How we developed our own global internal payment fraud system at Delivery Hero

10.04.24 by Kentaro Uscategui

How we developed our own global internal payment fraud system at Delivery Hero

Delivery Hero Logo

5 min read

Experience the multi-year story of how Delivery Hero developed its own internal fraud detection system to mitigate bad actors across our platform.

At Delivery Hero, our global presence spans over 70 countries with millions of orders flowing through our system every month. While the majority of these transactions are placed by loyal customers, we also encounter instances of fraudulent behaviour on our platform. These range from the typical use of stolen credit cards to complex social engineering tactics such as fraudulently gaining access to accounts. The ingenuity of fraudsters knows no limits. 

Customers across our brands around the world always expect a smooth experience. From the moment they open their app to when they receive their order, fraudsters are constantly trying to take advantage of the seamless experience that Delivery Hero provides. 

In order to mitigate this behaviour on our platform, the Payment Fraud Team at Delivery Hero is committed to ensuring a secure and frictionless customer experience while continuing to minimize fraud exposure on our platform. Over the last couple of years, we have utilized several third-party fraud and risk tool solutions to fulfil these commitments. 

The significant growth in our business since 2020, along with evolving and new fraud trends, has required us to be more forward thinking and explore the possibility of developing our own in-house fraud and risk tool solution. An in-house solution could provide Delivery Hero with greater flexibility, control over our fraud setup, and cost savings compared to the many third-party solutions we have used. 

The Decision

“We paid millions of dollars to our third-party fraud and risk tool solutions and these costs will continue to go up. While on the other hand, with an internally built system, we can control the costs, the tech, and not be reliant on someone else” – Mayank Gupta, Manager of Risk Engineering

It became clear that our reliance on third-party fraud solutions was becoming too extensive and there were limitations within the third-party products that hindered Delivery Hero’s technical needs and ambitions for the future.

A few years ago, we made the decision to collaborate internally with another team that was building its own risk engine for vouchers and other non-payment fraud types. This already built engine provided us with a strong foundation to expand this to other fraud types relevant to our team.  

From the beginning, the complexity of this project extended far beyond its scale. With millions of orders across several global regions, different payment methods, and regulatory requirements, comprehensive planning was essential.

“One of the biggest initial challenges was understanding how simple or difficult it would be from both the technical and cost perspectives to build a replacement for our third-party solutions. We needed a solid assessment otherwise we could end up building both the team and the internal system to figure out 1 to 2 years later that we’re much more expensive.” – Philipp Merker, Director of Global Payment Fraud

For the successful development and completion of the project, collaboration among specialized teams at Delivery Hero was crucial. This included Data Foundation, Payment Fraud Operations, Fraud Data Engineers, Fraud Data Science and Fraud Data Analytics teams. 

The Challenges

During the development phase, we encountered many hurdles that could impact the long-term efficacy of the process. To minimize adverse impact, continuous coordination among our key stakeholders was required. 

Simultaneously, we faced the challenge of integrating one of our major brands into our existing third-party solutions while accommodating additional requirements that were not initially planned. This compounded workload placed strain on the Fraud Data Engineering team limiting further our capacity to handle all tasks effectively.

As soon as the fundamental components of the internal fraud system became available, we started enabling it to deliver value as early as possible. In the beginning of 2023, cash payments within foodora, our European brand, became the first group to utilize features of the system by deploying rules to take action on risky cash transactions. Following the successful implementation of the cash payment method in Europe and APAC, we started designing the system to launch in foodpanda, our APAC brand, to accommodate other payment methods. 

Despite various challenges encountered throughout the process, by the end of 2023, the development of our internal fraud system was completed.  All planned regions were successfully migrated into our new setup as scheduled.

The Result

The new system brought benefits in several areas including significant cost savings per transaction once our third-party products were turned off. 

By eliminating our reliance on third-party support for adjustments to our fraud system, we removed the need to wait for external support which often comes with longer SLAs, significantly improving our response times. However, increased ownership and flexibility are not without their risks. Instead of relying on a third party with fixed fees, we are now accountable for any potential increases in cloud costs and for any temporary or even permanent loss of crucial engineering capacity. On the other hand, utilizing our in-house teams allows efficient resource allocation, scalability and flexibility that is vital to our business model and order volume.

“Training our Machine Learning models in our internal system is much faster and we can go from scoping to deploying in one month. We have the autonomy to create and serve complex features in real time, which was not possible before. The many improvements in our in-house solution have brought further autonomy in the decisions that the Data Science team makes.” – Vitor Louzada, Senior Manager of Data Science

One team benefiting from our new setup is our Data Science team, which has seen a reduction in the time to deploy new machine learning models as well as in the fully automated retraining of an existing model. In addition, there is a lot more stability and precise control of the model usage by treating it as a piece of code that must be subject to all software engineering best practices as part of our Reliability Manifesto. More information regarding the Reliability Manifesto can be found in another Delivery Hero Tech Blog post here

From the initial launch related to cash transactions in our European foodora market, we saw immediate improvements in data efficiency. 

“Data is extremely important in the context of fraud prevention and without it, you cannot have a setup to make decisions on. By having data from all the components of the flow – from the data that is being ingested, to data that is being processed, and the data outcome, we now have full control of this process” – Matias Arnoletto, Manager of Risk Analytics

Leveraging this data, along with our fraud prevention knowledge, enabled us to finally gain complete control and autonomy over all aspects of our anti-fraud setup.

The Future

While we successfully developed our internal fraud system, there is more work to be done. The focus now shifts to ensuring stability in the system, expanding into further regions, and continuously monitoring and assessing the various fraud types that we experience at Delivery Hero. 

As more Delivery Hero brands are incorporated into our new system, we will increase our focus on experimentation to continuously challenge and optimize our KPIs. Given the constantly changing behaviour of fraudsters, it is critical to stay one step ahead. 

Furthermore, we will enhance data collection throughout the customer journey to gain a more comprehensive understanding of customer behaviour, which is a key priority for us. This approach will enable us to expand our assessment to other critical points in the customer flow.

Our journey to enhance fraud prevention at Delivery Hero will empower us to proactively protect our operations, ensuring a secure and seamless experience for our customers while minimizing fraud risks on our platform.


If you like what you’ve read and you’re someone who wants to work on open, interesting projects in a caring environment, check out our full list of open roles here – from Backend to Frontend and everything in between. We’d love to have you on board for an amazing journey ahead.

How we developed our own global internal payment fraud system at Delivery Hero
Kentaro Uscategui
Fintech Fraud Manager
DeCoding the Journey: My First 90 Days as a Junior Data Engineer

Next

Engineering

DeCoding the Journey: My First 90 Days as a Junior Data Engineer

Delivery Hero Logo
5 min read