MATCHED_TO.
3. Populate the Graph: Use Didit's output to create and update nodes and edges in Amazon Neptune. For instance, when a user completes ID verification and liveness, create a Person node, a Document node, and a Liveness_Session node, along with edges connecting them. Add properties like liveness_score, document_type, or is_blocklisted to these nodes and edges.
4. Querying for Fraud Patterns: Develop Gremlin or openCypher queries to identify suspicious patterns.
   • Shared Devices/IPs: Find multiple Person nodes connected to the same Device or IP_Address node.
   • Synthetic Identities: Look for Person nodes with differing document details but strong facial similarity (from Didit's 1:1 Face Match) or shared addresses/emails.
   • Blocklist Matches: Immediately flag Person nodes where Didit's Face Search or AML Screening indicates a match to a blocklist or watchlist.
   • Low Liveness Scores: Identify Liveness_Session nodes with low scores or LIVENESS_FACE_ATTACK warnings, especially when combined with other suspicious connections.
5. Real-time Alerts and Actions: Integrate your graph queries with an alert system to notify fraud analysts or trigger automated actions (e.g., placing a transaction on hold, requesting additional verification, or declining an account) when a fraud pattern is detected.

How Didit Helps

Didit is uniquely positioned to be the cornerstone of your graph-based fraud detection strategy. As an AI-native, developer-first identity platform, Didit provides the precise, structured identity data necessary to feed and enrich your Amazon Neptune graph. Our modular architecture means you can pick and choose the verification primitives you need, from ID Verification and Passive & Active Liveness to 1:1 Face Match and AML Screening & Monitoring. This flexibility allows you to build highly customized and effective fraud detection workflows.

Didit's advantages are clear: we offer Free Core KYC, enabling you to start verifying identities and collecting valuable data without upfront costs. Our AI-native approach ensures high accuracy and resilience against new fraud vectors, while our clean APIs and instant sandbox make integration straightforward for developers. With Didit, you're not just getting a verification service; you're getting the foundational identity layer that automates trust and empowers your fraud prevention efforts, all without setup fees.

Ready to Get Started?

Ready to see Didit in action? Get a free demo today.

Start verifying identities for free with Didit's free tier.