A human Video KYC agent can run one session at a time. A Video KYC Moderator AI runs 1,000. Not by replacing the human verification decisions that the RBI's V-CIP framework requires — but by handling every element of the session that does not require a human: the pre-session checks, the real-time document verification, the liveness confirmation, the geolocation validation, and the session recording and logging. The human agent verifies. The AI conducts.
The V-CIP bottleneck — and why it has always been at the agent layer
Since the RBI introduced the Video-Based Customer Identification Process (V-CIP) guidelines in January 2020 and expanded them through subsequent circulars, the regulatory framework has been clear: the video session must be conducted by an authorised official of the reporting entity, must be recorded end-to-end, must capture the borrower presenting their Aadhaar and PAN live on camera, and must include specific verification steps performed in real time.
The requirement for an authorised official — a human — in every session created an instant throughput ceiling. An agent who can run ten sessions per hour, with reasonable session length and handoff time, sets the maximum KYC capacity of the institution. Scale the application pipeline and you scale the staffing requirement proportionally. During peak onboarding periods — product launches, festive season demand surges, DSA-driven campaign peaks — the human agent layer becomes the rate-limiting constraint on disbursements.
The Video KYC Moderator AI does not eliminate the human agent from the V-CIP session. It restructures what the human agent does within it — from conducting the session manually to supervising and approving an AI-moderated session that has already performed the majority of verifiable checks automatically before the agent's attention is needed.
What the AI handles vs what the human agent decides
The division of responsibility in a V-CIP session
The Video KYC Moderator AI handles: pre-session eligibility checks, geolocation validation, face match against Aadhaar photograph, liveness detection, document OCR and data extraction, cross-check of extracted data against application fields, session recording start/stop, real-time compliance flag generation, and audit trail logging. The human agent handles: starting and closing the session, confirming their identity as the authorised official, making the final pass/refer/fail determination, and signing the electronic session record. The regulatory requirement for human involvement is met — at a fraction of the time cost of a fully manual session.
The live session dashboard: 1,000 sessions in progress
The session timeline: what happens in the 7 minutes of a V-CIP session
How 48 agents handle 1,000 simultaneous sessions
With a fully manual V-CIP process, 48 agents running ten sessions per hour each produce a maximum throughput of 480 sessions per hour. The Video KYC Moderator AI changes the arithmetic. Because the AI handles 85–90% of the session autonomously, the human agent's required engagement per session is not seven minutes — it is approximately 68 seconds: open the session, review the AI-generated compliance summary, confirm their identity as the authorised official, and sign off.
At 68 seconds average handle time, each agent can handle approximately 52 sessions per hour. Across 48 agents, that is approximately 2,500 sessions per hour — and the 1,000 simultaneous session load is managed with meaningful headroom. The agent's role shifts from operator to supervisor: they are not running the session, they are certifying it.
The 9.4% of sessions that generate a compliance flag require longer agent engagement — typically 3 to 5 minutes of active review. The Video KYC Moderator AI prioritises these sessions in the agent queue automatically, ensuring that the flag cases receive the attention they require without creating a queue that delays the routine cases behind them.
Scale is not the point — consistent quality at scale is
A video KYC operation that runs 4,284 sessions per day and applies the RBI compliance checklist inconsistently — because agent 31 is more thorough than agent 12, because the 4 PM sessions are less carefully reviewed than the 10 AM ones — is not a compliant operation at scale. It is an inconsistent operation that happens to be large. The Video KYC Moderator AI applies the same checklist to every session, regardless of time of day, agent, or session volume. The consistency is not a feature — it is the compliance argument. The RBI inspects whether your process is consistently applied, not whether it is applied by humans or machines.