Marco Ridder is Director of Global Operations for Mitek Systems, digital identity verification, and mobile deposit solution provider. One of his responsibilities is heading up a team of about 130 document authentication experts based in Amsterdam, Barcelona, San Diego, and Vietnam. Marco talked with Mike Sasaki, leader of the Mitek global customer success team, about how human expertise fits into today’s machine-learning-based IDV.
Let me start with this high-level question: What is the role human document experts play in Mitek identity verification?
Marco: Sure, this is an option Mitek provides for clients to add another layer of IDV assurance.
They can choose to have the document images submitted by end-users, as well as the data
our computer vision algorithms and other AI technology automatically extract from them, passed to a human expert for further examination.
It’s interesting that some companies opt for human review. Speed is a top concern for most Mitek clients—doesn’t human review slow things down?
Yes, of course it does, but only to a small extent. In fact, one of the things I’m really proud of is that my team has recently pushed response time down from five minutes to one minute in many cases.
The reason we can return high-assurance answers so quickly is that the AI assists human review by classifying the document image as to what type of ID it is, and extracts data from it. Agents don’t have to key in anything. They can immediately start examining the image and comparing it to the template corresponding to that particular government-issued document.
That said, I think for every Mitek client it’s different. There’s a “sweet spot” we help them find where they’re hitting the right balance of high speed and high assurance for them.
What are some specific use cases where companies value this additional layer of assurance?
The most common one is to lift conversion rates. There are situations where auto IDV isn’t able to recognize the ID type. This could be because our auto coverage doesn’t yet have a template for that particular ID. Or it could be due to bad image quality— the user was in a very dark room or blocked a corner of the ID with a finger. The company could reject the applicant or ask for the ID to be resubmitted, but there might be some drop-off because of that extra friction in the process. So another option is to pass the image and data to a human expert. They’ll be able to quickly determine what kind of ID it is and if it’s real and hasn’t been altered. And they’ll also provide their results back to our machine learning engines so auto IDV will recognize that type of document next time.
Another use case is where there is a requirement for a human expert to be involved because of regulations or the client’s preference. Some companies want to know that humans, not just algorithms, are looking at incoming documents. So, for example, one of our clients is a global platform for short-term property rental. The safety of their users is of even greater concern to them than speed. And in other situations, companies will refer only a fraction of submitted IDs for human review, such as those machine learning identifies as highest risk or where the auto result is borderline.
How is it that humans are able to make such accurate judgments about ID validity when AI can’t?
Experience. I’m happy to say our team includes people with as much as 30 years of experience examining IDs in the off-line world, including for border patrols and airports. A lot of these people have examined tens of thousands, if not hundreds of thousands, of IDs over their professional lives. Their brains have “libraries” filled with mental templates of what legitimate IDs should look like. So as soon as there’s something off, they spot it immediately. They know something’s going on there.
These are our top experts, but if you look across the entire team, we have many, many individuals who were trained by those top people and have now racked up a lot of experience themselves after being with us for five to eight years. And we really are a global team: Our people in Europe, the US, and Asia know each other, talk to each other and share their knowledge.