AI-Powered Fraud Defense: Company Slash Scammers by 25%, Speeds Up Detection 20x

ML-powered fraud detection platform enabled to process and handle 20 times more jobs per day and reduce scammer activity by no less than 25%, with almost 97% of all jobs handled automatically.

ML-powered fraud detection platform enabled to process and handle 20 times more jobs per day and reduce scammer activity by no less than 25%, with almost 97% of all jobs handled automatically.

Client:

Appen, a supplier of high-quality training data, aids organizations in constructing powerful AI systems on a large scale. Utilizing crowd workers, the company labels datasets crucial for training ML/DL models. Appen was founded in 1996 with a corporate headquarter in Chatswood, Australia. Their fraud detection system is used to ensure data quality by filtering out low-quality contributions. Appen sought to automate its manual fraud detection process to enhance the efficiency of identifying and preventing malicious activities on its platform.

Problem Statement:

Appen sought to automate its manual fraud detection process to enhance the efficiency of identifying and preventing malicious activities on its platform. Their goal was to scale the monitoring capacity to oversee a greater number of distributed workers each day, thus elevating the standard for identifying and preventing malicious activities on their platform. Additionally, Appen sought to expand their existing capability in training ML/DL models for their platform by automating data labeling, annotation, and categorization processes. Reducing the manual workload performed by distributed workers to boost the speed and efficiency of data processing and eliminate human error also presented a significant challenge for Appen. For Appen, effectively monitoring more than 50 jobs per day in a predominantly manual manner posed considerable challenges. The team contemplated hiring over 20 data analysts to cope with the platform’s expansion. The company required a solution to expand their fraud detection capabilities, enhance the productivity of their crowd workers, and entice potential enterprise clients.

Results:

• Enhanced satisfaction, safety, and trust among major enterprise clients.
• 25% decrease in scammer activity on the platform and a 5x reduction in false positives.
• 20 times increase in jobs monitored daily, with 97% handled automatically.
• Five times reduction in data-related disputes.
• Improved staff productivity and efficiency.
• Long-term reduction in operational expenses.

AI Solution:

Provectus engineered and constructed an automated fraud detection platform with human-in-the-loop functionality, employing TensorFlow and AWS products. To provide an automated fraud detection platform, Provectus established and implemented data pipelines to streamline the labeling, annotation, categorization, and moderation of data for the Appen team. Highly accurate ML/DL models, serving as the AI backbone of the fraud detection solution, were designed, trained, and fine-tuned. Furthermore, a user-centric web application was created for the Appen team to streamline the processing and oversight of data and notifications, enhancing efficiency.

All components of the fraud detection solution were automated and seamlessly integrated to ensure maximum efficiency and user-friendliness. Before initiating platform development, the Provectus team conducted thorough due diligence, researching the latest publications in crowdsourcing and fraud detection domains. The ML and DL elements, constituting the AI foundation of the solution, were created using TensorFlow. Deployment, serving, and monitoring of DL models were facilitated through Hydrosphere.io deployed on Amazon ECS. Java microservices, SQS events, and a React.js UI application were employed to automate data pipelines and ensure a comprehensive user workflow from end to end.

References:

1. Appen overhauls its fraud detection operations, efficiently detecting and preventing scam with a scalable, ML-powered fraud detection platform designed and built by Provectus. https://provectus.com/case-studies/appen-fraud-detection-platform/
2. Appen’s Transformation: From Manual to Automated Fraud Detection with AI/ML. https://www.iotone.com/case-study/appen-s-transformation-from-manual-to-automated-fraud-detection-with-ai-ml/c2650
3. Automated Fraud Detection Using AI. https://careers.provectus.com/tech-news/automated-fraud-detection-using-ai

Industry: Finance
Vendor: Provectus
Client: Appen
Publication Date: 2020
Keywords: fraud detection, data processing, AI/ML-driven platform, scam.