Fraud Graphs

Globally each year, organizations lose tens of billions of dollars to fraud. Fraud can be carried out by a single bad actor, or a network of participants using multiple identities colluding with each other for transactions with business institutions. Multiple people can collude to commit fraudulent transactions, creating fraud rings, which may have hundreds of members, making it challenging to find the bad actors and detect fraud. Transactional data, for example credit card transactions, has basic attribute information, but does not include the relationships such as those between people, shared addresses, and others. This makes it hard to identify fraud that is committed by groups of coordinated actors or by a single actor over time.A fraud graph stores the relationships between the transactions, actors, and other relevant information to enable customers to find common patterns in the data and build applications that can detect fraudulent activities. Using a fraud graph, organizations can identify a network of connected users and items such as e-mail accounts, addresses, and phone numbers that they have in common. This results in a highly connected and complex network of information, which can be queried, visualized, and analyzed to detect fraud. Fraud graphs are complementary to other techniques used to detect fraud, for example using Machine Learning models to identify potentially fraudulent transactions, and querying a fraud graph about the actors related to the transactions. You can build your fraud graph solution using Amazon Neptune.

Relational databases, built for storing and analyzing tabular data, are not efficient at storing and querying the relationships between billions of interconnected entities where you need to explore and visualize connections and groups within the data. Using a relational database to query large relationships can be complex, where using SQL to query the database can result in multiple complex joins leading to poor performance.

Graph databases, which are purpose-built to store and navigate relationships, and their query languages, are designed to work with data that is highly connected, making querying the data for patterns and connections simple, fast, and reliable. Graph databases treat relationships as “first-class citizens,” have flexible schema, and provide higher performance for graph query traversals. This makes graph databases capable of sophisticated fraud detection and prevention. With graph database, you can model relationships between people, places, and financial transactions in real time and discover additional relationships that may not be obvious.

Infinitium refactored their increasingly complex rule-based system to leverage Amazon Neptune, a managed graph database together with DocumentDB and MemoryDB to more efficiently detect fraud for their customers.

Using Amazon Neptune, Rappi was able to use Neptune for their fraud detection solution, replace their 3rd party solution and bring down their overall cost significantly.