The UPI Playbook Doesn't Export. That's the Point.

At central-banking conferences, the same slide has become familiar. It plots the monthly volume of India's Unified Payments Interface — an emphatic line up and to the right — above a single word: "Lessons." The implication is rarely in doubt. India built a public payments rail; other countries should build one too.

UPI deserves the admiration. It is among the most consequential pieces of public digital infrastructure built this century. But the lesson usually drawn from it is wrong. What India created cannot be reduced to a rail, and the conditions that made it work are not readily exportable.

That distinction has been lost as the export narrative has hardened into consensus. Brazil's Pix, Thailand's PromptPay and Nigeria's recent NIBSS overhaul are all discussed alongside UPI. NPCI International, which is taking the system abroad, has live corridors in seven countries and continues to sign new agreements. "Stack" has become part of the diplomatic vocabulary. Beneath the enthusiasm lies an assumption that is seldom made explicit: that UPI is, at heart, software.

The panel that runs at every conference

The case for UPI is strong. In less than a decade, India moved from an economy in which cash dominated small consumer purchases to one in which a vendor selling ₹40 of coriander will expect a customer to scan a QR code. The scale is extraordinary.

₹0 merchant discount rate — the per-transaction fee that funds card networks everywhere else — set, deliberately, to zero.
~16B transactions a month — more monthly transactions than the rest of the world's instant-payment systems combined.
7 live export corridors — countries with a working UPI link or acceptance deal, with more under negotiation.

It is perfectly reasonable for a finance ministry in a middle-income country to look at this and want something similar. The error lies in the shopping list. Governments reach for the visible artefacts — API specifications, QR standards and switch architecture — because these can be procured. A switch can be licensed, integrators hired and a sandbox launched with a ribbon-cutting in 18 months.

The reasons UPI worked do not fit in a procurement document.

You cannot import a political settlement through an SDK.

What the rail actually sits on

UPI rests on three preconditions, none of them technical. The first is identity. Aadhaar, the biometric identity programme, came before the payment rail. The Indian state spent much of a decade building it, defending it in court and spending political capital on it. When UPI launched, the difficult question of identifying the person at the other end of a transaction had largely been addressed. Countries seeking to emulate UPI often treat that step as optional. India took ten years and a Supreme Court case to get through it.

The second is demographic. UPI arrived as a young population was coming online through inexpensive smartphones, just as mobile-data prices collapsed. A new entrant brought effectively free data to the market in 2016. This was not a payments policy, or even a planned ingredient of one. Yet the adoption curve would look very different without cheap connectivity for hundreds of millions of first-time internet users.

The third is the least comfortable to acknowledge: the regulator was prepared to make banks bear the cost. Zero MDR was not a clever pricing tweak. It removed, by decree, the fee pool that supports payments infrastructure in most markets and required banks to absorb the loss as the price of participation. That was not a product decision. It was a state using its leverage over the financial sector in pursuit of a public objective.

Together, those conditions make the UPI "stack" look less like a technology than a snapshot of one country at one moment, backed by a state able to do particular things. The software is real and well designed. It was also, comparatively, the easy part.

Pix didn't copy the API. It rebuilt the politics.

Pix, Brazil's instant-payment system, provides the useful comparison. It is a striking domestic success — arguably UPI's equal — and it did not copy India's API. Banco Central do Brasil built its own system, on its own timetable and with its own technical choices. What it reproduced, or independently reached, was the political structure beneath it: a central bank able to mandate participation, a decision to make payments cheap or free for users, and an existing if imperfect identity layer to link accounts to people. The resemblance to UPI lies in the settlement, not the software.

The chart below captures a point the export panels rarely make: adoption at this speed is not incremental. It comes when the groundwork is in place and a switch is finally thrown.

Figure placeholder — asset key: upi-monthly-volume

The wall, not the slope. The curve everyone screenshots is the output. The preconditions — identity, cheap data, a regulator willing to impose a loss — are the inputs, and they were assembled years before the line moves at all.

The distinction among imitators is therefore fairly simple. The successful ones have done the unglamorous, multi-year work of assembling these conditions, using India as evidence that the outcome is possible. The others buy the switch and wait for the curve.

The corridor counterargument

The people running UPI's international expansion see this differently. Their public case is not primarily about replicating UPI at home. It is about building corridors first — for remittances, tourist payments and cross-border settlement — and allowing domestic use cases to follow once the link is useful.

This is a serious argument. The Singapore–India PayNow–UPI link is its proof of concept: two systems with different domestic politics joined at the settlement layer to move remittances more cheaply. Over 15 years, a corridor-first strategy may achieve some of what domestic replication cannot, through cross-border utility rather than a domestic mandate.

But it also concedes the central point. It does not promise that a country can buy a stack and reproduce India's domestic adoption curve. It offers something narrower, and more credible: a connection to India's rail and a share of cross-border flows. That is a real prize, but it is not the product advertised on the conference panel.

What the five-year read misprices

For investors, the important gap is between the 15-year story and the five-year one. That is where capital is most likely to be misallocated. Four errors stand out.

The "build a UPI" mandate as a procurement — Governments fund a switch and a sandbox, declare victory at the ribbon-cutting, and then quietly watch adoption flatline because nobody bought the preconditions. The write-off shows up two budget cycles later, attributed to "execution."
Fintechs pricing in zero-MDR economics elsewhere — India's free rail is a subsidy in disguise. Models that assume the same merchant economics in a market where someone still has to fund the rail are quietly assuming a political decision that hasn't been made.
Treating identity as a fast-follow — Plans that skip the identity layer to save time are mispricing the single most expensive precondition. Aadhaar took a decade and a constitutional fight. It is not a sprint item.
Reading corridor volume as domestic traction — Cross-border remittance flow is real revenue, but it is not the hockey stick. Confusing the two is how a corridor business gets valued like a domestic-network business — and then re-rated, hard.

None of this is an argument against UPI. It is an argument for care in deciding what UPI proves. The rail is a triumph. The broad generalisation taken from it is a category error — and category errors can be expensive because they often feel like insight.

The rail is the visible ten per cent. The preconditions are the ninety per cent nobody can screenshot.

A method, not a manual

There is a lesson worth exporting, but it is a method rather than a manual — and it is less convenient for those selling the stack.

Countries should identify the local conditions that perform the work India's identity layer, data-price collapse and regulatory resolve performed in India. A country may already have a strong national ID but lack acceptance density. It may have widespread smartphone use but no political appetite to make the rail free. Its banks may be too weak to absorb a decade of zero-MDR, making bank reform more urgent than a new app. The work is diagnostic, country-specific and slow. It bears little resemblance to buying a switch.

That is also a useful way to read Asian technology more broadly. The copyable object — the rail, the super-app, the QR sticker at a paan stall — is rarely what made the system succeed. It is the visible result of conditions that were political and demographic before they became technical. Read only for the rail and it is easy to buy switches and wait for curves that never arrive. Read for the conditions, and there is at least a chance of building something that lasts.

The conference slide will keep reappearing, and the line will keep climbing. The word beneath it deserves to change. Not "Lessons." "Preconditions."