AI-Driven
Fraud Detection
for a Leading
US Delivery Platform

A crowdsourced delivery platform that enables urgent, same-day, and local next-day delivery services across the US. The platform provides a set of tailored applications for business senders, individual senders, and drivers. Boasting a network of 200,000 drivers, they’ve successfully expanded their reach to over 20,000 zip codes making them the largest same-day delivery provider in the US. Facing accelerated growth and the need to ensure optimal applications performance, they engaged Svitla Systems to address performance bottlenecks and enhance the satisfaction of their ever-expanding user base.

Svitla Systems partnered with the platform to address these challenges through a multi-faceted approach that included AI-driven fraud detection and performance optimization.

We initiated our work with a Proof of Concept for the AI-powered fraud detection system. The PoC was completed in just two months and demonstrated that new driver photos could be accurately compared to existing driver records to eliminate fraudulent registrations successfully. Encouraged by these results, the fraud detection system was deployed into production.

Simultaneously, the performance optimization work ensured the platform could scale without sacrificing speed or reliability, while new integrations expanded its service capabilities.

• Enhance the platform’s performance to support scalability and business expansion plans.
• Undertake technological upgrades to boost system capabilities and maintain a competitive edge.
• Revamp and optimize existing testing processes to streamline operations and cut costs.
• Expand the skillset of the existing QA team to address the lack of expertise in performance testing.
• Mitigate the shortage of experienced personnel well-versed in complex system maintenance to ensure seamless in-house QA operations.

• AI-Powered Fraud Detection: The platform experienced an increasing number of fraudulent driver registrations, where individuals created multiple accounts using different selfies. Svitla developed a Proof of Concept (PoC) leveraging facial recognition technology to address this issue. By extracting facial embeddings and comparing them with existing records using distance metrics like cosine similarity, the system identified duplicate accounts and prevented them from being created. This solution reduced the occurrence of phishing and fake accounts, improving the overall security of the platform’s operations.
• Performance Optimization: To handle the platform’s growing demand, Svitla Systems optimized the performance of their core applications. We migrated the testing environment from a slow and costly tool to Locust and CircleCI, which allowed for faster and more efficient testing. This process included creating new testing scenarios for warehouse and driver applications and revising existing ones to ensure optimal performance across the platform. As a result, the platform achieved better scalability and substantial cost savings​.
  
• Expanding Service Capabilities through Integration: The platform sought to extend its reach by integrating with major logistics providers. Svitla led key integrations with UPS and Delta Cargo, which allowed the platform to expand its service offerings to include air cargo and same-day deliveries. These integrations were instrumental in securing new business, such as handling Amazon Returns, and positioned the platform for continued growth​.