Digital identity - Onboarding with OCR and facial biometrics
Integrating ICAR's identity verification platform, now Mitek, into onboarding apps, and the UX Match that placed second at the first CIAB FEBRABAN Hackathon.
A note before we start
This case is different from the others in the portfolio, and it is worth saying so up front.
ICAR, the Spanish identity verification company acquired by Mitek Systems in 2017, appears here as a technology supplier and partner, not as a client. X-Apps is the one who integrated their platform into product, took the solution to market in Brazil and built a product of its own on top of it.
The problem: opening an account without visiting a branch
Digital onboarding is the moment an institution has to be sure the person on the other side of the screen is who they claim to be, using only a phone camera.
That means solving three things at once, in seconds: reading the document without typing, verifying the document is genuine, and confirming the person holding the phone is the same one in the document photo.
The integration, in 2016
The app captures three images: the front of the document, the back of the document and a selfie. Each capture carries a device information block, used in face detection.
Those images are encoded and assembled into a request sent to the verification service, which returns automatic document reading, the holder's photo, the signature, the fingerprint and the result of the facial comparison between the selfie and the document photo.
The integration work had two technical points that decided the quality of the result:
Making the calls asynchronous. The original flow was synchronous, and that degraded the experience for whoever was waiting on the other side. Refactoring to asynchronous calls is what made the response time tolerable on a phone.
Handling the path where recognition fails. Not every document is recognized automatically. When that happens, verification moves on to manual operator processing. That exception flow is not a detail: it determines whether onboarding works for everyone or only for the easy cases.
Development was completed in mid-2016, with the platform supplier itself confirming it worked.
When the problem is outside your code
One episode is worth recording, because it is common in integration work and rarely told.
In the manual fallback flow, the operator received only the front image of the document, even though the app was correctly sending front and back. The defect sat on the platform side, outside X-Apps' reach, and still it blocked product validation.
There was also a case where the response came back as document not recognized simply because no operator was available at that moment for manual processing, which forced test windows to be agreed in advance across all three parties.
That is what integrating a third party platform means: part of the work is isolating what is yours from what is not, proving which is which, and still moving when the fix depends on another company in another time zone.
UX Match: two days, second place out of 400
In 2017, X-Apps took the knowledge built in that integration to the first CIAB FEBRABAN Hackathon, at Latin America's largest financial technology fair.
The product built was UX Match, an automated onboarding and document authentication tool on top of ICAR's reading technology, developed in two days, the full length of the hackathon.
The result was second place out of 400 participants, with the final held at the fair itself.
UX Match was shown at the stand in Android and iOS versions, and interest was captured on the spot: more than twenty companies signed up to see the solution, among them financial technology firms and lending fintechs.
The result led to a joint press release with ICAR, and their chief executive gave a formal testimonial about the work, which went into X-Apps' own material.
Stack and practices
- Native Android and iOS apps. Integration with ICAR's cloud identity verification platform, through a structured request carrying the encoded images. On-device document reading through the platform's mobile SDK. Facial comparison between selfie and document photo, reading the similarity score returned by the service. Asynchronous calls, adopted on the supplier's own documented technical recommendation. * Explicit handling of the semi-automatic flow, for when automatic recognition does not complete.
What carries over today
Digital onboarding became a market standard, but the problems are the same: poor framing, unrecognized documents, inconclusive facial comparison, and what to do with the user while that happens.
X-Apps has worked on that flow since 2016, integrating third party platforms and taking on the part that is usually underestimated, which is the exception path.
Integrating identity verification into your product?
The hard decision is not which recognition vendor to use. It is how your product behaves when recognition fails, and how much of that is visible to the person on the other side of the screen. X-Apps integrates the platform and designs that flow with you.