AI-Powered Video KYC for Banking
Rebuilding Digital Onboarding & Revenue During COVID
Domain: Fintech · AI/ML · Digital Identity · Banking
Executive Summary
Role: Senior Product Designer
Scope: End-to-end product strategy, AI interaction model, customer mobile and operator dashboard
Team: Product Managers, Engineering, Compliance, Legal, Business, and Operations stakeholders
Impact: 15% credit card conversion uplift · 20% faster verification · 30% less operator handling time
The Problem Nobody had a Playbook for
COVID closed bank branches overnight. Credit card onboarding stopped completely. The ask was to move KYC to video, but the real challenge was harder: how do you design AI-assisted verification in a regulated, high-stakes environment without losing human accountability?
No industry precedent. Built from scratch under compliance constraints and zero tolerance for fraud risk.
What I Owned
I was the sole product designer across two very different surfaces, customer mobile flows and relationship officer web dashboard, partnering directly with Product, Engineering, Compliance, and Legal on every decision.
The Call that Defined the System
There was pressure to automate verification end-to-end. I pushed back.
Research showed officers did not trust black-box AI decisions. They needed to understand why a case was flagged. I made the call to design AI as a support layer, not a replacement. Every output, document match, liveness score, location signal, surfaced as a confidence indicator, not a verdict. Officers retained final decision authority on every case.
That single principle defined the architecture of both surfaces.
What I Designed
Customer mobile: OCR-based PAN and Aadhaar verification, liveness detection, seamless video call handoff. Designed to feel fast, safe, and human despite being AI-driven.
Operator dashboard: Built around exception handling, not routine processing. AI handled checks silently and only surfaced cases needing human attention, with confidence scores, session access, and fast paths for approval or escalation.
Every AI failure state was designed before any success state. Liveness failure, unreadable documents, low confidence thresholds, all mapped and handled explicitly. This is where trust is built.
Key Decisions
Confidence scores over binary outcomes gave officers calibrated signals they could act on with judgment, not just verdicts to rubber stamp.
Progressive disclosure over dense compliance screens surfaced critical signals first, detail on demand. Reduced errors and processing time.
Held the line on regulatory steps when there was pressure to shorten the customer journey. Designed required steps to feel purposeful, not bureaucratic.
Outcome
15% increase in monthly credit card conversion 20% reduction in verification completion time 30% reduction in operator case handling time 25% decrease in customer complaints on verification delays 15% reduction in onboarding drop-off
The platform moved the bank from zero digital onboarding capability to a scalable, compliant, AI-assisted operation during the most disruptive period in recent banking history.