Human & machine: why a hybrid approach gets the best results

When choosing an identity verification solution, the best approach to verification plays a key part in the decision making equation. Historically, approaches have been either fully automated, fully manual, or mostly automated, with reliance on a human fallback. In this blog, we look at why the best approach is when there is a hybrid of tech and human, deeply intertwined. 

A hybrid approach delivers the best of both worlds

A hybrid approach combines human expertise and machine-learning power to deliver a fast, effective, scalable solution. This is the approach we use at Onfido. We use every weapon to fight fraud, providing the expertise of human capabilities, along with the speed and scalability of machines. That’s something other IDV solutions simply can’t do.

World-class team of document fraud experts + machine learning techniques

We have some of the world’s most experienced document experts. They guide our machine-learning efforts and lead the teams that provide manual verification. Experts continually monitor system performance and detect some of the more sophisticated fraud attacks.


And our platform was designed and built using deep learning, a machine-learning technique designed to mimic the way the human brain works. 

Our trained experts review any IDs which the machine system hasn’t seen before. They classify them to help develop our deep learning machine models. By using our human experts to train the automated models, we ensure a continuous process of improvement.

This is how we’re able to continually improve our fraud exposure rate (FER). The industry average of fraudulent applicants is 1.5%. So out of 10,000 applicants, 150 will be fraudulent. During the first half of 2019, we delivered an FER of 0.0195% for our clients. This means that out of the 150 fraudulent applicants, we’d catch 148 of them, miss two, and flag an additional 150 for manual review. In total, our AI will flag and categorize a total of 298. Our human experts then confirm the fraudulent 148 and categorize the additional 150 as safe.

Crucially, our human experts will then train the models, based on the characteristics and patterns of the 148 caught and 2 missed. This drives continuous improvement.

Frost & Sullivan applauded our innovation in a recent report, stating they are: “impressed with Onfido for optimally addressing customer as well as industry  requirements by developing such an advanced AI-based verification and authentication solution.” Read the full report on Frost & Sullivan’s best practices.

Ultimately, a hybrid approach is the best approach to identity verification. It combines the expertise of human manual checks, with the efficiency of automated machine learning techniques. It’s a scalable, effective process, proven to balance both the speed and security you’ll want, while catching fraud.


For some of the latest insights from our fraud experts, why not take a look at our Global Fraud Index.

Previous Flipbook
Are you using the best approach to verify customer identities?
Are you using the best approach to verify customer identities?

Learn how Onfido’s identity verification technology can help you verify your users’ identities with confide...

Next Article
Synthetic identity fraud: definition and explanation
Synthetic identity fraud: definition and explanation

In this blog, find out how synthetic identity fraud differs from traditional fraud methods, how it impacts ...