Remote identity verification & KYC

Open an account in under two minutes, with no branch visit and flawless compliance.

+45%

Onboarding completion rate

<2 min

Verification time

98.5%

Document fraud detection

-70%

Cost per file reduction

The challenge

Across West Africa, account opening is still throttled by manual KYC: a branch visit, photocopied documents, repeated data entry and approximate visual checks. The result is digital onboarding drop-off rates above 60% and a per-file processing cost that erodes the margin on low-deposit accounts.

Regional ID documents are heterogeneous — Mauritanian national ID, ECOWAS card, passport, driving licence — often bilingual Arabic/French with uneven print quality. On top of this sits mounting regulatory pressure (BCEAO, GABAC, anti-money-laundering) that demands a complete audit trail and reliable detection of document forgery and impersonation.

Our approach

ADST deploys a fully remote KYC journey that combines three real-time AI building blocks. First, intelligent document reading: document-type detection, deskewing, bilingual Arabic/French OCR of MRZ and visual fields, then consistency checks (dates, MRZ checksum, number formats).

Next comes passive and active liveness detection to defeat photo, replay and mask attacks, followed by face matching between the selfie and the ID photo with a calibrated similarity threshold. Every decision carries a score and an explicit reason, automatically routed to human review whenever there is doubt.

The whole flow plugs into your core banking or telecom platform via API, with end-to-end encryption, hosting compliant with local data-residency requirements, and a time-stamped audit trail ready for regulatory inspection.

Architecture

Document detection & classification: object-detection models + PaddleOCR fine-tuned on Arabic/French
Liveness: anti-spoofing CNN (texture, depth, micro-movement analysis) certified PAD level 2
Face matching: ArcFace embeddings compared by cosine distance with a calibrated threshold
Orchestration: real-time API (<2 s), human-review queue, immutable audit log

Models used

Anti-spoofing CNN (Liveness/PAD)Face recognition (ArcFace embeddings)OCR (PaddleOCR fine-tuned AR/FR)Object detection (document detection)Vision Transformer (document classification)

Data required

Front/back ID photos (national ID, passport, ECOWAS card)Selfie and short video stream for livenessMachine-readable zone (MRZ) of documentsPer-country document format referencesLabelled document-fraud history

Return on investment

By automating 90% of checks, a retail bank cuts its acquisition cost while tripling the number of accounts opened per month.

Relevant sectors

BankingTelecomFintech

Related services

Intelligence ArtificielleFintechConseil IA

A similar project?

Let's discuss how AI can transform your organization.

Get in touch