The bill of exchange: paying elsewhere without shipping metal
A bill of exchange is a written order in which a banker instructs a correspondent in another trading city to pay a set sum to a named payee. It solved a problem of long-distance trade: distance made shipping coin expensive, slow (it took weeks), and vulnerable to theft. With a bill, a debt could be settled without a single coin changing places. A merchant paid funds to a banker in one city and received a written order instructing a correspondent in another city to pay out, in another currency, while the metal stayed put. Only the document traveled from one city to the other.
Two later inventions turned the bill into a circulating instrument. Endorsement let the payee transfer the bill with a simple signature on the back, so one piece of paper could discharge several debts before it ever reached the drawee. Discounting turned it into cash before maturity, in exchange for interest kept by whoever advanced the funds. A claim on a third party thus became transferable from hand to hand and convertible into money before it fell due. In other words, it circulated as a payment instrument.
Giro banks were the period’s second contribution. The Amsterdam Wisselbank, founded in 1609, kept merchants’ accounts and settled their debts to one another by book entry on its own ledgers, without a single coin changing hands. Settlement took place in bank money, that is, in the liability of a third party that everyone involved accepted as final. T2, Fedwire, and BOJ-NET work on the same principle today, applied nationwide and executed in seconds.
1950–1979: from cardboard cards to the four-party network
Diners Club was founded in New York in February 1950. Department stores already issued cards, but each one worked only at the store that issued it. The Diners Club card, by contrast, was accepted at merchants independent of one another, and cardholders were billed monthly. A single third party stood between merchant and buyer, guaranteed payment, and took a commission. American Express still uses this three-party setup today.
The defining invention of the period was the license the network granted its member banks. A bank in Denver could not sign a contract with every merchant in the country. So the network sold it the right to issue cards under a shared brand, on condition that it honor transactions from other members’ cardholders. The four-party model, also called four-corner, grew out of that constraint. It brings together cardholder, issuer, merchant, and acquirer, four parties bound by a single set of rules, with no direct contract between the two banks.
| Three-party model | Four-party model | |
|---|---|---|
| Examples | American Express (direct), Diners Club | Visa, Mastercard, UnionPay, CB, girocard, Interac |
| Who issues and who acquires | The same company, on both sides | Two separate banks, connected through the scheme |
| Interchange | None: a single all-in fee, negotiated with the network | Paid by the acquirer to the issuer, often capped by regulators |
| Effect on cost | One fee, historically higher | Cost split into interchange, scheme fees, and acquirer margin |
| Negotiation | With the network itself | With the acquirer, over its margin only; the rest is a fixed cost of entry |
| Regulatory reach | Often exempt from interchange caps | Directly targeted by caps (EU IFR, US Regulation II) |
The magnetic stripe makes the card machine-readable
The magnetic stripe is a recording strip on the back of the card that holds the account number and related data in machine-readable form. It replaced cards that carried only printed information, which merchants had to accept by hand: an imprint taken with a manual imprinter, a phone call above a floor limit, and carbon copies mailed to the bank. Machine reading changed what a card was. Because stripe data could be sent over a network, a terminal could query the issuer before accepting a transaction, which made online authorization possible.
International standards followed, and they still hold. ISO/IEC 7810 sets the card’s physical dimensions, ISO/IEC 7811 the magnetic recording techniques, ISO/IEC 7813 the content of tracks 1 and 2, and ISO/IEC 7816 the integrated circuit interface. These four standards are why a card issued in Lagos works in an ATM in Osaka. The physical format they define has not changed in 50 years.
;4970100000001234=28122010000012345678?
; start sentinel
4970100000001234 PAN, up to 19 digits
= field separator
2812 expiration YYMM (December 2028)
201 service code (3 digits)
2 = international interchange, chip present
0 = authorization per issuer rules
1 = no restrictions, PIN not required
0000012345678 issuer discretionary data
? end sentinel (followed by the LRC check character)
40 characters maximum. No element changes from one
transaction to the next: that is the medium's core weakness.The ATM arrived in the same wave. Barclays opened one in Enfield, near London, on June 27, 1967. It worked on the same logic as a merchant terminal: a machine that reads a card, queries an authorization system, and dispenses value. In almost every country, cash withdrawals were automated before in-store payments. Most national switches also started out with ATM networks, including 1LINK in Pakistan, BancNet in the Philippines, Prosa in Mexico, and Multibanco in Portugal.
The chip, from a French patent to a global standard
A chip card carries an integrated circuit that can run a program, whereas a magnetic stripe can only play back the characters recorded on it. Roland Moreno patented a secure memory card in 1974, and Michel Ugon at Bull added a microprocessor in 1977. What sets the chip apart from the stripe is that processing power: it stores data and also computes new data every time it is used. For each transaction, it generates a cryptogram based on the amount, an internal counter, and a random number supplied by the terminal. The cryptogram changes from one transaction to the next, so a recording of a past transaction can no longer be used to create a new one.
France served as a nationwide testing ground. In 1984, French banks created the Groupement des Cartes Bancaires CB, France’s domestic card scheme, which required full interbank interoperability: one card worked at every participating merchant and every ATM, whatever the issuing bank. Chip and PIN became universal in 1992. France remains the first major market to have made the switch, and card-present fraud there collapsed earlier than anywhere else. The scheme is still running, with 77 million cards and 14.5 billion transactions in 2024, according to GIE CB.
Chips went global through EMV, named for Europay, Mastercard, and Visa, whose specification was published in 1996. EMVCo, set up in 1999 to govern it, now has six members: Visa, Mastercard, American Express, Discover, JCB, and UnionPay. Rollout happened country by country, almost always through the same lever, the liability shift, which puts fraud losses on whichever party has not invested in chip. The US only made the switch on October 1, 2015, nearly two decades after the specification was published.
| Instrument | Data presented | What it prevents | What it does not prevent |
|---|---|---|---|
| Paper imprint | Embossed PAN, signature | Nothing technical: visual check only | Replay, stolen cards, insufficient funds |
| Magnetic stripe | Static track data | Keying errors; enables online authorization | Cloning: a copied track is indistinguishable from the original |
| EMV chip | Dynamic cryptogram for each transaction | Cloning and replay at the point of sale | Remote fraud, where the chip plays no part |
| Token (tokenization) | Substitute number tied to a device or a merchant | Use of a PAN stolen from a database | Social engineering and account takeover |
The results of the migration show up in how fraud is split across channels. The chip did not eliminate fraud. It displaced it. Once in-person payments became hard to attack, fraud volumes moved to card-not-present sales, where no chip communicates with a terminal. Every market that migrated saw the same shift, a few quarters later, and the next wave of work (online cardholder authentication, tokenization, behavioral analytics) followed directly from it.
Electronic transfers, from overnight batches to messaging
An automated clearing house is the infrastructure where banks submit batches of payment files, an operator calculates net positions, and settlement happens at fixed times. Bulk payments grew up on this model, separately from cards and at far higher values. Payroll, pensions, and direct debits run through it. Batch processing means holding payments until they can be exchanged together, so the time between the payment order and settlement depends on how often the cycles run, not on how fast the computers are.
Three pioneering systems show three different approaches to the same service. Bacs, launched in the UK in 1968, set a three-day cycle the market has never shaken off. The US ACH network, created in 1972, runs under a single rulebook issued by Nacha but has two competing operators, FedACH on the central bank side and EPN on The Clearing House side, a setup rarely seen elsewhere. Japan’s Zengin System, launched in 1973, opted for near-immediate processing four decades before anyone spoke of “instant payments.”
Cross-border interbank payments need a common language, so that a French bank and a Japanese bank read the same messages the same way. Swift was founded in 1973 as a cooperative under Belgian law, and its network went live in 1977. It only carries standardized instructions, which means Swift settles nothing. The money moves elsewhere: through nostro and vostro accounts, in an RTGS system, or in CLS. More than 11,500 institutions in over 200 countries are connected to it.
| Card type | Settlement | Availability | What breaks |
|---|---|---|---|
| Batch clearing (Bacs, ACH, Elixir, SLIPS) | Deferred net, one to three times a day | Business days, strict submission windows | Missing a cutoff costs a full value day |
| Gross settlement (Fedwire, T2, CHAPS, BOJ-NET) | Gross, one payment at a time, in central bank money | Business days, market hours | Intraday liquidity: every payment must be funded |
| Instant (Pix, UPI, FPS, SCT Inst, FedNow) | Immediate and irrevocable, 24/7/365 | Always on, including weekends and holidays | Irrevocability: no recall once executed |
1974: Herstatt and the invention of gross settlement
On June 26, 1974, in the middle of the business day, German authorities revoked the banking license of Bankhaus I. D. Herstatt. Its counterparties had already paid in their deutsche marks in Frankfurt and were waiting for dollars in New York, where the trading day was not yet over. The dollars never came. The time gap between two settlement systems had turned an ordinary foreign exchange trade into a total loss.
Foreign exchange settlement risk has been known as Herstatt risk ever since: the loss suffered by a party that has delivered its currency without receiving the other side. The failure reshaped payment system oversight worldwide. G10 central bank governors set up the Basel Committee at the end of that same year. Two principles then became standard. The first is to settle gross, one payment at a time, in central bank money, instead of letting exposures build up until the end of the day. The second requires both legs of an FX trade to settle simultaneously or not at all.
This protection remains partial. Daily FX turnover reached $9.6 trillion in April 2025, up 28% in three years, according to the Bank for International Settlements (BIS) Triennial Survey published on September 30, 2025. CLS covers 18 currencies. All the others, including most emerging market currencies, still settle bilaterally without that safeguard. The BIS estimated that $2.2 trillion was still exposed to settlement risk on a given day in April 2022.
M-PESA, payments without banks
M-PESA is a money transfer and payment service tied to a mobile phone line, launched by Safaricom in Kenya in 2007. Three local conditions explain its uptake: few bank branches, dense mobile coverage, and urban workers sending money home to rural families. It requires no bank account, no smartphone, and no card. A USSD menu on a basic feature phone is all it takes. The rail that came to dominate the entire country is thus run by a telecom operator, outside both the banking sector and the public sector.
The model has spread across the continent, with variations from one operator to the next. MTN Mobile Money launched in 2009 and covers about 15 markets. Airtel Money followed in 2011 in 14 markets. Orange Money, launched in 2008, leads in the CFA franc zone, through e-money subsidiaries licensed country by country by the BCEAO and BEAC, the West and Central African central banks. Wave arrived in 2018 with far lower prices, charging 1% on transfers and nothing for deposits and withdrawals, which forced Orange to match it in Senegal and Côte d’Ivoire.
Regulators have converged on two requirements. The first separates the e-money issuer from the telecom operator that controls it. Mobile Money Limited, a subsidiary of MTN Ghana, holds its own license from the Bank of Ghana, and handled GHS 4.1 trillion in transactions with a float of GHS 38.4 billion in 2025. The second requires interoperability between competing wallets. GIMACPAY enforces it across the six CEMAC countries under BEAC Instruction 001/GR/2018. Tanzania made it mandatory in TIPS, which its central bank operates directly. That system processed 651 million transactions in 2025, up from 453 million in 2024.
Public instant payments, pioneered in the UK and scaled in Brazil and India
An instant payment rail executes an account-to-account transfer in seconds, at any hour, and irrevocably. Two systems led the way without being widely copied at first. SPEI, operated by the Banco de México since 2004, settles in seconds at very low cost. The UK’s Faster Payments Service, launched in 2008 and now run by Pay.UK, was the first 24/7 rail deployed at scale in Europe. The designers of India’s UPI, Brazil’s Pix, and TIPS openly drew on it. FPS processed 5.55 billion transactions worth £4,838 billion in 2025.
Brazil’s Pix, launched in November 2020, added something its predecessors lacked: a regulatory mandate to participate. The Banco Central do Brasil operates it through the SPI infrastructure, and every institution with more than 500,000 accounts must join. The service is free for individuals, payments are addressed by key (CPF, CNPJ, phone number, email, or random key), and the EMVCo QR code format is mandatory. Five years on, the rail handled 79.8 billion transactions worth R$35.36 trillion in 2025, accounted for 54.7% of retail transactions in the second half of the year, and reached about 175 million users.
India’s Unified Payments Interface, launched in 2016 by NPCI under a Reserve Bank of India mandate, took a different path: opening the rail to third parties. A nonbank app can initiate payments on the rail as a third-party application provider (TPAP), with public APIs and full interoperability between banks and apps. Fiscal year 2025–2026 closed with 241.62 billion transactions, up 30% by volume. Two extensions broaden the rail’s reach. UPI 123PAY brings it to feature phones, and RuPay Credit on UPI links a credit card to the merchant QR code, with no terminal needed.
| Rail | Operator | Since | Addressing | What drives adoption |
|---|---|---|---|---|
| Pix (Brazil) | Banco Central do Brasil | 2020 | Key: CPF/CNPJ, phone, email, random | Mandatory participation above 500,000 accounts; free for individuals |
| UPI (India) | NPCI, under RBI mandate | 2016 | Virtual payment address (VPA), mobile number | APIs open to third-party apps; no fees on P2P |
| FPS (UK) | Pay.UK, operated by Vocalink | 2008 | Account number and sort code | First mover, near-universal bank coverage |
| SCT Inst (SEPA area) | EPC scheme, settled through TIPS and RT1 | 2017 | IBAN, with verification of payee | Regulation (EU) 2024/886: receiving by January 9, 2025, sending by October 9, 2025, price parity |
| FedNow (US) | Federal Reserve Banks | 2023 | Account number and routing number | Open to any institution eligible for Fed services |
| PromptPay (Thailand) | National ITMX, mandated by the Bank of Thailand | 2017 | Mobile number, national ID, tax ID | Mandatory free transfers below a cap, which ended paid P2P |
The US is set up the opposite way, with two competing rails and no regulatory mandate to participate. The Clearing House’s RTP network, launched in 2017, topped $1.3 trillion in 2025, up from $246 billion in 2024, after its transaction limit was raised to $10 million. FedNow, launched in 2023, processed $853.4 billion in 2025, with an average transaction of $101,435. An average ticket that size points to corporate treasury and business-to-business (B2B) payments rather than person-to-person (P2P) transfers.
QR codes, mandated interoperability, and rail interconnection
QR code payments replace the terminal with a graphic code read by a phone: the displayed or printed image carries the payee’s details, which the payer’s app turns into a transfer order. China showed that a country could skip the payment terminal entirely. Alipay launched in 2004 on the back of Alibaba, and Tenpay and WeChat Pay in 2005 on the back of the messaging app. A code printed on a sheet of A4 paper is enough to accept payments, with no hardware, no acquiring contract, and no maintenance. The two companies now hold about 54% and 42% of China’s mobile payment market, according to an OECD background note from June 2025.
Chinese regulators then pulled these flows back into a single clearing house. NetsUnion Clearing Corporation was set up in 2017, and since June 2018, nonbank online payments have had to go through it instead of connecting directly to banks. Super-app flows became visible to the central bank and subject to its regulation again. QR interoperability among Alipay, WeChat Pay, and UnionPay followed between 2021 and 2023.
Southeast Asia scaled QR payments through standards rather than through one dominant company. QRIS, mandated by Bank Indonesia with the payment industry association ASPI in 2019, requires every wallet and every bank to read a single code. The central bank reports 50.50 million users and 32.71 million enrolled merchants, and an NFC extension, QRIS Tap, has rounded out the system since 2025. DuitNow QR, run by PayNet in Malaysia since 2019, applies the same rule and supports four cross-border links. The underlying rail is still an instant credit transfer: DuitNow since 2018, PromptPay and PayNow since 2017, the latter built on Singapore’s FAST infrastructure, launched in 2014.
The current phase is about linking these rails directly, without going through correspondent banks. The UPI-PayNow link, launched in 2023, was the first direct connection between two national instant payment systems for person-to-person remittances. It has 19 participating Indian banks since 13 more joined in July 2025. Nexus Global Payments, a nonprofit set up in Singapore on March 26, 2025, by six central banks, extends the idea with a hub-and-spoke model, in which a single connection gives access to every member. The project is not yet live. In Africa, PAPSS settles in local currencies with daily netting and covers 28 countries since the BEAC joined in July 2026, but it does not publish its volumes.