Video-based Customer Identification Process began as a pandemic accommodation and has become the standard remote onboarding route for Indian financial institutions. It is not a video call with a screenshot. The RBI has progressively tightened what a compliant session must contain, and the requirements shape the architecture far more than the video streaming does.
Start with who conducts it, because this is the constraint that surprises product teams most. The official performing the verification must be on the regulated entity's own payroll rather than an outsourced agent. That single rule determines your capacity model: throughput is bounded by trained internal staff, not by servers, and any growth plan that assumes onboarding scales like software is wrong. It also means the queueing, scheduling and shift management around the session is a real part of the product rather than an afterthought.
Liveness is required to be active rather than passive. The system must prompt the customer to perform randomised actions — blinking, turning the head, reading a number displayed on screen — so that a recorded video or a still photograph cannot pass. The randomisation matters: a fixed sequence is a sequence an attacker can prepare for. With synthetic video now cheap to produce, the direction of travel is toward stronger challenge-response, and building the prompt logic so new challenge types can be added is worth doing at the start.
Location and network origin are checked, not merely logged. The session must capture live GPS coordinates and verify that the IP address originates from an Indian internet service provider. Both are things a customer's browser or app must be able to supply, which means the permission prompts, the fallback when a customer declines location access, and the handling of VPN users all need designing as product decisions rather than left as error cases.
Storage and audit are where most of the engineering effort actually lands. The session is end-to-end encrypted, with AES-256 the expected standard, and the recording and associated data must be stored in India with cross-border transfer restricted. Auditors will later review video recordings, chat transcripts and decision audit trails, which means every step — what the officer saw, what they asked, what document was captured, what OCR extracted, what decision was recorded and by whom — needs to be reconstructable years afterwards. Design the audit record first and the user interface around it.
Document verification runs alongside: officially valid documents captured on camera, read by OCR, and checked against the issuing database rather than accepted at face value. The failure mode to design for is not fraud but ordinary life — poor light, a glare on a laminated card, a customer holding it at an angle, a network drop halfway through. A session that must be restarted from the beginning because of a dropped connection is the single largest cause of abandonment, and resumability is worth more to your conversion rate than any amount of interface polish.
