The Adyen way of engineering: Our tech stack is open source or built in-house

Being the most money-involved domain of FinTech, payment processing systems like Adyen are often targeted by individuals or sophisticated organizations. Fraudsters try to generate many synthetic identities like hundreds of different email addresses or hundreds of spoofed IP addresses hoping to slip-in our payment systems, but with the help of our in-house Adyen Graph Database, we can always track them in real-time and block their fraud attempts even though they keep trying new tricks.

Adyen Graph Database is one of our in-house products that helps us to catchfraudattempts as well asidentifyingindividuals and businesses trying to open an account in Adyen. You don’t want to conduct business with shady companies or sole proprietors!

Use cases like catching fraud attempts or identifying a business or a person during their onboarding to Adyen and continuously checking if they comply with Adyen and payment regulations, pattern matching is a must. To see and explain how payments are connected to each other, relational data structures are not the best choice on their own. There can be changing patterns, many hops between fake identities, recursive and exhaustive search helps us to see all the patterns and connections in real-time, enabling us to track criminals such as fraudsters or shady business owners attempting to open an account and start to process payments with Adyen.

With the payments data, you should be able to look at that from two different but combined angles: as atransactionaltime-series data and as aconnected graphto closely track potential criminals.

Having a pure traditionalrelational database systemcannot help as they are not good at handling connections and patterns.Graph databaseson the other hand are not good for processing high volume transactional data.

At Adyen, we need a completely different solution thatcombinesboth properties of graph databases and relational-transactional databases in a way that they also serve our specific requirements such as:

All these unique high-end requirements affect how we make decisions at Adyen. This resulted in the Adyen Graph Database being built on top of community accepted, open-source, relational databasePostgreSQLand open-source persistence frameworkMyBatiswhich supports custom SQL, stored procedures advanced mappings, and programming-language-like control statements that is managed by our internalgraph-enginewritten inJava.

At Adyen engineering, we believe in the simplicity of our code and our systems. Combining all the powerful features of PostgreSQL such as handling high amounts of transactional data with features of MyBatis, such as being simple, explicit, and close to Java POJO objects allowed us to write our tailor-made SQL queries and stored procedures that are very high-performant.

The different parts of the Adyen Graph Database work as below

TheJava application:

MyBatis:

PostgreSQL/plpgsql

The most powerful aspect of our graph database is that unlike a traditional graph database,it doesn’t traverse the whole graph each time during a payment – which in our use cases is an unnecessary and yet very expensive operation. But instead, for the sake of performance, each time a payment is happening we asynchronously generate an overview of statistics and analytics information that is compact and contains all the information we need to detect criminal activities such as fraud. 

Storing the relevant information in very simple, normalized key-value pairs we have a huge performance gain even on very big graphs.

With the exponential growth in our processed payments volume, we constantly look for improving our bespoke graph database solution to always scale better, be more extensible without scarifying the ability to be always ahead of the game before criminals as they constantly try new ways to penetrate into payments domain.