Haute Lumière
Commerce · III.09 · MMXXVI · daylight
TEN CONCEPT BRIEFS · Chapter III.09 — Digital Settlement Without Extraction
One page each. A reader who reads only these ten pages has the chapter.
BRIEF 1 — Interchange
The idea. Interchange is a fee the merchant's bank pays the cardholder's bank on every card sale. It is not a fee for work done. It is a transfer between two banks, set by the card network that sits between them, and it is the largest component of what a merchant pays to accept a card.
Nobody at the till chose it. The merchant did not negotiate it — the network publishes a rate card and the merchant's acquirer passes it through. The cardholder never sees it, because it is charged on the merchant's side.
Worked example. A $50 basket, paid by an American credit card in 2025, at a combined Visa and Mastercard credit interchange rate of 2.36 per cent, carries $1.18 of interchange. The same basket in the European Union, where Regulation (EU) 2015/751 caps credit interchange at 0.3 per cent, carries $0.15 — 7.87 times less, on the same networks, the same silicon and the same message format. The whole difference is regulatory.
The number to carry. American merchants paid $187.20 billion in card processing fees in 2024, and $198.25 billion in 2025 — growth of 5.90 per cent in one year.
You already know this because you have seen a shop with a "card minimum" sign and understood instinctively that something about small card payments does not work for the shopkeeper. That sign is interchange, written in marker pen.
BRIEF 2 — Ad valorem versus per-message pricing
The idea. A payment fee can be charged as a percentage of the amount (ad valorem) or as a fixed amount per transaction (per message). The choice looks technical. It decides almost everything.
A payment does not become harder to make as the number attached to it grows. The same messages are exchanged, the same fraud checks run, the same record is stored. So a percentage fee has no engineering basis — it is a claim on value rather than a price for work, and it silently indexes the payment industry's revenue to the price of housing, fuel and food.
Worked example. Take a representative European card rate of 1.8 per cent and a representative flat scheme fee of €0.29.
basket € 40 ad valorem € 0.72 flat € 0.29 = 0.725 % of basket
basket € 4,000 ad valorem € 72.00 flat € 0.29 = 0.007 % of basketThe ad valorem fee is 100 times larger on the big basket. The flat fee is identical. On the €4,000 basket the percentage fee is 248.3 times the flat one, and nothing about the payment was 248 times more difficult.
The trade-off, honestly. A flat fee is regressive for very small payments — €0.29 on a €2 coffee is brutal. The workable answer is a small flat fee with a floor exemption, or a capped percentage, which is what India adopted: 0.4 per cent above ₹2,000, capped at ₹300, so the cap bites at ₹75,000.
You already know this because you have never once been charged more to send a heavier email.
BRIEF 3 — The wedge: settlement cost against settlement price
The idea. The cost of settling a payment and the price charged for settling it are two different numbers, and until somebody publishes the first one, the second one looks like a fact of nature.
Brazil published the first one. The Banco Central do Brasil charges participating institutions R$0.01 for every ten credits settled through the instant payments system — a tenth of a centavo. Against Pix's average ticket of R$443.11, that is 0.000226 per cent of value.
Worked example. Three numbers, in ascending order:
Banco Central's settlement charge 0.000226 % of value
what Pix costs a Brazilian merchant 0.22 %
Brazilian credit card merchant discount 2.2 %The middle number is 975 times the first. The top number is 9,748 times the first.
Read the middle ratio first, because it is the honest one. The gap between 0.000226 per cent and 0.22 per cent is real work: fraud screening, disputes, onboarding, support, the app, the refund desk. Nearly all of the 0.22 per cent is somebody doing something useful. Which is exactly what makes the top ratio hard to defend — if the entire apparatus of a modern payment business fits inside 0.22 per cent, the other 1.98 per cent is not apparatus.
You already know this because you have compared the price of an airport bottle of water with the price of the same bottle in a supermarket, and understood that the difference was not in the water.
BRIEF 4 — Finality
The idea. Finality is the moment a payment stops being reversible. It is the only question a treasurer actually asks about a payment rail, and the four answers available are genuinely different.
- Bitcoin — probabilistic. Six confirmations at ten minutes a block is 60 minutes, and even then it is a probability, not a guarantee.
- Ethereum, proof-of-stake — deterministic, at two epochs of thirty-two twelve-second slots: 768 seconds, or 12.8 minutes. Bitcoin takes 4.69 times as long, and offers less.
- Card — authorised in seconds, settled in days, and reversible for months through chargeback.
- Pix, UPI, Faster Payments — seconds, and irrevocable, because the payment settles in central-bank or settlement-bank money.
Worked example, and it is a corrective. What is an hour of waiting actually worth? On a $10,000,000 payment at a 5 per cent annual funding cost, one hour of float costs $57.08. Ethereum's 12.8 minutes costs $12.18.
Those numbers are small, and saying so is the point. The case for instant finality is operational, not financial. It is about counterparty risk, about reconciliation staff, about whether the goods can leave the warehouse — not about the interest on the float.
You already know this because you have waited for a bank transfer to clear before handing over keys, and the problem was never the interest.
BRIEF 5 — The two-sided market, and why interchange exists at all
The idea. A card network serves two groups who need each other: cardholders and merchants. Jean-Charles Rochet and Jean Tirole's work on two-sided markets showed that in such a market the level of total fees and the structure — who pays which side — are separate decisions, and that a platform can rationally charge one side heavily to subsidise the other.
That is the honest case for interchange, and it should be stated before it is answered. Interchange charges the merchant to subsidise the cardholder, which recruits cardholders, which makes acceptance worth more to the merchant. For a network starting from nothing, this works.
Where the argument stops working. It is a launch argument. Once a network carries the overwhelming majority of a country's payments, the subsidy is no longer recruiting anybody — everybody already has the card. At that point the structure persists not because it builds the network but because it funds the rewards, and the rewards are the reason the cardholder does not switch.
Worked example. Cards carried 64 per cent of all United Kingdom payments in 2025 and American card volume was $11.92 trillion in 2024. Neither market needs recruiting.
You already know this because you have kept a loyalty card long after you stopped choosing the shop for the points.
BRIEF 6 — The regressive transfer
The idea. Merchants recover acceptance fees in the shelf price, because almost none of them price differently by payment method. So the fee is paid by every customer, and the rewards funded by the fee go only to the customers with the rewarded cards. That is a transfer, and it runs uphill.
Worked example. Schuh, Shy and Stavins, at the Federal Reserve Bank of Boston, calibrated it. The average cash-using household pays $149 a year to card-using households. The average card-using household receives $1,133 — a ratio of 7.60.
Sorted by income, it is sharper. The household earning under $20,000 a year pays $21. The household earning over $150,000 receives $750 — a top-to-bottom spread of $771 a year, moving from the poorer household to the richer one, collected at the till of the shop where both of them buy.
The reframe. The merchant discount rate is not a price for moving money. It is a reverse-means-tested transfer with a payment network as the collection agent. That is why it survives: the people best placed to object are the people being paid.
And for scale: $187.20 billion of American card fees in 2024, across 131.4 million households, is $1,424.66 per household per year — a number on nobody's statement, because it is in the shelf price.
You already know this because you have watched somebody pay for a flight with points and not wondered who bought them.
BRIEF 7 — The instant payment rail as a public utility
The idea. An instant payment rail run as public infrastructure is defined by three properties, and all three are testable from published documents: it settles in the safest money available, it recovers its costs rather than earning a margin, and it returns what it over-collects.
The third is the one that proves the other two.
Worked example. In March 2026, Pay.UK returned £27.5 million to its participants because it had collected more than it needed. Across the 5.5 billion Faster Payments made in 2025, that over-collection was half a penny per payment. A commercial rail does not return an over-collection. It books it as profit and its shareholders applaud.
The cost ceiling. Pay.UK's whole cost base in 2025 — Faster Payments, Bacs, the Image Clearing System and the standards work together — was £161.1 million. Attribute all of it to Faster Payments alone, which overstates it substantially, and one instant payment costs at most £0.0293, under three pence, or 0.00336 per cent of the £4.80 trillion moved. For scale, the American card fee rate of 1.57 per cent is about 468 times that — different countries, so read it as a scale comparison rather than a price.
You already know this because you have received a water rates rebate and understood, without being told, what kind of company sends one.
BRIEF 8 — Acceptance, and who pays to build it
The idea. A rail priced at cost has no salesforce. Nobody's commission depends on your shop signing up, so nobody knocks on your door. Acceptance is the one thing extraction was genuinely buying, and a design that removes the extraction without replacing the acceptance function does not get adopted.
Worked example. Three national rails, three answers.
rail launched measured years share mechanism
------------------------------------------------------------------
FPS 2008 2025 17 11.1 % neither mandate nor commission
UPI 2016 FY2025 9 83.4 % zero MDR plus state subsidy
Pix 2020 2025 5 54.7 % mandatory participationShare gained per year: United Kingdom 0.65 points, India 9.27, Brazil 10.94 — 16.8 times and 14.2 times faster where somebody bought the acceptance. (Different denominators in each country, so read it as orders of magnitude, not a league table.)
What India's speed cost. 1,389 crore rupees in the year to March 2022, then 2,210, then 3,631, then an outlay of 1,500 — 8,730 crore, about $992 million — and even the largest of those years covered only 18.2 per cent of the industry's estimated 20,000 crore annual cost of running UPI. On 15 October 2026, zero MDR ended.
You already know this because you have seen an excellent free tool nobody uses, sitting beside a mediocre paid one with a salesperson attached.
BRIEF 9 — Energy per transaction, honestly
The idea. "Energy per transaction" is the most quoted and least careful figure in payments. It is worth getting right in both directions, because a number that overstates the case gets the whole argument dismissed.
The figures. Bitcoin's network draws about 138 TWh a year, which at 1,100 kWh a transaction implies 125.5 million transactions annually — 3.98 per second. Ethereum after the Merge uses 35 Wh a transaction and about 2,600 MWh a year, a cut of 99.988 per cent with no change to what the ledger does. Visa is about 1.5 Wh. So Bitcoin is 733,333 times Visa per transaction and 31,429 times Ethereum, and Ethereum is 23.33 times Visa.
The scale test. Run Pix's 2025 volume of 79.8 billion transactions at Bitcoin's per-transaction energy and you need 87,780 TWh a year — 2.93 times all the electricity the world generates.
The correction, and it matters. Proof-of-work energy is spent per block, not per transaction. One more payment in a block costs almost no extra energy, so 1,100 kWh is an average and not a marginal cost, and batching or a payment channel moves it a long way. The figure without that problem is the security budget: 3.125 coins per block × 144 blocks × 365 = 164,250 coins a year, which at a round $100,000 is $16.43 billion a year, or $130.92 per transaction — paid by holders as dilution rather than by payers as a fee, which is why it is on nobody's receipt.
You already know this because you have seen a per-head cost quoted for a building that would have been heated anyway.
BRIEF 10 — Fee-to-profit: the merchant's decision number
The idea. A payment fee expressed as a percentage of revenue sounds small. The same fee expressed as a percentage of net profit is the number that moves a board, and it is one division away.
( MDR_old − MDR_new ) × revenue × share migrated
------------------------------------------------
net profitWorked example. A merchant with R$1,000,000 of card-borne revenue at a 4.0 per cent net margin earns R$40,000. The card fee at 2.2 per cent is R$22,000; the same volume on a rail at 0.22 per cent costs R$2,200. The saving is R$19,800 — 49.5 per cent of net profit. At half migration, 24.8 per cent. At a 3 per cent margin and full migration, 66.0 per cent.
The threshold worth carrying. R$19,800 on R$1,000,000 of revenue is 1.98 per cent. So at any net margin below 1.98 per cent, the payment fee is larger than the entire profit of the business. That is not rhetoric; it is a division, and for a great many food retailers it is the correct side of the line.
At corporate scale. A firm with $240.0 million of revenue, 70 per cent of it card-borne, at a 6.0 per cent net margin, pays $2,637,600 a year in acceptance. Migrating 30 per cent of that volume from 1.57 per cent to 0.22 per cent saves $680,400 — 4.72 per cent of net profit in year one, and $2,268,000, or 15.75 per cent, at full migration.
You already know this because you have watched a small recurring cost destroy a thin-margin business while a large one-off cost did nothing at all.