Figure Eight: Fraud Detection Platform
Figure Eight efficiently detects and prevents scam with a scalable, ML-powered fraud detection platform
Satisfaction, safety, and trust of major enterprise customers achieved
25% reduction in scammers on the platform
12x more jobs monitored per day
5x fewer tainted judgments about data
Reduced operational costs
Figure Eight aimed to automate its manual fraud detection mechanism in order to:
- Scale the amount of distributed workers it can monitor per day to more efficiently detect and prevent malicious activity on the platform
- Train the platform’s machine learning and deep learning models more efficiently due to automated data labeling, annotation, and categorization
- Cut the amount of manual redundant work and eliminate human error
Figure Eight used crowd workers to label data sets to train ML and DL models. SQL and Python scripts were used to manually activate the fraud detection mechanism. That approach was complex, inefficient, and not scalable. The number of manual tasks resulted in human error and did not allow for efficient coordination of distributed workers.
Figure Eight could not manually monitor more than 50 jobs per day and was pressed to hire 20+ data analysts to be able to keep up with the platform’s growth. The existing solution was not able to ensure the proper quality level needed to win the enterprise clients.
Figure Eight approached Provectus to design and build an automated ML-powered fraud detection platform, with a scalable SaaS architecture and a GUI for optimal user experience.
To deliver an automated, end-to-end fraud detection platform with human-in-the-loop, Provectus had to:
- Design and build data pipelines to ensure efficient data labeling, data annotation, data categorization, and content moderation
- Design, train, and optimize machine learning and deep learning models
- Develop a web application for the fraud detection platform
- Ensure automation of these components
As a first step, the Provectus team researched the latest academic papers in crowdsourcing and fraud detection to design an effective scammer and fraud detection platform.
TensorFlow was used to develop the machine learning and deep learning components of the platform. The DL models were deployed, served, and monitored with Hydrosphere.io installed on top of Amazon ECS.
To automate data pipelines and ensure an end-to-end user workflow, Java microservices, SQS events, and React.js UI application were used.
Prometheus and Grafana were used for continuous monitoring. They provided complete visibility into a system’s performance to engineers, fraud analysts, and business stakeholders.
Figure Eight’s ML-powered fraud detection platform allowed to manage 12x more jobs per day, deliver 5x fewer tainted judgments, and weed out 25% more scammers than its manual fraud detection mechanism.
The fraud detection platform improved productivity of Figure Eight’s employees. This allowed the company to forego hiring 20+ data analysts and to considerably reduce operational costs.
The quality of Figure Eight’s training data and service improved, which satisfied existing and attracted new enterprise clients to speed up the company’s expansion.