How Onfido Optimized AI Fraud Detection
“Appen was paramount in the development of Onfido and has become a critical component of our operations. We are thrilled to have found a reliable partner in Appen.”
– Francois Jehl, Director of Data and Analytics, Onfido.
Introduction: Enhancing Fraud Detection with AI and Machine Learning
In today’s increasingly digital world, AI fraud detection plays a critical role in ensuring security and preventing identity theft. Onfido, a global leader in AI identity verification, sought to enhance their anti-fraud capabilities by leveraging multimodal AI models. As online fraud continues to grow, Onfido needed to stay ahead of sophisticated fraud tactics, such as 3D silicone masks and display attacks. The company’s Real Identity Platform, which verifies users through biometric and document checks, required precise, secure data labeling to train their AI security models for identity verification.
To meet these challenges, Onfido turned to Appen for a solution. With Appen’s custom on-premise data annotation tools, Onfido improved the accuracy of its AI fraud detection systems, optimizing their machine learning models with the precise data needed to fight increasingly complex identity fraud.
About Onfido
Onfido is a global leader in AI identity verification, enabling businesses to securely verify users online. Their Real Identity Platform combines biometric and document verification with built-in AI fraud detection, allowing users to prove their identities remotely. Onfido’s solutions cater to businesses that require secure, user-friendly AI security processes, such as banks, fintech companies, and e-commerce platforms. By incorporating cutting-edge technologies, Onfido helps businesses reduce fraud, improve customer acquisition, and enhance overall security.
Goals: Achieving Secure, Scalable, and Accurate AI-Driven Solutions
Onfido’s primary goal was to enhance the security of their machine learning models by using high-quality labeled data to train their AI security systems for identity verification. Specific goals included:
- AI Fraud Detection: Onfido needed to stay ahead of emerging fraud techniques by continuously improving their anti-fraud models using accurate labeled data.
- Security: Given the sensitive nature of biometric data, Onfido needed an on-premise labeling tool that ensured compliance with data privacy regulations.
- Scalability: Onfido required a solution capable of processing large volumes of data while maintaining high accuracy across diverse data types, including images, videos, and documents.
- Customization: The company needed a partner that could provide a flexible, customizable labeling tool to handle the specific requirements of their AI identity verification models.
Meeting these goals was crucial for helping Onfido stay ahead of increasingly sophisticated fraud tactics and improve the effectiveness of their AI security systems.
Challenge: Secure and Customizable AI Data Labeling
Onfido faced several challenges in its search for the right data labeling solution. As they work with highly sensitive biometric data, most of which contains personally identifiable information (PII), the company needed an on-premise solution to ensure compliance with AI security protocols. Each time Onfido shared data with a third party, they were required to obtain special permissions, adding another layer of complexity.
In addition to security concerns, Onfido needed a flexible labeling tool that could handle a wide variety of data types, from images and videos to documents. The company’s AI fraud detection systems required precise annotations of diverse data formats to train their machine learning models effectively. With increasingly sophisticated fraud techniques emerging, ensuring accuracy and flexibility in data labeling was essential.
Solution: A Custom On-Premise Labeling Tool from Appen
To address these challenges, Onfido partnered with Appen to implement a fully customizable on-premise data labeling tool. Appen’s AI Data Platform (ADAP) provided the flexibility, security, and accuracy that Onfido required to meet its goals.
Collaborative Approach
Appen collaborated closely with Onfido for a custom data labeling solution that met their specific requirements. This involved weekly meetings where both teams collaborated to fine-tune the solution. Onfido needed the tool to handle a variety of data formats, including complex biometric and video data. Appen’s solution offered flexible data annotation capabilities across images, documents, and videos, providing Onfido with the exact functionality needed to label their data accurately and securely.
On-Premise Deployment for Enhanced Security
Given the sensitive nature of the data involved, Onfido needed to maintain full control over the labeling process. Appen provided an on-premise deployment option that allowed Onfido to label data securely within their own infrastructure, eliminating concerns about third-party data access and ensuring compliance with AI security and privacy regulations.
Scalable and Collaborative
Appen’s tool also supported scalability, enabling Onfido to process large volumes of data without sacrificing accuracy. Onfido leveraged Appen’s data collection services to source identity data tailored to specific demographic requirements. The teams worked in tandem, with continuous feedback and weekly status meetings, ensuring that the tool was delivering precise data that could train Onfido’s machine learning models for AI identity verification effectively.
Result: Improved AI Fraud Detection with 10x Performance
“We appreciate all the support, from the customized project plan to the weekly status meetings. Appen is always there for us.”
– Francois Jehl, Director of Data and Analytics, Onfido
The partnership between Onfido and Appen led to the successful deployment of a data labeling tool that significantly improved Onfido’s AI fraud detection capabilities. The tool played a pivotal role in training the machine learning models that power Onfido’s product, Motion, which uses simple user head movements for identity verification.
Enhancing AI Anti-Fraud Technology
To train Motion’s anti-fraud models, Onfido launched a dedicated lab that generated hundreds of thousands of fraud videos using masks, mannequins, and videos of videos. These fraud videos were labeled using Appen’s tool, enabling Onfido to improve the performance of their AI fraud detection models by 10x. Appen’s solution provided the precise data annotation needed to ensure that Onfido’s models could accurately detect even the most sophisticated fraud techniques.
Continuous Model Retraining with Human-in-the-Loop
Onfido also built a continuous human-in-the-loop retraining pipeline using Appen’s API. This system automatically generated annotation jobs from production data, allowed for manual labeling, and fed the labeled data back into Onfido’s machine learning models for retraining. This continuous improvement loop enabled Onfido to maintain the high performance of their AI security models over time.
A Reliable Partnership
The partnership between Onfido and Appen highlights the importance of flexible, secure, and scalable data labeling solutions in building effective machine learning models for AI identity verification and fraud detection. By working closely with Appen, Onfido was able to enhance their AI security systems and deliver more reliable identity verification services to their customers.
Conclusion
Onfido’s collaboration with Appen was instrumental in developing the AI fraud detection models that power their Real Identity Platform. Appen’s expert support enabled Onfido to meet their security, scalability, and accuracy requirements, improving their machine learning models by 10x. Through close collaboration and continuous support, Appen helped Onfido stay ahead of increasingly sophisticated fraud techniques and deliver enhanced AI identity verification services to businesses worldwide.