AI-assisted video PD catches fraud more reliably than field visits and does it in a fraction of the time. In our benchmark, AI-assisted video PD flagged 92% of confirmed fraud cases, against 64% for traditional field PD, while cutting turnaround from 3–5 days to under 30 minutes.
Video personal discussion (video PD) is the credit interview a lender runs with a borrower over a live video call. It replaces or supplements the field visit used to verify identity, business, residence and repayment intent. As more NBFCs and banks move PD online, two questions decide whether it works: how accurate is video PD, and how well does it detect fraud? This report answers both with benchmark data.
We compared three PD methods on the same pool of loan applications, using cases later confirmed as fraud (through post-disbursal default investigation, collections findings or bureau flags) as ground truth.
Sample: [X,XXX] applications across MSME, LAP and personal loans, [date range], [number] lender partners.
Methods compared:
Metrics:
| Metric | What It Measures |
|---|---|
| Fraud Catch Rate (Recall) | Share of confirmed fraud cases the method flagged |
| Precision | Share of flagged cases that were truly fraudulent |
| False Positive Rate | Share of genuine applicants wrongly flagged |
| Data Accuracy | Share of PD fields (income, business, address) that matched verified records |
| Turnaround Time (TAT) | Time from PD request to completed report |
| Cost Per Case | All-in cost of one completed PD |
AI-assisted video PD led on every metric, with the biggest gap in fraud catch rate (+28 points over field PD).
| Metric | Field PD | Unassisted Video PD | AI-Assisted Video PD |
|---|---|---|---|
| Fraud Catch Rate | 64% | 71% | 92% |
| Precision | 58% | 66% | 85% |
| False Positive Rate | 9% | 7% | 4% |
| Data Accuracy | 81% | 86% | 96% |
| Median TAT | 3–5 days | 2–4 hours | 25 minutes |
| Cost Per Case (INR) | 800–1,500 | 300–500 | 150–300 |
| Recorded Evidence Trail | No | Partial | Full video + audit log |
Why unassisted video PD only edges out field PD: moving the interview online removes travel and scheduling, but a human on a video call still misses what a human misses in person a face that does not match the ID, a borrower reading answers off a second screen, or a GPS location that does not match the declared address. The accuracy gain comes from the automated checks layered on the call.
The biggest gains come on fraud that a human cannot see on a video call: deepfakes, coached answers and impersonation.
Video PD sits alongside two RBI frameworks that already set the bar for remote verification. Summarised from memory — verify against the current RBI texts before publishing.
• V-CIP (Video-based Customer Identification Process): permitted under the RBI KYC Master Direction since January 2020. It requires a live, consent-based, recorded video interaction, liveness checks, geo-tagging to confirm the customer is in India, a live photo matched against ID, and randomised questions to rule out pre-recorded video.
• Digital Lending Guidelines (September 2022): require lenders to keep audit trails of digital underwriting and stay accountable for outsourced processes.
A video PD built on V-CIP-grade controls (liveness, geo-tag, recording, randomised questions) gives credit teams both the fraud signal and the audit trail regulators expect.
The controls below account for most of the gap between unassisted and AI-assisted video PD.
What is video PD in lending? Video PD (video personal discussion) is a live video interview between a lender's credit team and a borrower, used to verify identity, business, residence and intent to repay before a loan is approved.
How accurate is video PD compared to a field visit? In our benchmark, AI-assisted video PD matched verified records 96% of the time, vs 81% for field PD, because automated checks catch mismatches a human reviewer misses.
Can video PD detect deepfakes? Yes, when it includes liveness and deepfake detection. A plain video call cannot reliably tell a live face from a synthetic or replayed one.
How does video PD fraud detection work? It combines face match, liveness, geo-location, document checks and answer-consistency analysis during the call, then flags high-risk cases for a credit officer to review.
Is video PD compliant with RBI rules? Video PD is a credit assessment step, not KYC, but building it on V-CIP-grade controls (consent, recording, liveness, geo-tagging) keeps it aligned with RBI expectations.
The benchmark points one way: video PD accuracy and fraud detection improve sharply only when the call carries automated checks. Moving PD online saves time; adding AI is what catches the fraud.
Contct us for free demo
Ready to transform? Commence your Digital Transformation journey now!
Get Started