NPS onboarding took 20 minutes. PensionBox made it 2, Got Acquired by Zerodha
Building India's first purpose-built NPS account opening experience from scratch, at a fintech startup with no PM, that cut industry time by 90% and was acquired by Zerodha.
Outcomes
~2 mins
Avg completion
90%
Faster than industry
2L+
Users onboarded
Zerodha
Acquired by, 2025
My role
Nishit Mangal · PM, Designer, Researcher, Barista ;)
Joined Day 0 as the only founding designer & first employee. No PM. So I became one.Owned the design end-to-end. Led prototyping, user testing, developer handoff and post MVP testing.
About PensionBox
PensionBox was a retirement fintech built to help working Indians plan for life after 60. The core product let users track their retirement corpus, set savings goals, and invest through government-backed instruments like NPS(similar to 401k).
Just founders and me operating in a space most fintechs ignored. Retirement planning had no Zerodha moment yet. No one had made it simple, trustworthy, or mobile-first for the average Indian salaried professional.
That was the gap PensionBox was built to fill. In 2025, Zerodha acquired us.
Overview
In 2021, opening an NPS account in India meant 20-minute ordeal of long forms, government jargon, and random redirects to third-party portals that looked like scam pages. Every platform plugged into the same government API, painted a UI on top, and shipped it.
Nobody questioned whether it could be different.
I joined PensionBox on Day 0 as first employee. No PM. No design system. No prior version to reference. Just a problem: make opening an NPS account feel like something a 26-year-old would actually do.
I reverse engineered it, each step, each field was questioned, that’s how we cut that 20-minute industry average to under 2 minutes, onboarded 2 lakh+ users, and was acquired by Zerodha in 2025. The NPS onboarding module was a core part of what made that conversation happen.
Let’s break down the problem
National Pension Scheme(NPS) is a Indian government investment instrument, that focuses towards saving finances for citizens retirement. Just like how equity market is regulated by SEBI. NPS is regulated by PFRDA and Gov. handles distribution & complaince of APIs to fintech apps to build tech layer.
Industry problem : accepted a broken experience as the default
Every fintech that offered NPS treated it as a compliance checkbox. Offer the product, check. Build the flow, check. The flow itself was never the point.
The result was a category-wide failure. Government portals took 20+ minutes. Bank websites were no better. KFintech, the government's own partner portal and the backend every platform ran on, had a UI that had not meaningfully changed in years. PaytmMoney, the closest thing to a considered experience, still asked users for the same information twice across different screens and treated NPS as a side door feature.
The 20 minute average was not an outlier. It was the standard. Everyone had looked at the problem and concluded: this is just what NPS onboarding is.
We disagreed.
Visualizing market speed
Talking to users before drawing anything
With the data model mapped, I ran card sorting sessions with 20+ users across all three knowledge levels. Two specific questions guided everything: where do users mentally group information, and where do they feel unsafe?
Users consistently separated identity information (who I am) from financial information (how I pay) from investment information (where my money goes). The government API organised fields by technical dependency. Users organised them by emotional weight.
Anxiety peaked at three moments: entering PAN and Aadhaar together on one screen, the payment step, and the redirect to KFintech. Those two insights — mental grouping and anxiety mapping — became the architecture.
Step by step journey
Iterations
API-native flow with clean UI
Chunked forms without pre-fill
Pre-fill without consent framing
Learnings
Each failure had a precise lesson. Iteration 01 said the problem was structural, not visual. Iteration 02 said the problem was trust, not just cognitive load. Iteration 03 said that pre-fill only works when users feel in control of it.
The final solution combined all three learnings: chunked screens, pre-fill, and explicit consent with visible edit controls at every step. None of it would have been possible to validate without going through the failures first.
