💹 FX and multi-currency payments

The four currencies in a payment, and where the margin hides

An international payment can involve up to four currencies, each tied to a step in the chain: the currency of the price the customer sees, the currency of the amount the network carries, the currency the acquirer pays out, and the currency the seller’s treasury receives. Nothing requires these four to match. Whenever two consecutive steps use different currencies, a conversion takes place at a rate that differs from the market rate. That difference pays whoever does the converting. It never shows up as a fee on an invoice, because it is built into the rate itself.

Keeping all four currencies the same is the only setup with no FX cost. A seller that prices, accepts, receives, and books in a single currency makes no conversion at all. As soon as one of the four diverges, an intermediary performs the conversion and takes its cut from the rate applied. Analyzing an FX cost therefore starts with pinpointing the step where the break occurs.

The reference rate: what a margin is measured against

An FX margin is the gap between the rate applied to a conversion and a reference rate chosen to measure it. The result therefore depends on the reference, and comparing two margins requires the same reference on both sides. The mid-market rate is the midpoint between the best bid and the best offer on the interbank market at a given moment. No customer ever gets it, because it is the middle of a spread, not a price anyone quotes. It serves as the yardstick.

The dollar’s dominance has a direct operational consequence. Many currency pairs have no liquid market, because volume is too thin for a direct quote to form. A naira-to-Philippine-peso flow is then handled in two successive legs: naira to dollars, then dollars to pesos. Each leg carries its own spread, and the two spreads add up. A corridor between two thinly traded currencies is therefore structurally more expensive than a corridor of similar volume that runs through the dollar or the euro, whichever provider is chosen.

Double conversion is the second source of extra cost. It happens when the transaction currency matches neither the cardholder’s currency nor the merchant’s settlement currency. A sale priced in US dollars by a European merchant to a Canadian customer can thus go from Canadian dollars to US dollars, then from US dollars to euros. Two spreads stack up on a single sale. The fix is structural: align the transaction currency with the cardholder’s currency or with the settlement currency.

Dynamic currency conversion: mechanics, rules, and abuses

Dynamic currency conversion (DCC) lets a foreign cardholder convert into their card’s currency at the point of sale, instead of having their issuer convert afterward. The terminal, ATM, or payment page shows the converted amount before the cardholder confirms, at a so-called “guaranteed” rate. That rate is set on the acquiring side by a specialist provider. The spread it contains is shared: part of it goes to the merchant or ATM operator as a revenue share. That revenue share, far more than any convenience to the cardholder, explains why the product persists.

• Detection: the first digits of the card number (the BIN) reveal the issuer’s country and currency.
• Offer: the screen offers to charge the card in its own currency, for a fixed amount, next to the amount in local currency.
• Choice: the cardholder accepts or declines. Visa and Mastercard rules require that the choice be the cardholder’s and that it be made before PIN entry.
• Disclosure: the rate applied and the markup must be shown, along with the amount in both currencies, on the screen or the receipt.
• Prohibited: preselecting conversion, choosing on the cardholder’s behalf, or steering the decision through the size or color of the labels.

Surveys by the European Consumer Centres Network and the European Commission put the DCC markup at 4% to 12%, with peaks at ATMs in tourist areas. Conversion by the network and then the issuer typically costs 1.5% to 3% all-in. Comparing the two ranges is simple arithmetic, and in almost every case it leaves no gray area. Declining conversion at the point of sale, in other words paying in local currency, leaves the cardholder with the lower cost of the two.

For the merchant, the revenue share follows its own path and has effects beyond its amount. It arrives in a separate flow from sales settlement, often monthly, and is hard to reconcile with the transactions that generated it. It also fuels complaints, because a cardholder who discovers the markup on their statement disputes the sale itself, not the conversion. Before turning DCC on, weigh three things: the expected revenue share, the cost of handling disputes, and the effect on repeat business.

Multi-currency settlement: PvP, CLS, and orphan currencies

An FX trade has two legs, in two currencies, on two settlement systems, in two time zones. Between the moment the first leg is paid and the moment the second arrives, the counterparty can default, and the institution that has already paid loses the amount of its leg. This is known as Herstatt risk, after the German bank whose closure in 1974 left counterparties paid on one side only. The mechanism that eliminates it is payment versus payment (PvP). It makes the settlement of each leg conditional on the settlement of the other: both legs settle together, or neither does.

CLSSettlement, operated since 2002 by CLS Bank International, is the main PvP settlement system for FX trades, covering 18 currencies. Its legal status matters as much as its technical architecture, because it determines which law governs settlement. CLS Bank International is an Edge Act corporation chartered under US federal law and based in New York, supervised by the Federal Reserve under a cooperative oversight arrangement among central banks. A treasurer who settles yen against Swiss francs through CLS is therefore operating within a US regulatory framework. CLSNet, launched in 2018, covers more than 120 currencies and calculates bilateral net positions without settling them itself. The distinction matters: netting reduces the amounts to transfer, and therefore liquidity needs, but it does not remove settlement risk, which remains until both legs have actually settled.

Messaging and settlement are two separate functions, often conflated in descriptions of cross-border flows. Swift carries payment instructions; settlement happens elsewhere, on correspondent accounts or in a real-time gross settlement (RTGS) system. The coexistence period for cross-border payment formats ended in late November 2025 with the retirement of MT 103 and MT 202 messages. Any integration built on those formats had to migrate to ISO 20022. Currency and charges fields are structured differently there, a frequent cause of reconciliation breaks in the first months of operation.

Collecting locally: local collection accounts and domestic acquiring

A local collection account is an account opened in the customer’s country, into which a provider collects funds in local currency on behalf of merchants based elsewhere. The traditional model did the opposite: each payment crossed the border through a chain of correspondent banks. Today’s dominant model avoids that. The provider opens accounts in every country it serves, collects and pays out locally, and only nets balances between its own entities. Wise built its position on this principle: funds stay inside each country, and only balances are netted. The trade-off is regulatory, since each country served requires a license, a banking partner, or both.

• Who holds the account: the merchant, the provider, or a safeguarding structure. The answer determines what happens if the provider fails.
• Beneficiary name: a virtual account in the provider’s name fails name-matching checks, which are now mandatory on several rails, including euro credit transfers since October 2025.
• Holding the currency: a contract that converts automatically on receipt leaves the merchant no room to choose. The merchant’s right to decide when to convert must be written into the contract, not left to commercial goodwill.
• Conversion reference in reports: rate, timestamp, source amount, and converted amount, for each transaction. Without these four fields, no FX audit is possible.
• Exit terms: how long it takes and what it costs to withdraw balances, and what happens to the funds if the provider loses a local license.

Exchange controls and repatriation: what traps money in-country

Exchange controls are the rules a country uses to restrict the purchase, sale, or transfer of foreign currency within its territory. In some markets, local currency collected can leave the country only on terms set by the central bank. The tools are well known and often combined: requirements to surrender part of export earnings, prior approval for transfers, and multiple exchange rates. On top of these come FX allocation queues and taxes or withholdings on outflows. A collection plan drawn up without reviewing these rules ends with balances stuck in the country that the seller can neither transfer nor use abroad.

Workarounds exist and are in use. Dollar-backed stablecoins serve as a substitute currency in economies with exchange controls or high inflation. Tether’s USDT, whose market capitalization was close to US$183.65 billion at the end of July 2026 according to CoinGecko, dominates this use on the Tron and Ethereum networks. The growth of this channel comes with no comparable legal certainty. USDT does not comply with MiCA and has been delisted from most platforms serving the EU. A regulated institution that uses this channel to get around local FX rules risks losing its license.

Settling in local currencies: rails that skip the dollar detour

Local currency settlement means executing a payment between two countries in their respective currencies, without going through a third currency. Neighboring businesses that trade with each other often take the opposite route: their payment leaves the region, transits through a correspondent in New York or London, and comes back. This detour adds two conversions and two FX spreads, lengthens settlement times, and makes the transaction dependent on a correspondent banking network that has been shrinking since the early 2010s because of de-risking. The BIS Committee on Payments and Market Infrastructures (CPMI) tracks this decline using Swift message data. The responses built over the past decade are regional.

China shows the limits of a rail-by-rail analysis. CIPS settles cross-border renminbi, with 210 direct participants and 1,619 indirect participants as of June 30, 2026, according to its operator. It still relies on Swift for messaging on most of its flows, so it is not a complete alternative. A seller collecting in China meets other layers before reaching that rail, including onshore foreign currency settlement through CDFCPS and merchant gateways such as Alipay+, which connects some 50 national wallets across more than 220 markets.

Measuring the true cost of a currency

The true cost of a currency can be calculated from just three numbers: the amount sent, the amount received, and the market rate at the conversion timestamp. No invoice presents it that way. The stated fee, fixed charges, and “zero fee” claims give only a partial picture, because they leave out the spread built into the rate. International bodies call the full measure the total cost, expressed as a percentage of the amount sent.

Transaction data (from the settlement report)
  source_amount         = 100,000.00 USD
  credited_amount       =  91,240.00 EUR
  fixed_fees_deducted   =      25.00 EUR
  market_rate           =    0.9210 EUR per 1 USD   (same date, same time)

Effective rate obtained
  effective_rate = (credited_amount + fixed_fees) / source_amount
                 = (91,240.00 + 25.00) / 100,000.00
                 = 0.912650

Spread to market (the true FX margin)
  spread = (market_rate - effective_rate) / market_rate
         = (0.9210 - 0.912650) / 0.9210
         = 0.90%

Total cost of the transaction
  total_cost = FX spread + fixed fees as a share of the amount
             = 0.90% + (25.00 / 92,100.00)
             = 0.93%

Takeaway: a contract advertised as fee-free charges 0.90% in the rate.
The only metric comparable across providers is the total cost.

The G20 targets also set a benchmark for retail cross-border payments: a global average cost of no more than 1%, with no corridor above 3%, by the end of 2027. The wholesale segment has no price target; the goal there is for every corridor to offer at least one option for sending and receiving payments. These benchmarks describe a corridor more than a provider. A corridor whose cost reaches 8% owes that level to a combination of an illiquid currency, a long chain of intermediaries, and a cash last mile, not to the quality of the chosen provider.

• Require, for each transaction, the rate applied, its timestamp, the source amount, and the credited amount. Without these four fields, nothing can be measured.
• Rebuild the market rate at the same timestamp from a single, stable source, and keep the series.
• Calculate the spread and total cost on a complete sample, not on the most visible transactions.
• Compare providers on total cost, never on the stated fee.
• Measure the time between the sale and the availability of funds: carrying cost is a cost, even when nobody bills for it.
• Check, corridor by corridor, whether the currency is eligible for PvP settlement and whether a regional local currency rail exists.