Helps people find the best credit card that fits how they actually spend.
Part of the CashKaro and EarnKaro group, India’s largest cashback platform. 20M+ members. About $720M in annual GMV (FY25). Live at great.cards.
150+ cards from 20+ banks. People couldn’t tell which one was right for them. The problem wasn’t knowing about cards. It was feeling sure about the choice.
I built a spend-based card recommender that works off how you actually spend. You pick a goal, choose your spending categories, add rough amounts, and get one clear pick with the savings shown.
A six-week launch reached over 50,000 people. 6,890 got a card that fit after bank checks, nearly 1 in 3 of everyone who saw their results. 18 of 20 tested users could explain why their card ranked first, and over ₹1.2 crore went back to users as cashback.
Lead designer. I owned the input flow, the ranking and results screens, and the patterns partners embed.
Every answer was written by me, not generated 😉
Every card has its own rewards, fees, exclusions, milestones, and rules. Sites like Paisabazaar and CardInsider list them all, but none of them answer the question people actually have:
“Which card is best for me?”
60%+ of card spending in India now happens online, which is exactly where the right card matters most.
70% of people never make the most of their card rewards.
Promos, bank push, creators, and generic “best of” lists.
They compared cards but still couldn’t work out the best option for how they spend.
Nearly half of new cardholders are under 30, and many are getting their first card. They need a clear answer, not a list of 100 cards.
Too many cards, too many offers, too much clutter. I don’t have time to compare it all. I just want to know which one fits what I actually spend.
People already decide on cards inside creator content, comparison sites, bank apps, and finance communities. So Great.cards didn’t need to win its own audience first. It needed to power the places where those decisions already happen.
Trusted creators plug in the engine.
Banks show fit-based results inside their own apps.
One link carries a full recommendation.
Our Bet: power other apps/sources (B2B2C) rather than build our own audience first.
A few constraints. It was a founder-led 0 to 1 build, card rules changed from bank to bank, and it had to work both in our own app and inside partners’ apps.
It was an early 0 to 1, so we learned fast instead of researching for months. We already had two products to learn from inside the group — CashKaro, India’s largest cashback platform, and BankKaro, its credit-card comparison site — and between them they see millions of real sessions a year. So we reused what those had already proven, ran spend scenarios, reviewed card rules with partners, and watched real users in moderated sessions.
People don’t trust a recommendation unless the value is spelled out.
Long forms kill intent before anyone sees the benefit.
Generic “best card” lists read as promotional, so they breed doubt.
People drop off when you ask them to type exact spends.
Sliders with a manual option, so rough numbers are fine.
Rankings feel paid-for, so people stop trusting them.
Showed the savings behind the pick, and kept cashback separate.
More options led to fewer decisions (from past data).
One clear pick, not a long catalog.
Card rules differ by bank, so one shared scale broke the math.
Gave each category its own slider range and steps.
People compared cards but still didn’t apply.
Added cashback, so there was a real reason to apply.
Before Great.cards, picking a card meant a long, uncertain hunt. We mapped that mess, then cut it to a short flow built around the questions people actually had.
We tested five ways to select categories before we found the one that felt effortless. Each had a clear reason it fell short.
People landed on an empty “start adding” screen. “It feels like a form,” one said, and a few dropped off there. Fixed: show the categories up front, no blank start.
People tapped the “selected” tag expecting to see their choices, and looked for a way to change spends later. Fixed: selected categories stay visible and editable throughout.
With questions stacked, people missed the ones lower down and over-scrolled. Fixed: one question in view at a time, with clear progress.
The final screen leads with the categories, clearly marked. People choose what they actually spend on, then enter amounts only for those, so it never feels blank or long.
Pick the categories you spend on. Nothing to type, and no blank screen to start from.
The recommendation needs real numbers, but nobody wants to type exact spends into a new app. I tested five ways to enter a number, each fixing the last one’s weak spot.
Fast to tap, but the fixed values were never personal, so the savings looked vague.
Dragging felt easy, but people had weak control over the exact number.
Snapping to steps improved precision, but the steps slowed people down.
Faster to move, but still hard to be precise on large amounts.
Free-flowing so no one is boxed in, tuned to real spend so the usual ranges are easy to hit, with tap-to-type and plus/minus for precision. The best balance.
It reads as one simple control, with the per-category logic tuned underneath.
Drag for a rough number, or tap the amount to set it exactly. Every category has its own range.
The slider felt simple because the numbers under it were set per sub-category. I set each range and its steps from that category’s average spend, then handed engineering a breakpoint spec, built with AI, covering every sub-category.
Representative examples; every sub-category shipped with its own range and steps.
People needed four things at once: which card ranked first, why, how much they’d save in each category, and a reason to trust it before applying. We tested three layouts.
We tested the card-first showcase — then pulled it back.
It looked cleaner and more premium, but people noticed the card and missed why it ranked first.
The card that ranked first, what it saves, and the fee it costs — before you scroll.
People could see the result came straight from their own spending, category by category, benefits added, fees taken out. It felt less like a paid ranking and more like a decision they could check for themselves.
Every number opens up: category by category, benefits added, fees taken out.
We rank by what fits the person, never by what a partner pays us. Cashback sits separately so it can’t change the order. The US regulator (CFPB) has warned that pay-to-rank “advice” can break the law.
We show fees, exclusions, and eligibility up front. These are the things people usually miss.
Every slider also takes keyboard and manual input, with proper labels and readable text. A full accessibility check and screen-reader testing are planned, not done yet.
CFPB Circular 2024-01, consumerfinance.gov, 2024
This was a blended launch across CashKaro campaigns, EarnKaro creators, and paid and organic channels, over the Diwali shopping weeks. There was no earlier Great.cards version to test against, and the channels carried different intent, so I read these as a launch result, not a controlled lift.
Nearly 1 in 3 people who saw their results applied and got qualified for a card.
Six-week launch, 20 Oct to 1 Dec. Total reach over the window was higher; this funnel is measured on a 50,000-user base. The 6,890 excludes rejected, reversed, cancelled, and duplicate applications; repeats were caught with an “already applied” state. Because the channels carried mixed intent and the window overlapped Diwali, this is a blended launch result, not an isolated measure of the design.
We shipped the recommendation flow first and confirmed it worked. Then we added cashback, a real reason to apply, paid after approval, and ran a second phase at the same reach.
At the same 50,000 reach, adding cashback lifted applications by about 18% and approvals by about 16%. This phase ran after the Diwali weeks, not during them, so the gain held even without that seasonal push.
These were sequential phases, not a controlled A/B, and cashback also makes the offer easier to market. So I read this as a real, directional gain from cashback, not an isolated causal lift.
Trust needs visible logic.
People act when they can see how the number was made.
Don’t ask for precision too early.
Rough input first, accuracy later, beats a wall of fields.
Local rules are the product.
Rewards, fees, milestones, and exclusions can’t be an afterthought.
It’s a decision tool, not a catalog.
It worked when it helped people choose, not when it showed more options.
We learned plenty from later signals, but cleaner baselines would have shown exactly what moved people: card fit, the value math, input speed, trust in the ranking, or the cashback. Next time I’d measure at three levels.
Can people say why a card was recommended to them?
Do more people move from the recommendation to actually applying?
Are they applying with a clear grasp of the value, fees, and cashback terms?
The logic isn’t tied to India. Swap the cards and the reward rules, and the same engine works. Here it is on US cards.
The launch numbers (reach, results, qualified cards, cashback) and the comprehension result are internal Great.cards figures from the six-week launch.
This case study is private.
Enter the 4-digit PIN from the link you were sent.
That PIN doesn’t match.