Onboarding that ends in the same session it starts

The client is a consumer fintech operating under central bank supervision in the Gulf, with a payments and money-management app acquired almost entirely through app stores. The product was mobile-first. Onboarding was not. Signup collected a form, then handed the applicant to a back office: emailed requests for ID documents, an analyst comparing a selfie to a scan by eye, sanctions checks run by hand against exported lists, and a final approval recorded in a spreadsheet.

The process broke where growth met compliance. Every acquisition campaign filled the review queue faster than analysts could clear it. Applicants who signed up in minutes waited days for approval, and many never returned. The compliance team had its own worries. Decisions lived in inboxes, name screening depended on which analyst ran it, and reconstructing why an account had been approved meant archaeology across systems.

Kentro was engaged to rebuild onboarding as a product: a flow that completes in a single session on the phone, automates verification and screening end to end, escalates only the cases that need human judgment, and leaves behind a record a regulator can walk through.

Growth exposed an onboarding process designed around the back office rather than the applicant. Each stage ran on manual effort, and none of it left a defensible record.

• Manual review as the default path.: Every applicant landed in the compliance queue regardless of risk. Analysts spent their days clearing obviously legitimate cases while the genuinely risky ones waited in the same line.
• Identity checks built for a desk, not a phone.: Documents arrived over email and web upload forms. Blurry photos, missing document sides, and abandoned sessions were routine, and nothing in the process could catch a photo of a photo.
• Screening that could not keep pace.: Sanctions and PEP checks ran by hand against exported list files, with no handling for Arabic-to-Latin transliteration. The same person spelled differently across documents produced inconsistent results depending on who ran the check.
• No defensible record of decisions.: Approvals lived in email threads and spreadsheets. Demonstrating why an account had been opened meant piecing the story together from fragments across systems.

Kentro built onboarding as an event-driven verification pipeline behind the client's app. A native capture SDK runs on the device. Stateless verification services for documents, biometrics, and screening run on Kubernetes and publish step results to Kafka topics. A decision engine consumes those results and either completes the account or routes it to a case queue. Session state is resumable and every step is idempotent, so a dropped connection resumes the journey instead of restarting it.

• On-device capture, server-side verification.: The SDK guides document capture, parses the machine readable zone under the ICAO travel document standard, and reads the chip over NFC where the document carries it, including Emirates ID and biometric passports. Server-side checks validate the chip signature and cross-check extracted fields against the visual zone.
• Passive liveness and face match.: Liveness detection certified against the ISO presentation attack detection standard runs during a short selfie capture, with no head-turning routine for applicants to abandon. The live face is matched server-side against the document portrait, and both artefacts enter the evidence pack.
• Transliteration-aware sanctions and PEP screening.: Applicants are screened against the UN consolidated list, local sanctions lists, and commercial PEP data through a fuzzy-matching service that normalizes Arabic and Latin name forms before comparison. Matches come back as scored candidates with the matched fields shown, not binary hits.
• Risk-based decisioning with a real exception path.: A rules-plus-scoring engine applies the FATF risk-based approach: clean applications pass straight through, borderline signals trigger step-up checks in the same session, and only genuine screening hits or document anomalies land in a maker-checker review queue with the evidence pack attached.
• An audit trail built as an event log.: Every pipeline step writes to an append-only, hash-chained event store. The full history of a decision (what was captured, what was checked, what the engine concluded, who reviewed it) can be replayed on demand rather than reconstructed.

The measure of the build is operational: what the applicant experiences, what the compliance team stopped doing by hand, and what the regulator can now see.

• Onboarding finishes in the session it starts.: Clean applicants open a working account before they put the phone down. The gap between signup and approval, where most drop-off used to happen, no longer exists on the straight-through path.
• Compliance reviews exceptions, not everyone.: Analysts stopped approving obviously clean cases by hand. The queue now holds only escalations, each arriving with documents, biometrics, and screening results already assembled.
• Screening became consistent and explainable.: Name matching behaves the same on every application regardless of transliteration, and every clearance or hit carries the scoring behind it. The compliance team defends decisions with evidence instead of memory.
• Regulator questions start from the log.: When supervisors ask why an account was opened, the team exports the event history for that application. Reconstruction from inboxes and spreadsheets ended with the old process.