Cashfree Payments . 2025
Redesigning Video KYC.
From a 60% drop-off industry
to a 95% completion rate.
A regulated identity verification system built for India's Tier 2, 3, and 4 cities. Where networks fail, documents go missing, and compliance leaves no room for error.
20%
Drop-off reduced
< 2 min
Avg. completion time
95%
Completion rate
Lead Designer (Acting Manager) . Strategy Lead + UX Architect + Mentor + Decision Maker
BACKGROUND
Video KYC sounds simple.
It isn't.
Video KYC is regulated by RBI. A live session, a trained agent, real-time recording. If the video drops for even 2 seconds, the auditor rejects the session. No exceptions, no appeals.
The customer needs their original PAN card physically present. A duplicate fails compliance. And in Tier 2, 3, 4 cities where this product matters most, a stable live video call is not a given.
The industry accepted 55 to 60% drop-off as normal. We didn't.
HOW THE PRODUCT ACTUALLY WORKS
Five actors.
One sequence.
No room for failure.
Most people see Video KYC as a single screen. It isn't. It's a handoff chain across five actors, and a failure at any point fails the entire verification.
Here's how it actually works.
DESIGN SCOPE
We designed 4 of these 6 touchpoints. Customer journey (steps 2 to 3), agent portal (step 4), auditor portal (step 5), and the merchant-facing dashboard. Each had its own portal, its own user, and its own failure modes.
01
Application
Customer applies with merchant
02
Redirection
Redirected to Cashfree verification
03
Verification
Security questions, Aadhaar, and video call
{ Handled by Cashfree }
04
Submission
Agent verifies identity and submits session
{ Handled by Cashfree }
05
Audit
Auditor reviews for compliance
{ Handled by Cashfree }
06
Outcome
Merchant communicates final decision
RESEARCH
We talked to merchants
before writing a single spec.
Before touching Figma, I sat down with early adopter merchants who were already running KYC at scale. Aditya Birla Capital, Kissht, and ClearTax. The goal was simple: understand where the system was actually breaking, not where we assumed it was.
Four things came out of those conversations that shaped everything we built.
01
The drop-off point
Customers were abandoning during personal detail collection. Not during the video call. The problems started earlier than anyone expected, because users had no guidance on what to prepare before joining.
02
Agent fatigue
Agents were context-switching between a video feed and a separate data entry form. The tools were disconnected. Mistakes happened. Sessions ran long.
03
Auditor errors
Review workflows were unintuitive. Auditors were manually reworking cases because the interface made it easy to miss compliance flags. This was slowing down approval rates and creating rework loops.
04
Merchant visibility
Merchants had no real-time view into how their customers were progressing. They found out about failures after the fact. They needed controls, not just reports.
THE REAL PROBLEM
Users weren't failing.
The system was failing them.
Three things kept coming up across every merchant conversation.
Users showed up to the video call without their original PAN card. Nobody told them a duplicate wouldn't work. They'd reschedule, not return.
Users in low-connectivity areas lost video for 2 seconds mid-session. Session rejected. User had no idea why.
Users were filling in the same information they'd already submitted at an earlier step. 73% of people abandon when they hit friction like that.
Three stages. Security questions, Aadhaar via DigiLocker, live video call. Each designed separately. Nobody had thought about the handoffs.
INDUSTRY CONTEXT
68% of users abandon onboarding due to repetitive document submissions. 73% abandon when they hit repeated friction. Industry average completion time: 7 minutes. Drop-off: 55 to 60%.
THE CONSTRAINT
Compliance wasn't the problem.
Working within it creatively was.
RBI says PAN capture has to happen during the call. Not before. Not uploaded separately. During the live session, while the video streams and records simultaneously. Two seconds of drop? Session void.
Every team before us had designed around this defensively. Build for the regulator, not the user.
We asked a different question. What if the compliance requirement is actually the design opportunity?
THE DECISION THAT CHANGED EVERYTHING
Pre-verify the PAN before
the agent even joins.
The biggest reason first attempts failed was simple. Users didn't have their original PAN card.
So we added a pre-call PAN verification step. Scan and verify before entering the live session. If it fails, you get a clear message and a redirect. No agent time wasted. No compliance rejection triggered.
Every user who entered a call was already verified. Agent bandwidth improved. Drop-off on first attempt came down significantly.
We also added multilingual real-time transcription on screen during the call. For users across Tier 2, 3, 4 cities who aren't comfortable in English, this removed the biggest comprehension barrier we had.
THE BUILD
"Let's meet in the middle."
My designs were detailed. Specific states, edge cases for low-network conditions, full user journeys. The FE lead looked at them and said it was impossible to build with the current stack.
The implementation was a no-code/low-code React setup. I'd pushed past what it could handle.
Instead of simplifying the designs, I pushed for a different conversation. We brought in a senior FE architect. The solution was custom components inside the existing environment. Full control where we needed it, without rebuilding anything from scratch.
The FE lead's exact words: let's meet in the middle, you reduce some UX and I'll build better. We didn't reduce anything.
"Let's meet in the middle. You reduce some UX elements and I'll try to build better."
- FE Lead, Cashfree Payments
HOW WE WORKED
Two designers. Four modules.
Three months.
I owned the customer-facing journey. Pre-call verification, the live session, the transcription system.
One designer on my team handled the agent and auditor side. I reviewed his work regularly and made sure both journeys worked as one system, not two separate products that happened to share a database.
For the agent portal, the key design decision was making live video controls float directly over the data entry form. Agents could verify documents without breaking eye contact with the customer. It sounds simple. It took several iterations to get right.
For the auditor portal, we built in timestamped video playback, auto-flagged discrepancies, and keyboard shortcuts for rapid approval and rejection. Auditors work high volumes. Every second of friction in their tool costs the customer on the other end.
Aside from design, I was in the room with the PM and tech lead from the start. Feature scoping, stack decisions, and research with the merchants we spoke to early on.
[ Add: UI screens — customer flow + agent workspace + auditor portal ]
BEYOND THE PRODUCT
Design leadership means
owning the narrative too.
Once the product shipped, I didn't stop there. I sat down with Varun (PM) and Shubham (Eng Lead) to record our vision for what conversion-friendly Video KYC could look like for the next 100 million Indians.
Not a post-launch debrief. A forward-looking articulation of where we thought the product should go and why. Product, engineering, and design aligned on a shared story.
This is the kind of work that doesn't show up in a Jira ticket but shapes how a team thinks about what they're building.
OUTCOME
From 60% drop-off to
95% completion.
35 to 40%
Our drop-off rate
(industry: 55 to 60%)
< 2 min
Completion time
(was 7 mins)
95%
Completion rate
across all attempts
Pre-PAN verification removed the single biggest reason first attempts failed. Multilingual transcription made the session understandable for users across Tier 2, 3, and 4 cities. The custom component approach meant nothing got cut.
Four modules. Three months. Two designers.
WHAT I'D DO DIFFERENTLY
I'd start the tech
conversation earlier.
The pushback on the tech stack came late. Designs were mostly done by the time we had that conversation. We figured it out, but we lost time.
Next time I'd bring the FE architect in earlier. Not at handoff. At problem definition.
The best ideas in this project came out of conversations, not documents.
Back to projects
Next: Global Fintech Festival 2025