← India FS Pulse

Memo

A price does not kill a rail, it segments it: cards kept the large ticket and lost the volume

Generated 2026-11-05 by analysis/13_cards_control.py. Every figure below is interpolated from a computed value, so the prose cannot drift from the data.

The answer

India runs the experiment this report has been reaching for. In 2026-07, cards and UPI moved money between the same merchants and the same customers under the same regulator, and only one of them was allowed to charge for it.

The free rail took the volume: UPI ran 33 times the card transactions and 12 times the value. Cards were left with 2.9% of the transactions between the two rails.

But cards kept 7.6% of the value, which is 2.6 times their share of transactions. The priced rail did not lose. It retreated to the large ticket, where a fee is small enough to absorb: Rs 3,482 against Rs 1,263, 2.8 times.

A merchant discount rate does not decide whether a rail survives. It decides which transactions it gets. Across the 3 months transcribed the card share of value moves by only 0.2 percentage points and the ticket multiple barely moves, so this is a standing split rather than one month that happened to look this way.

Three supporting arguments

1. Acceptance is the clearest thing a price has ever moved. There are 803 million UPI QR codes in the country against 10.0 million card terminals: 80 to one. A terminal is hardware that costs money to deploy and carries a fee on every swipe. A QR code is a printed sticker that carries nothing. Where a merchant can be paid has been decided, and it was decided by price rather than by technology.

2. The two rails are not competing for the same transaction. At Rs 3,482 against Rs 1,263, they serve different purchases. That is why zero MDR did not simply destroy the card business: a fee of a percent or two is invisible on a Rs 3,482 purchase and impossible on a Rs 1,263 one. The report’s earlier finding, that the merchant leg is volume-heavy and value-light, now has a mechanism attached to it rather than only a description.

3. It bounds what re-pricing UPI could ever achieve. Sub-module C prices the merchant leg at rates nobody charges. This is the same country actually charging one, and the priced rail holds 7.6% of the value between them. Anyone underwriting a return of MDR is underwriting a migration of value from the free rail to the priced one, and cards show what that side looks like at its current size.

So what

Method and its limits

Card figures are RBI’s own national Total row, transcribed by hand from the monthly workbook because RBI’s document links return an interstitial to a script. That Total was reconciled against the sum of its 63 bank rows at 0.00% on every column transcribed, which is the check that makes a hand-transcribed figure usable. Cash withdrawal is excluded throughout: it is a cash logistics event, not a merchant payment. UPI is the NPCI series this repo already carries, for the same calendar month, so no period is stretched.

What this does not say. It does not price the card rail. RBI publishes no blended effective MDR; the ceiling differs by instrument, merchant category and ticket, so multiplying a statutory rate by total value would publish an estimate as a measurement, and this module refuses that exactly as sub-module B refuses a rate across mismatched periods. It rests on 3 months, 2026-05 to 2026-07, each a separate workbook downloaded and transcribed by hand. That is enough to show the split is standing rather than incidental, and not enough to call a direction: three points do not make a trend, and every further month is another manual transcription. And card value here excludes transactions on cards routed over UPI, which RuPay credit-on-UPI makes a growing and separately unpublished category, so the boundary between the two rails is blurring in a direction this data cannot yet see.