UPI skipped the card

In 2016 India had close to a billion debit cards and a few million card terminals. The cards were never the constraint. Within a handful of years a vegetable seller was taking digital payments through a printed square taped to his cart, and the transaction cost him nothing. The interesting part is not that the queue at the cash counter got shorter. It is what the design skipped on the way.

Every payment system has two ends, and a card system builds its economics at the payee's end. That is where the cost sat, and therefore where the exclusion sat. UPI moved the cost of accepting a payment to roughly zero and made settlement immediate, and then had to answer for three things the card had been quietly providing: a reversal window, a rule about who pays for fraud, and a pool of money to fund rewards.

The card's bottleneck was the merchant, not the customer

Cards are cheap to issue and expensive to accept. A terminal is bought or rented, needs a data line, needs a merchant account and an underwriter's approval, and takes a cut of every ticket. In India the debit-card fee was capped at 0.90% above ₹2,000 from January 2018, and 0.40% below that. On a ₹40 cup of tea, two per cent is eighty paise — less than the paper roll in the printer. The merchant refuses, and the refusal is rational.

So acceptance concentrated where tickets are large: supermarkets, fuel stations, hotels, hospitals, electronics. India issued close to a billion debit cards over the same period, but most of them were ATM cards in all but name; the average debit card was used at a terminal a couple of times a year, if that. A card network is a two-sided market, and the side that never showed up was the merchant side.

Settlement made the arithmetic worse for exactly the shops that needed it most. A card payment credits the merchant in T+2, later T+1, minus the fee. A small shop that restocks daily is lending the difference to the bank for two days at no interest. The fee is visible and gets argued about; the float is invisible and does not.

A QR code is a piece of paper

UPI replaced the terminal with the customer's phone and a printed code. The merchant needs no device, no underwriting, no rental, no minimum volume. The marginal cost of adding one more acceptance point is the price of printing a square. That single substitution is why acceptance could be installed across the whole merchant distribution in a few years rather than a few decades — from a few million terminals to several hundred million codes.

The rail underneath already existed. UPI settles on IMPS, the immediate payment service NPCI has run since 2010, which moves money between bank accounts in real time, around the clock. UPI added three things on top: an address that is not an account number, an app layer that any bank or licensed provider can build, and one authorisation standard. It is not a wallet. The money never sits with the app; it sits in the bank account, which is why banks adopted it instead of fighting it.

Then the price was set to zero. From 1 January 2020 the RBI barred banks and NPCI from charging a merchant fee on UPI and RuPay debit, following an income-tax provision that obliges large merchants to accept at least one electronic mode with no fee. Zero MDR removes the merchant's objection completely, and removes the revenue that pays for the network at the same time — a hole the system covers with a per-transaction incentive for small tickets, which is a subsidy wearing a different name.

What the short path gives up

Reversal first. A card payment is a pull: within a defined window the issuer can take the money back from the merchant without asking. That is what a chargeback is, and it is the reason a stranger's card is worth accepting. A UPI transfer is a push authorised by a PIN. Once it lands, undoing it requires the recipient's agreement or a dispute process, and the scheme is built to protect the receiver. So the fraud that grows is the kind that gets the payer to approve: a caller who talks someone through a collect request, or an app that takes over the screen. The RBI's limited-liability rules give a customer zero liability when a third party breaches the bank's systems; they do not cover a payment the customer authorised, even under deception. Most arguments now happen on that line.

Failure is the second difference, and it is subtler. A card declines before any money moves. A UPI transaction can debit the payer and then fail at the credit leg, which leaves the money in the air. The 2019 turnaround-time rules require automatic reversal of a failed transaction within a day, which converts a loss into a delay — but a delay at the wrong moment is still a cost for somebody buying stock that afternoon.

Rewards are the third, and the least discussed. Credit cards are funded by interchange, roughly 1.8–2% of the ticket, which pays the network, the issuer's risk and the points. UPI has no interchange, so there is nothing to fund points with. A zero-fee rail and a rewards economy are arithmetically incompatible: you can have one, or you can charge the merchant. Putting a RuPay credit card on UPI, from 2022, keeps the credit line and drops the fee, which moves the question rather than answering it.

It worked because the card network was thin

The same design in a country with a dense terminal network and a large interchange-funded rewards business would have had to displace something with revenue and a lobbying budget. In India the merchant side was mostly absent, so there was almost nothing to displace, and the QR code did not have to win a fight — it arrived in a segment the cards had never reached.

There is a structural difference too. The switch is a not-for-profit owned by the banks, not a listed network whose revenue is a percentage of the ticket. That is why a zero price was technically possible at all: no shareholder was being asked to give up an interchange stream. It also means the cost of the rails is carried by the banks and offset, in part, by a government incentive for small-ticket payments. Naming that is not a criticism. It is the reason the price could be zero, and the reason the price is now a live question.

Where the bottleneck moves next

The general lesson is about which part of a system you decide to make cheap. Where cards reached, the payment problem was solved by adding security and rewards to an expensive rail. Where they did not reach, it was solved by deleting the expensive part and accepting the losses that part had been absorbing. Both are rational. They produce different fraud, different disputes and different distribution of who gets paid.

And the bottleneck has already moved. With acceptance solved, the constraint is no longer the merchant's terminal but the payer's balance — which is why credit on UPI, and the question of who funds it, is the next argument rather than a detail.

This is one of the Modern India entries in About India. The pillar's argument — that each of these systems has a stated reason and an actual one — is on the pillar page.