KYC Onboarding at a UK Challenger Bank — From 5 Days to 4 Hours, Dropout from 42% to 11%

We deployed OnboardX (Ox) as the unifying onboarding pipeline, with DocuMage as the document-intelligence engine underneath, FaceMatch (Fm) for the liveness and biometric step, and Doc Pair Matcher (Dp) for the cross-document consistency checks. The four accelerators are deeply integrated as a single straight-through flow.

Phase one was the customer-journey rebuild. We worked with the bank's product team to design a single, unified onboarding journey that absorbed the previously discrete steps — application data capture, document submission, liveness check, sanctions and PEP screening, source-of-funds questionnaire, and account activation — into a continuous flow with clear progress indicators and minimum friction.

Phase two was the verification pipeline. OnboardX orchestrates: ID document capture (passport, driving licence, national ID — automatic detection of document type), DocuMage extraction of all visible fields including the MRZ on passports, FaceMatch liveness check (the customer is asked to perform a randomised series of head movements and the system performs an active anti-spoofing check), Doc Pair Matcher consistency check (does the name on the document match the name on the application; does the address match the proof-of-address; does the photo on the document match the live selfie), and sanctions / PEP screening via the bank's existing third-party screening provider.

Phase three was the exception-handling layer. Cases that pass every check on the first pass go straight through to account activation — typically within four hours of the customer starting the application. Cases that fail any check are routed to the KYC operations team with the specific failure reason, the relevant document images, and the system's confidence scores — meaning the human reviewer can resolve most exceptions in under five minutes versus the previous twenty.

OnboardX runs in the bank's AWS environment in the London region, with all PII processing happening inside a dedicated VPC. The platform is built as a microservices stack — orchestration, document service, biometric service, screening service, and exception-management service all running as independently scalable services on EKS.

Document intelligence uses DocuMage with the UK-document specific extractors enabled — passport MRZ parsing, UK driving licence layout (front and back), UK utility bill template variations, HMRC documents, Companies House extracts (for business onboarding). The MRZ parser is critical: it provides an independent check on the human-readable zone of the passport and detects manipulation of the visual fields.

The biometric layer uses FaceMatch with an active liveness check — the customer is asked to follow a randomised sequence of head movements while the system captures multiple frames. The frames are processed through a deepfake-detection model and a 3D facial-structure validator before being passed to the 1:1 matching engine that compares the live frames to the photo on the submitted ID. Confidence scoring and the underlying frame data are stored for audit.

Sanctions and PEP screening uses the bank's existing third-party data provider — we integrated rather than replaced — and feeds the screening results into OnboardX's adjudication layer. PEP scoring is risk-weighted; not every PEP match triggers manual review, only those where the position, jurisdiction and risk profile combine above a defined threshold.

The entire pipeline is built to be FCA-auditable. Every decision (auto-approve, manual review, decline) is logged with the underlying data, model versions, confidence scores and reasoning. The bank's internal audit team has direct access to the audit log and the model-card documentation for every ML component in the pipeline.