On the 12th April 2021, the People’s Bank of China (PBoC) summoned Ant Group for the third time in under six months. In the published records of the discussion, one line, translated from Mandarin, caught my eye.
Ant was instructed to sever what PBoC deemed improper connections between Alipay and Huabei and Jiebei, its consumer credit products, and to correct what the translation termed “violations such as nesting” of its credit business inside the payment chain.
Five years on, the same question of whether credit belongs inside the payment chain, that is the “nesting of credit”, has arrived at three of the world’s largest public rails. India’s Unified Payments Interface (UPI) is run by the National Payments Corporation of India (NPCI) and Brazil’s Pix by the Banco Central do Brasil (BCB), and both have already had to decide how credit gets onto the rail. The euro area is different. Its public rail is the Single Euro Payments Area (SEPA) Instant Credit Transfer, or SEPA Instant for short. It is a scheme standard that the European Payments Council (EPC) writes, that EU law obliges every payment provider in the euro area to offer, and that settles in central bank money through the Eurosystem and EBA Clearing. Nobody owns it the way NPCI and the BCB own theirs.
For SEPA Instant, the credit decision sits one layer up, with the European Payments Initiative (EPI). The EPI runs Wero, the digital wallet and account-to-account (A2A) scheme built on top of the SEPA Instant rails, and it faces the same decision. So when I say “rail owner” in this article, I mean the body that decides the rules credit rides under: NPCI, the BCB and, for Wero, the EPI.
Meanwhile, fintechs across those markets remain hopeful that the credit arriving on these rails is theirs to run. Take Falcon, an Indian fintech that runs credit programmes for banks, which wrote in December that credit lines on UPI would “reshape the entire ecosystem”, with fintechs like itself doing the underwriting and collections for the banks that hold the loan. In my recent LATAM wallets article, I had made a bet of my own and argued that the wallet providers that matter on these instant payment rails would be the ones that put credit and real-time underwriting back on top. I framed it this way.
Private credit layered onto public payment infrastructure is, in my view, the business model hidden within every one of these growth forecasts.
That claim is what I wanted to test in this article, and to do so, I built a Wardley map. For those unfamiliar, a Wardley map is a diagram of everything a user’s need depends on. Each component is placed from top to bottom according to how visible it is to that user. It also shows how far the component has evolved and become commoditised, moving left to right through four stages, namely, genesis, custom-built, product and commodity, from a novel experiment to a metered utility. Placing things that way shows where a competitive advantage might reside. By competitive advantage, I simply mean a position that a rival can’t copy at the same cost. A commodity can still generate profits for whoever operates it, but it can’t be an advantage for anyone building on it, because every competitor builds on the same thing on the same terms.
It’s also important to understand the map’s perspective before you read it. A Wardley map is anchored on the needs of the people being served, not on whoever is using it for strategising. In the map I created, the people being served are the shopper paying at the checkout and the merchant being paid for the goods or services being sold. Thus their needs sit at the top, and every component is placed by how visible it is to the shopper.
The map lays out the chain of things a wallet provider or fintech needs in order to put credit on top of the rail and make credit on UPI, Pix, or Wero its main checkout product over the next five years. Read left to right, the map indicates which of those things the fintech should build itself and which it should buy or rent.
With that clarified, I used the map to help me verify my claim by answering two questions. Firstly, “Is credit on public rails a category in its own right?” If it is, the business model in my claim exists, and the map should show where a competitive advantage can sit within it. Secondly, “Should the rail owner be the one doing the lending?” If it should, the business model exists but isn’t a private one.
What the map says: the competitive advantage isn’t in the rail
So here is the map. It has 54 nodes: four are the user needs (the shopper, the merchant, the rail owner and the state), and the other 50 are the components those needs depend on, from checkout down to compute. It is one map across the three markets, and where a component sits differently in one of them, its rationale in the interactive version says so. Every component carries a rationale and the toggle shows where I expect things to sit in 2031. A bar beside a component marks inertia and provides a reason its owner would resist that move.
Of the 50 components I plotted on the map, three weigh heavily in determining the answer, because the business in my claim would depend on all three, and a fintech can own at most one.
The first is the instant rail itself, and it sits in the commodity band. It has mandated reach, standardised messages, a single switch under it in India and Brazil, and a price to the payer that is zero by rule in those markets and capped by law in the euro area at the price of an ordinary bank transfer. The euro area is a step behind the other two, because paying instantly from an account is not yet the habit there that it is in India and Brazil. Instant transfers were 25% by count and 8% by value of euro-area credit transfers in the second half of 2025. The card networks show that a competitive advantage can sit above a commoditised rail, but theirs is the scheme rulebook, the dispute framework and the fraud and risk scoring around it. On a public rail, those belong to the rail owner, or are rented from the card networks.
The second component is the credit rulebook on the rail, sitting at custom-built. This is the set of rules for how credit may ride the rail, including, but not limited to:
- who is allowed to fund it
- what it is called and how it is classified
- where it may be used and how it may be offered to the shopper
- what fee it carries
- who is liable when it fails
A lender sets the terms of its own loan, but only the rail owner writes these rules, within the limits its regulator sets, and of the three, only NPCI has done so thus far. It wrote the operating rules when the Reserve Bank of India (RBI) allowed pre-sanctioned credit lines at banks to be linked to UPI in September 2023. The BCB had Pix Parcelado, an instalment feature of the scheme itself, on its roadmap from 2023, postponed it three times and on 4 December 2025 decided not to regulate it at all. The EPI has said since September 2025 that Wero will gradually cover payment “in several instalments”, with no lender, no product and no date attached.
The third component is the lending balance sheet, with licensing, regulatory oversight, and capital rules beneath it. It sits at product, low on the map where the shopper never really sees it, and almost every credit product on the map depends on it. Whoever holds it holds the risk and earns interest as their reward.
Now going back to my first question. Is credit on public rails a category in its own right? On a map, a category is a component that can’t be reduced to one already there, and that still has room to evolve. So I tested credit on public rails against its three nearest neighbours: a feature of the rail itself, a card, and ordinary bank lending.
It isn’t a feature of the rail, because none of the three rail owners has made credit part of the rail itself. The RBI ruled in June that a credit line drawn through UPI takes the prudential treatment of the underlying facility, so it is still just that loan. The BCB declined in December to put instalments into Pix. And NPCI carries credit on UPI as a card instrument, RuPay, which is where almost all of the volume is. In August 2024 NPCI’s chief executive, Dilip Asbe, put credit on UPI at about Rs 10,000 crore a month, a crore being ten million, so about $1.1 billion, of which the rail’s own credit-line product was Rs 100 to 200 crore ($11 to 21 million). The rest was RuPay credit cards paying through the same QR codes, and a year later the credit-line product was still only at roughly Rs 500 crore a month ($53 million).
It isn’t a card, because a card has a scheme rulebook around it, with an interchange fee the merchant’s side pays and a dispute right the shopper can invoke. A Pix funded from a loan has neither, and a credit line drawn through UPI has no dispute right and no published interchange.
And it isn’t ordinary bank lending, because of where the loan is drawn and whose rules it is drawn under. A personal loan is paid into an account before the shopper spends it. Credit on a public rail is drawn at the point of checkout, one payment at a time, through the rail’s own credential, the QR code or the Pix key, and it runs under the rail owner’s rules on top of the lender’s terms. So yes, it is a category in its own right.
Since it is a category, it needs a name, and the translation of the PBoC’s records from the beginning of the article provides a useful one. I’ll call it unnested credit, a loan made by a licensed lender separate from the rail owner, paid out, accepted and repaid over the public rail, under rail rules that the rail owner defines. That last point is what makes this different. Without the rail owner’s rules, what’s left is a bank loan that the shopper happens to spend over a public rail. Unnesting is what the RBI, the BCB and NPCI each did in the decisions above, and it’s what the PBoC forced on Ant in 2021.
You will notice on the map that NPCI’s own card on the rail sits well inside the product stage, and that the line of credit a bank provides and the rail pays out sits only at the start of it. In Brazil that line is what Nubank sells as Pix no crédito, a Pix funded from the customer’s card limit and repaid in up to 12 instalments with interest, and by Serasa’s count Itaú, Santander, Inter, Bradesco and Mercado Pago run versions of it at monthly rates of around 5%. It is one product per bank, with no scheme standard between them, and the merchant receives an ordinary Pix and never learns that credit was involved.
So what does that do to my claim? Credit on public rails is a category, and the lenders in it so far are licensed, banks and small finance banks on UPI and the banks behind Pix no crédito, so the business model in my claim is real. But my claim pictured private credit layered onto the rail, a fintech adding credit at its checkout button and owning it, and the evidence says the credit isn’t layered on the rail at all. It is unnested from it, a licensed lender’s loan under the rail owner’s rules, and a fintech that is a tenant of the rail owns neither the loan nor the rules. The card-issuing bank that already owns the app the shopper pays from has the licence, the shopper and the checkout, and can copy the button. Read as a lender’s business, the claim passes. Read the way I wrote it originally, as the wallets’ business, it fails.
Credit is coming to the public rails, and where it has arrived, it has done so unnested, with a licensed lender holding the risk and a rulebook for the rail owner to write.
So where does the competitive advantage lie for a fintech that does become a lender? In two places the rail doesn’t reach, and neither is equally certain. The first is the balance sheet, which is where the interest on these rails has gone so far. In India, Slice, which merged with a small finance bank in 2024 and reported its first profitable year in the year to March 2026, is the one fintech I can find holding the loan it distributes on the rail. The others distribute a bank’s card or line, and without a licence a fintech’s position on UPI is distribution alone.
The second is underwriting on the transaction data a lender can lawfully see, which means the rail’s history reached through the shopper’s consent, and this one is a forecast rather than an observation. A bank sees its customers’ account history, and a credit bureau sees repayment records. Neither sees what passes over the rail, which is every payment the shopper has made on it, from every account they hold. Today that history stays with the rail and the shopper’s individual banks. India’s Account Aggregator, the consent layer built to carry it to any lender the shopper chooses, had fulfilled about 493 million consent requests by June 2026 with nothing on the public record tying any of them to a credit line drawn through UPI, and every lender whose method is on the public record still underwrites from a bureau or its own card book. That is why I’ve placed the underwriting practice at custom-built, the least evolved of the three positions the map’s title names.
The reason I still expect it to become an advantage is the size of the loans. RuPay credit on UPI runs at about Rs 850 ($9) a transaction, by my arithmetic from the figures given to Parliament, and Pix’s median transfer is R$36 ($7). A loan that size, dozens of times a month, at a merchant who pays nothing towards it, earns its keep on the shopper’s interest and on a loss curve that a bureau’s monthly repayment records are too coarse to price. Whoever learns to price that curve from the rail’s history first has the one advantage on the map that the shopper’s own bank hasn’t already got. That arithmetic sits on the map as small-ticket economics, at custom-built, because nobody has published it.
One more thing on the map. The dashed arrows show where I expect the components’ 2031 positions to be. RuPay credit on UPI moves to the edge of commodity, ordinary card credit that happens to ride the rail. The credit rulebook moves into product, carried by India, because NPCI has written one and the BCB and the EPI haven’t. The underwriting practice crosses into product, and the rail’s own data barely moves, because nothing on the record says a rail owner will share it, so the lender’s route to it is the consent layer. If a documented credit decision on a public rail is using Account Aggregator or Open Finance data by 2031, that arrow is right. If not, the fintech’s remaining case goes with it.
The lending balance sheet doesn’t move at all, and that is the point. The loan, the licence and the capital sit on a lender’s balance sheet in 2031 as they do today.
Should the rail owner lend? Write the rulebook, stay off the balance sheet
That leaves the second question, whether the rail owner should be the one lending, and the answer comes in three parts.
For starters, a rail owner that lends becomes a regulated lender sitting inside its own switch, and the China record shows what regulators do about that. The PBoC didn’t ban the credit layer. It forced Ant to move Huabei and Jiebei into a licensed consumer-finance company, with renminbi (RMB) 23 billion ($3.2 billion at about RMB 7.2 to the dollar) of registered capital against a credit book of about RMB 311 billion ($43 billion) by the end of 2024, and made the partner banks carry their own share under their own brands. A central bank that lends on its own rail would have to supervise itself doing it, and the United States Trade Representative already calls the BCB’s “dual role as regulator” and operator a conflict of interest for the payment leg alone.
Secondly, the rail owner’s own stated need runs the other way. Asked about the US tariffs in July, governor Gabriel Galípolo said Pix would remain “gratuito, seguro e aberto”, free, safe and open to every financial institution. Neutrality is what a public rail sells to the banks that have to join it, and a rail owner underwriting credit would have to see what its participants see.
And finally, the one rail owner that does carry credit shows why the others don’t lend. NPCI carries credit on UPI because it owns RuPay, a card scheme, so it can carry the instrument without carrying the loan. The BCB owns no card scheme. The EPI is authorised as a payment institution, so it can’t lend at all, and its shareholders are mostly lenders, so a Wero credit product is really a question of which shareholder’s balance sheet it sits on. Riverty, the Bertelsmann instalments business, drew that conclusion in May when it turned itself into a Luxembourg bank. The one scheme operator in Europe that does lend on its own A2A rail, BLIK in Poland, did PLN 59.4 million ($16 million) of it in 2025 alongside the bank PKO BP, about 0.03% of the value BLIK carried in eCommerce by my arithmetic.
One thing the map doesn’t show but the evidence does is that declining to write the rulebook has a cost, and in Brazil the shopper pays it. A Pix funded from a card limit is a card-statement debt on the shopper’s side, so they expect a card’s right to dispute a purchase that never arrived, but on the merchant’s side it is a Pix, and a Pix has no chargeback. The rail’s only reversal route, the special return mechanism (MED), excludes “conflitos comerciais”, commercial disputes, so the shopper has a card issuer to argue with and no chargeback to invoke. Brazil’s consumer body Idec called the step-back unacceptable and said shoppers were left with “produtos de crédito heterogêneos”, heterogeneous credit products without minimum transparency. That is the gap a rulebook would close, and a rulebook without a balance sheet is the position I’d argue for.
The card networks are renting their craft to the rails
The fair objection to all of this comes in two forms. The first is that what I’ve been calling unnested credit already has a name, and it’s embedded finance. A retailer, a wallet or a ride-hailing app puts a loan inside its own checkout, a licensed lender behind it holds the licence, the balance sheet and the risk, and the brand on the front never touches any of the three. India’s regulator even licenses the arrangement in exactly that shape, as a lending service provider acting for a regulated entity, with the regulated entity’s obligations undiminished. So I’d accept the label, with one correction. Embedded finance describes the tenant version of my claim, the wallet fronting a bank’s line of credit, and that is precisely the version the map fails, because the surplus goes to whoever holds the loan and writes the rules, and the front end holds neither. What a public rail adds is that the front end holds even less than it does on cards. On a card, an embedded brand at least gets a scheme dispute right and a published interchange for its shopper. On Pix and UPI the dispute right doesn’t exist for a credit-funded payment and the interchange is nil or unpublished, so embedded finance on a public rail is the weaker version of the position, not the stronger one. It’s no accident that the lenders who have made it work, Klarna on deposits, Slice with a banking licence and Riverty with one of its own, stopped being embedded and became the lender.
The second objection is that I’ve described buy now, pay later (BNPL) and given it a new name. Two things separate them. Klarna sets its own instalment terms, merchant fee and dispute process, and is regulated as a lender, whereas on UPI the rules a credit line may be used under are NPCI’s and the RBI decides who may fund it at all. And the BNPL leaders are themselves moving to the unnested position from the other side. Klarna funded 95% of its lending from deposits in the year to June 2025 and had 6.5 million active Klarna Card users at the end of June against 1.3 million a year earlier, and Affirm’s card runs on Visa’s Flexible Credential, which lets one credential switch between debit and credit funding.
Affirm’s card is also the clue to what the networks are doing. The story headlines replaying is from EBANX’s “Pix overtakes card dominance” which gives the illusion that card networks are losing ground. Visa’s own annual report instead sets out a plan to “capture value and drive yield” from payments already routed through real-time networks like Pix and UPI, and what it is selling onto the rails is the craft that cards took decades to build. Its fraud product for A2A payments was piloted on Pix with five participants carrying more than 20% of Pix transactions and, by Visa’s own account, scored nearly $500 billion of Pix volume in six months. In the UK, Visa A2A is a rulebook, a liability framework and a dispute service laid over open-banking payments, with the first live transaction completed last November. The network is unbundling its brand, its disputes and its risk scoring from its own rail and renting them to someone else’s. The standing objection, put by Ron van Wezel of Datos Insights, is that cards “do several things” at once and A2A rails can’t replicate the bundle. On these rails the bundle is being sold in parts. The rail moves the money, a licensed lender extends the credit, and the guarantee is being sold separately.
One more component on the map, and it’s the one I’m least sure of. Agentic checkout, where a software agent pays on the shopper’s behalf, sits high on the map and at custom-built. The published protocols I could find, from Google, OpenAI, Stripe, Visa and Mastercard, pay with card credentials or, in Stripe’s newest, with stablecoins as well, and none lets the agent take on credit at the checkout. Stripe’s documentation requires that “the buyer must be” the one navigating a loan application, not the agent. NPCI’s own protocol is expected at Global Fintech Fest in the second week of September, and nothing reported ahead of it mentions a credit line. Whether it carries one is the single fact that would move this map most.
Where the rails end up
Unnested Credit turns out to be less of a Chinese translation peculiarity than a description of where each of these public rails ends up. Credit nested in the payment chain gets unnested, licensed and capitalised, and the rail goes back to being the rail. On the evidence so far, the fintech that wins on a public rail in the next five years is the one that stops trying to own the rail’s credit and becomes a lender that happens to use it. Others saw the contest before I did. Edson Santos of Colink said in June that the main contest of this half-decade in Brazil will be over “crédito no trilho Pix”, credit on the Pix rail. What the map adds is the form the contest takes, and who gets to hold the loan.