Most Americans assume that the rest of the world’s payment systems are like our own, to the extent they think about them at all. But in fact, the US system is unusual, both in the sheer dominance of card networks and in its rules that shift fraud risk away from consumers and towards merchants and banks.

These differences are easiest to see when payments go wrong. To illustrate, let’s (hypothetically) buy and then refund one thing, three times: a pair of fake Air Jordan 1s, $185 on a Visa card in San Francisco, ¥1,399 by QR code in Hangzhou, and R$1,299 by Pix in São Paulo.

The street vendor rings the Jordans up on their Square terminal, turns it around, and has us tap our card. They get the little green checkmark about two seconds later and start bagging the shoes.

In those two seconds the terminal asked Square; Square asked Visa; Visa asked Chase; and Chase looked at our credit limit, ran a fraud check, and put a $185 hold on our account. Then Chase told Visa it was good for it, Visa told Square, and Square told the merchant. No money has moved yet.

That night, Square submits its processed transactions into the card network for clearing. Visa validates the transactions, calculates what each party owes, groups the transactions by bank, and adds up the results. Instead of millions of individual payments moving between banks, each participating bank ends up with a single net amount it owes or is owed.

On the next business day, Chase, along with every other bank that owes money that day, wires its net position to Visa’s settlement bank. Visa instructs its settlement bank to disburse to the net creditors, one of which is Square’s sponsor bank. Both transfers occur over Fedwire, the Real Time Gross Settlement (RTGS) system that moves actual money between banks’ accounts at the Federal Reserve.

The vendor sees $180.04 of the $185 land a day or two later. The missing $4.96 (2.6% + $0.15) is split three ways: roughly $3.99 to Chase as interchange, about $0.25 to Visa, and $0.72 ($0.57 + $0.15) to Square.

We buy the same shoes from a stall in Hangzhou. Instead of using a terminal, we scan an Alipay QR code, enter ¥1,399, and confirm with Face ID. A speaker behind the counter announces the payment out loud, and the vendor starts bagging the shoes.

Unlike with the Visa transaction, no authorization request is sent through a card network to our bank. Alipay merely reduces our balance in their system by ¥1,399, increases the vendor’s by ¥1,390.60, and keeps ¥8.40 (0.6%) as the fee. The vendor can spend that Alipay balance immediately.

If the payment is funded from our bank account, Alipay still needs to collect the underlying ¥1,399. Before 2018, it would have done so by talking to our bank directly. Alipay and WeChat Pay each maintained a web of bilateral connections to dozens of banks that regulators had limited visibility into.

As one might expect, the Communist Party decided that this arrangement was 不好 (no good) and required third-party payment companies to route these transactions through a new state-controlled central clearing network called NetsUnion. Instead of Alipay talking directly to our bank, the payment instruction now goes from Alipay to NetsUnion to the bank, giving the central bank visibility into the flow.

We buy one more pair from a street vendor in São Paulo, once again by scanning a QR code, albeit this time from inside our regular bank app. We see the vendor’s registered name, type in R$1,299, and receive a confirmation a few seconds later.

Our bank sends the payment into SPI, the Banco Central do Brasil’s RTGS system (like Fedwire in the US). The central bank debits our bank’s account and credits the vendor’s bank’s account immediately. SPI settles each payment individually, rather than collecting transactions and netting them later.

By the time the vendor starts bagging the shoes, the Brazilian vendor’s bank has already received the money in its account at the central bank.

This speed is a relatively recent phenomenon in Brazil. Before the Banco Central launched Pix in 2020, merchants accepting credit cards typically had to wait around thirty days to receive their money. This was because Brazil’s card system was structured around the consumer’s monthly credit-card bill, meaning that merchants, rather than credit card issuers, were providing the float to the consumer.

Pix is generally free for individuals, and while businesses can be charged by their banks, these rates (0.22% on average) are far lower than card acceptance fees (see the 2.6% from the US purchase). Large financial institutions are required to participate in the system by the Banco Central, even though they make considerably more money when their customers pay by credit card.

A week later, and we've humiliated ourselves three times in public. For one pair, the stitching gives out. For another, we notice the logo is slightly too caked-up. For the third, a friend gently informs us that 'Air Jodan' is not a Nike sub-brand.

Enraged, we open the Chase app, tap “dispute” on the transaction, and fill out a brief form. A day or two later, Chase provisionally puts $185 back on our account, and from our perspective, the refund is complete.

Behind the scenes, Chase files a chargeback through Visa’s dispute system. Visa pulls $185 back out of Square’s sponsor bank, and Square debits it from the vendor’s balance. If the vendor was using a processor other than Square, they’d likely also be hit with a $10 to $25 dispute fee.

The vendor gets seven days to fight back with evidence, like proof the shoes were real, or a signed return policy. In case you ever wonder why DoorDash requires the driver to take a photo of your order once it has been dropped off, this is why.

Merchants do not contest most disputes, as assembling the packet often costs more than the merchandise itself did. In this case, they don’t bother because they know they would lose regardless since they’re selling counterfeits.

If you stop to consider what just happened, you’ll realize that we just used a privatized small-claims court where the burden of proof largely falls on the defendant.

This “court” has additional punishments for repeat offenders. If more than ~1% of a merchant’s transactions are disputed, the merchant ends up in Visa’s monitoring programs, which mean fines and may eventually lead to losing the ability to accept cards, which is a death sentence for a US retailer.

As one may imagine, this guilty-until-proven-innocent bias results in unjustified chargebacks, which are referred to as “friendly fraud” in the industry. Data varies widely across different sources, with claims that anywhere from 20-35% of all chargebacks are fraudulent.

Back in Hangzhou, we open Alipay looking for the equivalent of a chargeback, but there isn’t one. Alipay can handle complaints against merchants, but it cannot simply reverse a completed payment because we tell it the shoes were fake. We go back to the stall and ask for a refund, but the vendor refuses.

We report the merchant through Alipay, which can suspend merchants that violate its rules, but that doesn’t give us our money back.

For the refund itself, we submit the transaction record and photos of the shoes to 12315, China’s consumer complaint system. Our complaint is routed to the local market-regulation authority, who can contact the merchant and mediate the dispute, with the possibility of separate penalties for selling counterfeit goods. If the vendor agrees to refund us, they send ¥1,399 back through Alipay as a new payment. The original payment is never reversed.

Chinese consumer law also gives buyers stronger remedies when a merchant knowingly commits fraud, including damages above the purchase price. Those rules have helped create a class of professional counterfeit buyers, known as 职业打假人 (zhíyè dǎjiǎrén), who deliberately buy suspected fakes and then pursue compensation.

If we had made the purchase on a marketplace such as Taobao, we may not have needed to turn to the government. Marketplaces control much more of the transaction, as they know what was listed, retain the buyer-seller messages, and can hold funds until the transaction is complete.

Back in São Paulo, we open our bank app again, looking for an equivalent to a chargeback, but don’t find one. Pix does have a dispute mechanism, called MED, but it doesn’t apply here. MED is for fraud, scams, coercion, and other cases where the payment itself was fraudulent. We knowingly sent R$1,299 to the right person and received the wrong product, meaning that as far as Pix is concerned, the payment worked as intended.

We go back to the vendor and ask for our money back, but they refuse. From here, like in China, the dispute leaves the payment system.

In Brazil, the next stop is Procon, the government consumer-protection system. We file a complaint with the São Paulo office and attach the Pix receipt and photos of the shoes. The dispute is now about Brazilian consumer law, not payment rules.

If that doesn’t work, we take the vendor to a Juizado Especial Cível, Brazil’s small-claims court. Since our R$1,299 claim is below the threshold at which a lawyer is required, we bring the case ourselves.

Eventually the vendor gives in and sends us R$1,299 by Pix. Just as in China, the original payment was never reversed. The refund is a new payment going in the opposite direction.

Why does the American vendor pay 2.6% when the Brazilian one pays 0.22%? Because the latter is mostly paying to move money, whereas the former is paying into a system that bundles four other products and services into the same card swipe.

The first product is a $185 micro-loan. Our ¥1,399 and R$1,299 left our accounts within seconds, but there was a delay of several days to several weeks between when Chase paid the merchant and when we paid our next credit card bill.

The second product is a fraud/dispute insurance policy. Both the payouts (chargebacks) and the premium (interchange) are ultimately funded by the merchant.

The third product is credit card rewards. If our card gives us 2% back, that money has to come from somewhere. Merchants recover card-acceptance costs through higher prices, and because most stores charge the same price regardless of how you pay (surcharges for card-users were mostly forbidden until recently), customers using cash or cheaper payment methods help subsidize rewards-card users.

The fourth product is credit history. FICO scores are built from debt, not payments, meaning if you paid all of your bills via bank transfer, you’d be invisible to the mortgage market, no matter how responsible you were.

The American system most favors the consumer who pays off their credit card bill every month, uses their rewards, and isn’t afraid to hit the chargeback button in their bank portal. Each of these benefits is paid by another party, be they Chase, the merchant, or other consumers that don’t pay by card. It least favors merchants, who pay higher fees and receive fraudulent chargebacks that aren’t worth the time or money to dispute.

American merchant-side payment fees of 2.6% are eye-popping in comparison to Brazil’s 0.22%, but this is apples-to-oranges. What we should really care about is the total economic burden of each system, inclusive of fraud and dispute management.

In 2024, US merchants paid about $187 billion in card fees on $11.9 trillion of purchases. That money funds the aforementioned loans, rewards, and dispute court. Much of this consists of transfers between participants (e.g., rewards) and doesn’t contribute to the total economic burden. Cardholders contested $9.8 billion of charges that year and got $5.9 billion charged back, a sizable fraction of which was friendly fraud.

China and Brazil have payment rails with much lower fees, but the lack of chargebacks means that the consumer pays for fraud (by getting scammed or ripped off) instead of the merchant. Brazilians lost around R$5 billion (~$1b) to Pix scams in 2024, of which the fraud mechanism recovered about 9%. Chinese citizens lost ¥35 billion ($5.1b) to telecom and online scams in 2020 alone. All of this is facilitated by irreversible payments.

By my rough math, the economic burden of the US payments system is around $55–80 billion a year all-in, about 0.19–0.27% of GDP, China’s is around $15–40 billion (0.07–0.20%), and Brazil’s is around $6–9 billion (0.27–0.43%).

This means that despite Pix charging merchants one twelfth as much as US credit cards, the US system appears to be cheaper than Brazil’s because Brazil runs both Pix and cards side-by-side. While by these numbers China has the cheapest system, it's also the one we have the least accurate numbers for, as Beijing hasn’t published fraud loss statistics since 2020.

What we’re really measuring when we estimate the total cost of a payment system is the cost of trust in that society. This cost appears to be on roughly the same order of magnitude wherever you look, roughly a few tenths of a percent of GDP.

Fraud, disputes, enforcement, and credit risk all exist regardless of the payment architecture. A country can only decide what form those costs take, and who gets stuck with them. In the US, more of the cost materializes as merchant fees and higher prices; in China and Brazil, more of it falls directly on scam victims or the state.

You can learn a lot about a country by looking at its payment system. America sanctifies the consumer, so our system protects them, even at the expense of the merchant; China desires state control, so it put the government inside the payment system via NetsUnion; and Brazil treats payments infrastructure as a public good, so its central bank built the rail and left the rest to the courts.

Thanks for getting to the end! This took maybe 2-3x as much time and effort to write as the last two, but I learned a ton. I’m not 100% sure what the next piece will be on, but I want to branch out from payments into less career-focused topics like metaethics or Deltarune.