Over the last year, Jumio has improved our verification response times by more than 33% thanks to AI, machine learning and OCR improvements. Jumio has been refining its OCR technology in order to better automate the data extraction process and apply that automation to an expanded set of ID document types that used to be manual. Our AI has enabled automatic recognition of the ID country and types to streamline the flow. We also automated the image quality check (which assesses the readability of an ID document) — a process that used to take 15 seconds now takes less than a second.
Thanks to Jumio AI Labs, we’ve also refined our machine learning models by fine-tuning them on millions of internally tagged ID images. This tagging process, performed using our compliant tagging infrastructure and in-house experts, provides a critical feedback loop that helps train the ML algorithms by correctly flagging legitimate and fraudulent IDs. This type of continuous model fine-tuning improves the accuracy of our ML algorithms which, in turn, leads to fewer mistakes and more consistent decisioning.
About 90% of the identity verification process has been automated through AI and machine learning. In fact, the only transactions that are currently routed to verification agents are those where the image quality of the ID document or selfie is poor (e.g., shot in bad lighting, too blurry or too much glare) and therefore requires human review to make a verification judgment. Most of our regulated customers actually want this extra human review, even though it adds a few more seconds of verification time, as it results in higher conversion rates and a better customer experience because they’re not automatically rejected by AI algorithms that simply cannot read the images.
One of the added benefits of the additional human review implemented on images with poor quality is that Jumio can provide feedback to help users course-correct and thus enable higher onboarding rates. Most competitive solutions do not return specific, actionable rejection codes — if the image is too blurry, they are simply rejected outright. Jumio provides more than 15 specific rejection codes which are being used to help the end user complete their online journey, giving them a second chance to recapture the image and effectively course-correct. We are seeing 15% increased conversion rates resulting from this improvement.
Deepfakes, bots and advanced spoofing attacks have made certified liveness detection a must-have feature in any biometric-based verification solution. Unfortunately, gesture-based gimmicks, like asking a user to blink or speak a random passcode, add friction to the experience, and are also easily fooled by basic spoofing techniques. In fact, Jumio was the first identity verification solution to leverage certified liveness detection — that is, biometric-based technology that passed Level-2 certification for ISO 30107-3 iBeta Presentation Attack Detection (PAD) testing.
Jumio is one of a handful of solution providers that offers an omnichannel experience. This means end users can verify themselves via a mobile SDK, webcam, or mobile web experience.
For example, Jumio provides the capability to complete the whole identity verification journey using a webcam. However, customers also have the option to enable a cross-channel journey. This is especially important in certain geographies (e.g., Southeast Asia), where end users are capturing images of ID documents and selfies with older-model webcams which offer poor image resolution (compared to modern smartphones). Jumio enables webcam users to continue the user journey on their mobile device simply by sending an SMS message or a QR code. After taking the photos on their mobile device, people can complete the journey on their computer. This option has increased online conversion rates of our webcam users by more than 10%.