e-Safer - Document capture with OCR and biometrics
Three generations of document capture for e-Safer, from the first iOS app in 2016 to the Android app integrated with Kofax TotalAgility.
Turning a photo of a document into trustworthy data
e-Safer works in information security and fraud prevention. Document capture is where those two meet: someone photographs a document with their phone, and out of that picture has to come data structured and reliable enough to support a decision.
The partnership with X-Apps spanned three generations of product, between 2016 and 2019, each answering a different question.
The constant challenge
- The photo is taken by a non-specialist, with whatever device they have, in whatever light there is. Automatic recognition fails in a share of cases, and the flow has to keep working when it does. Text recognition and liveness are different steps, and treating them as one stalls the entire process.
First generation: separating what was glued together
The 2016 iOS app had concrete usage problems: it froze on the second photo and miscounted photos on some devices.
The work went beyond fixing the defect. There were three structural changes:
- Separating text recognition from biometrics. They were coupled, which meant a failure in either took down the other. Made independent, each step could fail on its own. Making the request asynchronous, to match how the certification service processed submissions, instead of leaving the user standing in front of the screen. Rebuilding the interface and translating the app, which was in another language.
Versions shipped as successive builds over weeks, rather than as one delivery at the end.
Second generation: capture integrated with Kofax
The second generation grew from a product question: what exactly did e-Safer need to have, compared to the demo app of the capture platform it used?
The answer became a formal study, screen by screen, mapping the reference app and proposing the equivalent for e-Safer's operation: sign-in, document capture, pending documents, assistant, help and about.
Out of that study came an Android app that captures the documents of the credit process, driver's licence, vehicle document and credit note, and sends them in the background to the capture platform, with a dedicated data model per document type.
Two design decisions are worth recording:
- No server configuration screen inside the app. All user and environment management stayed in e-Safer's back end. The app in the field exposes no server address and does not ask the user to pick an environment. * Background submission, so capture was not held hostage by the platform's response time.
The study also put in writing what was not clear, including reference app behaviours that needed confirmation from the vendor. Documenting the doubt stops it from becoming an assumption in the code.
Third generation: the full platform
The third front was the largest: a vehicle financing platform, with a mobile app, a web portal for the lenders, an admin panel and an API.
In the app. Registration and sign-in, vehicle make and model, driver's licence photo with text recognition, vehicle document photo with text recognition, photos of the vehicle from every angle, and the formal consents of the process.
In the API. Token authentication, access control by role and permission, customer and per-lender user registration, email delivery, push notifications, integration with the document recognition service, cloud image storage, and the proposal and counter-proposal flow between parties.
One field detail changed the camera design: driver's licence capture gained a native framing mask, integrated with the system camera. Guiding the framing at capture time is cheaper than correcting afterwards, because it avoids the resubmission.
Stack and practices
- First generation in Ionic, with builds distributed through a testing platform. Second generation on Android, integrated by SDK to the document capture platform, calling the platform service with port switching according to secure channel use. Third generation in React Native on the app, including a camera with a native mask, and React with Redux on the web portal and admin panel. Token authentication and access control by role and permission. External document recognition service wrapped in a dedicated layer, so the vendor could change without spreading modifications across the system. Cloud image storage. Push notifications and text message validation.
Each wave started with a written survey, with effort estimated item by item, before development.
Need to turn documents into data?
Document capture looks like a camera problem and is actually a flow problem: what happens when recognition fails, who confirms, where configuration lives, how long the user waits. X-Apps builds that entire flow, from the framing on screen to the structured data arriving on the other side.