Pepperstone Uses Machine Learning on AWS to Deliver a Seamless Global Trading Experience

When choosing an online broker, trust and customer service are important criteria. Trading platforms with long histories and multiple regulatory licenses tend to be safe bets. In its Broker Awards for 2021, DayTrading.com named Pepperstone the Best Forex Broker, largely for its global footprint and “ultra-reliable platforms.” Pepperstone is a foreign exchange (forex) and securities online trading platform based in Melbourne with more than 40,000 unique visitors a month.

Since its founding in 2010, Pepperstone has relied on the Amazon Web Services (AWS) Cloud. “AWS offers frequent innovation and flexible, out-of-the-box solutions that help us run our business more efficiently,” says Tony Gruebner, chief marketing officer at Pepperstone. Containerization, and the use of out-of-the-box managed services such as Amazon Elastic Kubernetes Service (Amazon EKS), have been critical to Pepperstone’s ability to rapidly grow its customer base and scale its services securely. During 2020 alone, Pepperstone reported 20 to 30 percent growth.

With Amazon EKS, Pepperstone can standardize operations across any environment, leveraging automation for consistent multi-region deployments as it expands to new countries. In 2020, Pepperstone obtained five new regulatory licenses from agencies such as the Dubai Financial Services Authority, up from the two licenses it held previously in Australia and the UK. The company’s regulatory push opened up new regions, including Africa and Europe.

Machine learning (ML) and artificial intelligence (AI) are core technologies in Pepperstone’s tech stack. In addition to an IT team of 70 spread across four countries, the company has a data science team in Melbourne dedicated to developing ML models. Initially, data scientists developed their own algorithms to run on AWS. They then switched to Amazon SageMaker in 2019 with AWS Fargate to automate the creation and deployment of ML models.

When switching to Amazon SageMaker for model training, Pepperstone’s data science team had no trouble adapting the tool into their workflows. “The learning curve for new AI or ML tools is usually quite steep, but we got started with Amazon SageMaker right away. We saved a lot of time from day one by hosting, training, and deploying within the AWS environment,” says Samuel Ellett, lead data scientist at Pepperstone. The time required to train ML models has dropped from 180 hours on local machines to 4.3 hours on Amazon SageMaker.

To obtain new regulatory licenses, Pepperstone had to demonstrate it had a rigorous know-your-customer (KYC) process to screen traders coming onto the platform. Before implementing Amazon SageMaker, much of the document review process was manual. Staff would personally check submitted documents, such as passport images, for authenticity. Many times, they’d spend hours onboarding a client only to discover days later that they weren’t who they said they were.

Pepperstone can now recognize potential fraud the same day new customers upload their documents by deploying fraud-detection models in Amazon SageMaker. Data scientists trained its ML models to compare documents submitted against millions of images stored in an Amazon Simple Storage Service (Amazon S3) data lake, including both authentic and altered images of common ID documents across the globe.

The system then assigns a score to each potential customer with a percentage revealing the likelihood of illicit tampering within their submitted documentation. The onboarding team receives the results and follows up with flagged submissions to request additional proof of identity. This improves the team's decision-making process, which leads to a reduction in time spent processing manual ID verifications.

On top of time savings, the level of detail and accuracy Amazon SageMaker delivers is far superior to what humans could achieve. “It wouldn’t have been possible for humans to compare 3 million documents, and many times the fraudulent elements are so slight that it’s very hard for a human eye to detect,” says Ellett. 

Speeding up the onboarding process also benefits new customers. Competition is stiff among online trading platforms and traders aren’t locked into a particular platform. Therefore, a seamless onboarding process is critical to establishing trust with Pepperstone. “Automating onboarding with ML has helped us not only operationally, but has also created a more seamless process that massively improves customer experience. Our goal is to make trading enjoyable, and part of this is avoiding unnecessary delays or barriers to entry,” explains Gruebner.

Pepperstone is also using ML models generated in Amazon SageMaker to assist the sales team with lead assessment and conversion. Assigning a score to each customer when they onboard the platform allows the sales team to target customer service efforts. The score updates in real time as their customer data accumulates while navigating the Pepperstone site. This helps the sales team better manage workloads and provide a bespoke level of service that differentiates the company in the online broker market. “Amazon SageMaker is a fantastic tool to ensure the whole data science pipeline is as integrated and automated as possible, pushing data when and where you need it,” Gruebner adds.

As a result of the out-of-the-box functionality gained with Amazon SageMaker, Pepperstone’s data science team is no longer beholden to busy DevOps engineers. Data engineers still rely on the DevOps team to set up a sandbox environment in Amazon SageMaker, but once that’s done, they can independently run proof-of-concepts. “We’ve effectively unblocked a resource clash between the data and DevOps teams. Amazon SageMaker has made it easy for us to build something quickly, test our hypothesis, and close it down right after so it doesn’t cost too much. We’re carrying out more experiments as a result,” Ellett concludes.

 To learn more, visit aws.amazon.com/machine-learning and aws.amazon.com/startups.