Accelerating Payment Approvals by 75% using Amazon Bedrock

Founded in 2008, Trustly empowers users to make secure transactions to merchants from their bank accounts. The company needs to authorize or decline payments quickly—in real time—but its previous system wasn’t set up to analyze such a large amount of data. The data science team needed a better way to pinpoint exactly which data needed to be analyzed by its fraud detection system.

Trustly had been outsourcing fraud detection and using private data centers. In 2015, it began moving to AWS. The company migrated its data to Amazon Relational Database Service (Amazon RDS), an easy-to-manage relational database service optimized for total cost of ownership.

Then, Trustly created its own fraud detection ML models using Amazon SageMaker, an integrated experience for analytics and AI. At one point, the company tested an open-source solution to run its ML model. But Trustly quickly decided to keep using Amazon SageMaker because it powered faster innovation and empowered Trustly to keep up with rapidly evolving fraud threats.

To maximize the speed of its ML model, Trustly needed to categorize the transaction data going into the model to help it better understand an account’s activity and authorize or decline the payment more quickly.

The company began using Amazon Bedrock, a service for building generative AI applications and agents at production scale, to categorize banking activity with AI. “Using Amazon SageMaker, we manage the whole lifecycle for our ML models,” says Denner Padilha, head of IT operations and security at Trustly. “With AI technology on Amazon Bedrock, we’re categorizing bank account activity to help the models be more accurate.” On AWS, this AI-powered feature, called Trustly Insight, is scalable and flexible to meet the company’s evolving needs.

Trustly has also gained better insights into its data that it’s using to innovate. The company adopted Amazon Quick, a service that businesses can use to answer questions and turn those answers into actions for AI-powered business intelligence insights. Now, Trustly provides data visualizations in convenient dashboards for its risk and fraud prevention teams. “At a glance, we have dashboards that provide better visibility into our metrics, the performance of our models, and our transaction approval rates,” says Padilha.

Amazon Quick has built-in AI agents for research and automation, so Trustly’s teams can act on their data directly from the dashboards. As a result, the company can measure the impact of its fraud detection models and continue to improve them. “Our employees like to work with cutting-edge technology and stay on top of the market,” says Padilha. “On AWS, we can run large, professional solutions in a way that’s resilient and that we can maintain in the future.”

As it developed its fraud prevention system, Trustly received support from AWS and discussed best practices for designing the solution with the AWS team. Trustly consistently trains its employees on the latest technology. “It’s important to have cutting-edge technologies on pace with our requirements,” says Padilha. “When you have a lot of opportunities and services, you can focus on solutions that actually help your business. We’ve been doing that successfully for years.”

On AWS, Trustly increased the resiliency of its infrastructure and enhanced fraud detection while reducing operational costs by 40 percent. The company reduced fraud attempts by 52 percent while maintaining 99.99% system availability.

Trustly has also reduced false positives by 43 percent using generative AI to identify emerging fraud patterns. And transaction risk assessments are 75 percent faster. In addition to internal monitoring tools, the company provides convenient dashboards for merchants to track activity. With faster fraud detection and better reporting, the company is helping merchants save money.

After the success of the fraud detection system in the US, Trustly is migrating additional infrastructure to AWS for its European business. And the company continues to innovate quickly. Previously, ML model deployments took weeks. Now, Trustly deploys models in hours, so it can respond to new threats rapidly.

“We’re always asking ourselves how we can use AI to further improve our product quality and the value we bring for clients and merchants,” says Padilha. “We’re looking to amplify our use of Amazon Bedrock in different areas, including customer support, merchant support, our knowledge base, and reporting.”