The Cost of a Payment: What the Average Really Tells You
One screen splits a payment into the takes on a settlement advice, showing what interchange, the scheme fee and the acquirer margin each took and what reached the business, proved to the paisa against the amount paid. A second works out the average payment in a system. Every charge in a payment chain is set by the Reserve Bank of India and moves, so neither states a charge of its own.
Where does each rupee of one payment land?
A card sale leaves the customer whole and reaches the business short. The gap is taken in named parts on the way, and a settlement advice lists them: interchangeThe part of a card sale that goes to the bank whose customer paid. What interchange is for, and who sets it, is covered separately., a scheme feeWhat the network charges for a sale running over its rails. The scheme fee is a separate take from the acquirer's, and where it goes is covered under card networks., and what the merchant acquirerThe party that stands behind a business's card sales and pays the business what its customers spent. What an acquirer does, and what stands behind it, is covered separately. keeps for standing behind the sale. Entering an advice into the boxes makes the screen do the one thing an advice never does: it sets the four takes and the amount paid out as a single piece of arithmetic that has to close.
One settlement advice, split four ways, proved against the amount paid
The first four boxes come off one advice and need nothing else. The last three come off the merchant agreement and are what lets the same schedule be tested at a different payment size. Every box opens on an invented advice belonging to Pushkar Tiffin Counter so that nothing is blank, and every box can be overwritten.
Rs 300.00/- is the payment being split, the payment on the advice, exactly
| Where it lands | Struck on the amount | Struck once for the payment | Its whole take | Share of the payment |
|---|---|---|---|---|
| Interchange | Rs 3.60/- | Rs 0.90/- | Rs 4.50/- | 1.50 per cent |
| Scheme fee | Rs 0.60/- | Rs 0.45/- | Rs 1.05/- | 0.35 per cent |
| Acquirer margin | Rs 1.20/- | Rs 0.30/- | Rs 1.50/- | 0.50 per cent |
| Taken on the way | Rs 5.40/- | Rs 1.65/- | Rs 7.05/- | 2.35 per cent |
| Reached the counter | not struck | not struck | Rs 292.95/- | 97.65 per cent |
| Amount the customer paid | — | — | Rs 300.00/- | 100.00 per cent |
Interchange at Rs 4.50/-, the scheme fee at Rs 1.05/-, the acquirer margin at Rs 1.50/- and the counter's own Rs 292.95/- come to Rs 300.00/-, against a payment of Rs 300.00/-. The parts and the payment agree to the paisa.
Pushkar Tiffin Counter is paid Rs 300.00/- on the card. Interchange takes Rs 4.50/-, the scheme fee Rs 1.05/- and the acquirer margin Rs 1.50/-, so Rs 7.05/- is taken on the way and Rs 292.95/- reaches the counter. That take is 2.35 per cent of the payment, and Rs 1.65/- of it was struck once for the payment rather than on the amount.
No change yet. Every box holds the invented advice as it was read.
Educational illustration. The seven boxes open on an invented advice written for teaching. Not one figure in them is a rate anybody charges: every charge in a payment chain, with any floor and any ceiling on it, is set by the Reserve Bank of India at rbi.org.in, and the account to account rails are described by the National Payments Corporation of India (NPCI) at npci.org.in. Every box can be overwritten with the figures on an actual advice. Nothing is kept: the numbers entered live in this tab and die with it.
Read the split off the screen once and it holds without the screen. Pushkar Tiffin Counter is paid Rs 300.00/- on a card. Interchange takes Rs 4.50/-, the scheme fee Rs 1.05/- and the acquirer margin Rs 1.50/-. The three takes come to Rs 7.05/-, the counter keeps Rs 292.95/-, and Rs 4.50/- plus Rs 1.05/- plus Rs 1.50/- plus Rs 292.95/- is Rs 300.00/-. The customer paid exactly that. The take is 2.35 per cent of that payment, and Rs 1.65/- of the Rs 7.05/- was struck once for the payment rather than on the amount. The flat Rs 1.65/- is what makes an unchanged schedule behave differently on a different sale. Hold the schedule and change only the sale: on Rs 30.00/- it takes Rs 2.19/-, or 7.30 per cent; on Rs 3,000.00/- it takes Rs 55.65/-, or 1.86 per cent; on Rs 1.50/- it takes Rs 1.68/-, which is 112.00 per cent of the sale, and the counter ends up Rs 0.18/- short for having been paid at all.
An advice shows Rs 4.50/- of interchange, Rs 1.05/- of scheme fee and Rs 1.50/- of acquirer margin on a sale of Rs 300.00/-. The identical schedule now meets a sale of Rs 30.00/-. What happens to the share of the sale that is taken?
Two kinds of cost sit inside every payment arrangement, and they behave in opposite directions. One kind lands once for each payment: it is the same amount whether the payment carried thirty rupees or thirty lakh, so read against the money that actually moved it is crushing on a small payment and almost invisible on a large one. The other kind lands on each rupee moved: it is the same proportion of every payment, so it does not care about size at all.
Which of those two kinds dominates a system is decided almost entirely by one number, and that number is the average payment. This is why a screen that computes an average is a cost screen even though an average is not a cost. The weight of the first kind cannot be got at without knowing what a payment typically carries, and the average is the only handle a reader has on that when a record gives them a total and a count and nothing else.
Everything below is worked on Setu Payments Limited, an invented payment system. Setu Payments reported two figures in its stated year and no others: 1,200 crore payments, and Rs 3,60,000 crore of value carried on them. Two numbers, one division, and a very large amount of thinking falls out of the answer.
A screen called the cost of a payment asks for a value carried and a number of payments. What can those two inputs actually produce between them?
What does this screen work out, and what does it refuse to work out?
A screen that hands over a confident number and explains its limits afterwards has already done the damage, so the boundary comes first. The two instruments do exactly this much and no more.
The split works out where one payment landed, from four figures off a single advice for one card instructionA message telling a bank to move money. An instruction is a separate event from the money actually moving, and that separation is worked through in full elsewhere., and then, once the flat part of each take is added, how the same schedule behaves on a payment of a different size. The panel further down works out the average payment in a system from a value carried and a count, and how heavily anything struck once for each payment falls per rupee moved, given as an index rather than as any sum of money.
Neither instrument states a charge. There is no rate of interchange here, no figure for the share of a sale that a business does not get to keep, no floor and no ceiling on any of it, and nothing about what a payment system spends. Every take in the boxes is one entered off an actual document, and the advice they open on was invented for teaching. Every charge, floor and ceiling is set by the Reserve Bank of India at rbi.org.in, with the account to account rails described by NPCI at npci.org.in, and every one of them moves, so a calculator that carried a level would give a wrong answer wrapped in a confident interface.
A blank invites a reader to go and look. A filled cell invites them to stop looking. A stale number on a screen is worse than no number at all.
Why does cost thinking start with an average that is not a cost?
Look at what a payment chain actually does when one payment goes through it. A message is composed and sent. Checks are run before it is accepted. A record is written, and then kept. Somebody may raise a query about that record eleven months later, and it has to be there to answer with. Somewhere underneath, an obligation between two banks is created and later discharged in settlementThe moment an obligation between two banks is actually discharged, as opposed to merely recorded. How that is arranged is covered separately and is assumed here..
Now ask how many of those things happen once for each rupee. None of them. All of them happen once for each payment. The message is one message whether it carries Rs 30.00/- or Rs 30,00,000/-. The record is one row. The query is one query. Almost everything a payment chain does is counted in payments. Almost everything a payment chain carries is counted in rupees. The average payment is the exchange rate between those two currencies of measurement.
The money moved therefore ends up in the denominatorThe number underneath in a division, the one the top number is being shared out over. Change it and the answer changes even when the top number has not moved at all. of everything: a fixed amount for each payment, read against the money that payment moved, is a constant divided by the average payment. The average decides whether the work done on a payment is trivial against the value it carried or comparable to it, and nothing else in the record decides that.
The everyday version of this is not an analogy at all but the same arithmetic in smaller clothes. A courier charges a flat delivery fee. On a grocery order of Rs 200/- that fee is a serious fraction of the amount spent. On a furniture order of Rs 40,000/- the identical fee is a rounding error. The courier did not change its charge between the two deliveries and did not need to. The order size did all the work, and the order size is the household's average payment.
Why does the average payment matter for cost at all, when the average payment is plainly not itself a cost?
Can the same average payment be two completely different pieces of news?
Setu Payments reached an average of Rs 300.00/- by carrying Rs 3,60,000 crore on 1,200 crore payments. Suppose that average moved to Rs 3,000.00/-. There are two ways to get there and the arithmetic cannot tell them apart.
The first route: the value carried rises tenfold to Rs 36,00,000 crore while the count holds at 1,200 crore payments. Rs 36,00,000 crore over 1,200 crore payments is Rs 3,000.00/-. The second route: the value holds at Rs 3,60,000 crore while the count falls tenfold to 120 crore payments. Rs 3,60,000 crore over 120 crore payments is also Rs 3,000.00/-. Same answer, and a reader who arrives at it twice by different paths and is told nothing will assume one of the paths is wrong.
Neither is wrong. An average is a quotientThe answer a division produces. Whatever it is called on the screen, a quotient is one number shared out over another. Two different changes can therefore move it by exactly the same amount., and multiplying the top by ten does exactly what dividing the bottom by ten does. The arithmetic is identical and the two events are opposite. More payments carrying the same money is a system doing more work for the same purpose. More money on the same payments is a system doing the same work for a larger purpose.
The distinction matters more than it looks. An average payment quoted on its own, without the two figures that produced it, is a number nobody can act on. When one is handed over alone, the only useful next question is which of the two inputs moved, and the honest answer to that question sits in a record rather than in a ratio.
Two systems both report an average payment of Rs 3,000.00/-. One got there by carrying ten times the money on the same number of payments. The other carried the same money on a tenth of the payments. Is that the same news?
Which costs land once for each payment, and which land on each rupee moved?
Sorting a cost into one of these two lists settles how it behaves before anything else about it is known.
Landing once for each payment: the message that carries the instruction; the checks run before it is accepted; the record written when it is accepted; the storage of that record for as long as it has to be kept; the query somebody raises about it later and the work of answering that query; and the share of capacity that payment consumed at the peakThe busiest moment a system is built to survive, counted in payments arriving at once rather than in rupees. Capacity is sized for that moment and paid for all year.. The share of peak capacity is easy to miss and it is not small. Capacity is bought for the worst minute of the year and paid for through every quiet one.
Landing on each rupee moved: the money that has to be available to discharge what is owed, and anything else whose size simply is the size of the amount. The second list is short, and being short is the whole finding.
Six items against two is why a payment system's cost tracks its payment count rather than its value, and why halving the average payment is much worse news for such a system than halving the value it carries. Halve the value with the count unchanged and the second list halves while the first list does not move at all. Halve the average by doubling the count instead and the long list doubles. The two changes look symmetrical in a ratio and are nothing like symmetrical in a cost base.
The same shape appears wherever work is counted in items and revenue is counted in rupees. Ten shops in one mall share one security guard and one power connection, so what each shop pays for the shared work depends on how much each of them sells, not on how much floor space the mall has. A merchant acquirer reading its own book faces exactly this question about the businesses it serves.
Sort three items. The message sent about a payment, the record kept of it, and the money that has to be available to discharge what is owed. Which sorting is correct?
The average payment falls from Rs 300.00/- to Rs 30.00/-, and something struck once for each payment has not changed by a single paisa. Per rupee moved, what happens to it?
How is an index read, and why is one used here instead of an amount?
The second panel's other reading is a comparison rather than a sum of money. Anything struck once for each payment is set at an index of 100.0 at the reported year, when the average payment was Rs 300.00/-, and the arithmetic is one line: take Rs 300.00/-, share it out over whatever the average payment currently is, and multiply by a hundred. At an average of Rs 30.00/- that gives 1,000.0. At Rs 3,000.00/- it gives 10.0. Nothing about the thing being weighed changed at any of those points. Only the money underneath it changed.
The charge itself is not stated, so the output is an index rather than an amount. How much heavier the same unnamed charge has become is what can honestly be given instead. That turns out to be more useful than it sounds, because how much heavier is exactly the question a business asks, and the answer does not require knowing the level at all.
Take the stall outside one office building that takes card payments over lunch. Whoever runs it does not know what will be charged on those sales next year, and cannot find out today. The stallholder can still work out, this afternoon, that if their average sale halves then whatever is struck once for each sale becomes twice as heavy per rupee they take. A real planning conclusion has been drawn from a charge nobody has stated, and the index is the shape of it.
The index is read against the reported year and never as a rupee figure. An index of 250.0 does not mean Rs 250/-, does not mean 250 per cent of anything a business pays, and does not mean a charge was raised. An index of 250.0 means the average payment has fallen to two fifths of what it was, so an unchanged fixed amount is now spread over two fifths as much money. Every reading of it carries that label beside it for exactly that reason.
The second reading comes back at an index of 10.0. What does that establish, and what does it not?
What is the reading if no control is touched at all?
Every figure the panel below produces at its declared settings is written out here as ordinary text, so a reader who never moves anything leaves with the whole of it.
At the reported year, Setu Payments Limited carried Rs 3,60,000 crore of value on 1,200 crore payments. Rs 3,60,000 crore shared out over 1,200 crore payments is Rs 300.00/-, and the index on anything struck once for each payment is set at 100.0 there by definition. The reading to keep is one line: an average payment of Rs 300.00/- and an index of 100.0, both computed from two reported figures and neither of them a charge.
Then the four declared endpoints. Hold the count at 1,200 crore payments and drop the value to Rs 36,000 crore, and the average payment is Rs 30.00/- with an index of 1,000.0. Hold the count and raise the value to Rs 36,00,000 crore, and the average is Rs 3,000.00/- with an index of 10.0. Now hold the value at Rs 3,60,000 crore instead and drop the count to 120 crore payments: the average is Rs 3,000.00/- again, index 10.0 again. Raise the count to 12,000 crore payments and the average is Rs 30.00/-, index 1,000.0. The two routes land on the same two answers, and the two ways of reaching one average arrive here as a table.
| Value carried | Payments | Average payment | Index |
|---|---|---|---|
| Rs 3,60,000 crore | 1,200 crore | Rs 300.00/- | 100.0 |
| Rs 36,000 crore | 1,200 crore | Rs 30.00/- | 1,000.0 |
| Rs 36,00,000 crore | 1,200 crore | Rs 3,000.00/- | 10.0 |
| Rs 3,60,000 crore | 120 crore | Rs 3,000.00/- | 10.0 |
| Rs 3,60,000 crore | 12,000 crore | Rs 30.00/- | 1,000.0 |
The shaded row is the reported year and the four below it are declared settings of the controls rather than anything that happened. The index column is a ratio to the shaded row and carries no unit of money at all.
Move the value, move the count, and watch which one did the work
Two inputs and two outputs make this a calculator rather than a control on a single relationship. The average payment is the value carried shared out over the number of payments. The index is Rs 300.00/- shared out over the current average payment, multiplied by a hundred. At the reported year it therefore reads 100.0. Both formulas are printed here rather than hidden behind the boxes. The two controls open at the reported year.
Rs 3,60,000 crore of value carried
1,200 crore payments
Setu Payments Limited carries Rs 3,60,000 crore of value on 1,200 crore payments, so the average payment is Rs 300.00/-. Anything struck once for each payment therefore reads an index of 100.0 per rupee moved, which is exactly what it weighed at the reported year. The index is a ratio to that year and is not a rupee amount.
Educational illustration. Invented payment system, two reported figures and nothing else, control settings at every other point, and an index rather than an amount. Every charge, floor and ceiling is set by the Reserve Bank of India and moves, so none of them is computed or implied at any setting of either control. The output is labelled the average payment and never the cost.
Why is the answer this screen gives not a price?
Rs 300.00/- is the average amount a payment moved. The figure is not what a payment costs anybody, not what anybody was charged, and not what the system spent to handle one. The panel gives what a payment carries, and the cost of a payment is a different number altogether.
The two numbers are not a precise version and a rough version of the same thing. The two numbers are about different things. One is money that moved from a payer to a payee and belongs to neither the system nor the parties in the chain. The other is money that leaves somebody to pay for the work of moving it. A payment carrying Rs 300.00/- has moved Rs 300.00/-, and every rupee of that Rs 300.00/- arrives at the far end.
Keep the labels in mind and the confusion cannot form. One output is labelled the average payment and the other a ratio to the reported year. Neither label contains the word cost, and neither sits beside a currency figure that could be mistaken for one.
The screen returns Rs 300.00/-. Finish the sentence correctly. Rs 300.00/- is the average amount a payment ...
The failure: an answer read as the thing the title names
The failure here is reading Rs 300.00/- as what a payment costs, and the interface is a large part of the cause rather than an innocent bystander. Somebody arrives at a screen called the cost of a payment. The visitor types in a value and a count. A confident number comes back. The sentence that forms in their head is that a payment costs Rs 300.00/-, and nothing on a badly built screen would stop it forming.
The number is the value carried shared out over the number of payments. The quotient describes money that moved between two other parties and contains no information whatever about what anybody was charged for moving it. The two readings do not differ in precision. The two readings differ in subject.
Who makes this reading: anybody, including people being careful. The cost of the mistake: a figure quoted onward as a cost, with every judgement built on top of it inheriting the error. The worst version is a comparison. Set two systems side by side on their averages, conclude one is cheaper than the other, and the comparison is between two numbers that differ for reasons with nothing to do with cost. A system moving wages and a system moving lunch bills will have very different averages and neither of them is a price.
The fix is built into the tool rather than written underneath it. The output is labelled the average payment on the screen. The index is labelled a ratio to the reported year at every reading. And what a payment costs is set by the Reserve Bank of India and read at rbi.org.in.
How does somebody running a stall actually use any of this?
Ten minutes with a till roll, before anybody quotes anybody anything
The routine works for a stall outside one office building, for a shop with a card machine, and for an analyst reading a payment system's annual figures. The steps are the same at every scale.
First, the two numbers: total value taken over a period, and the number of separate payments that carried it. A till roll has both. Dividing one by the other gives the average payment. The average payment belongs to the business and nobody has to supply it.
Second, the costs sort into the two lists, with one settlement advice filling the first three rows: anything struck once for each sale goes left, anything whose size is the size of the sale goes right. As a payment system finds, the left list turns out longer than expected and the peak sits in it.
Third, and this is the step the index exists for, what happens to the left list if the average sale halves can be worked out without knowing a single charge: whatever is struck once for each sale becomes twice as heavy per rupee taken, and that conclusion is available today from a level nobody has disclosed. A lunchtime stall whose average sale falls from Rs 300/- to Rs 150/- has doubled the weight of everything on the left list, and it can plan around that before any schedule is published.
The routine will not settle which way of taking money to use, whether what a business is charged is reasonable, or whether any route was obvious. None of those are arithmetic questions. Nobody who has been surprised by a charge failed to do something they should have done.
What is not settled here, and where does the answer sit?
Three things sit just beyond the edge of these two screens, and naming them as a list makes the refusal a route rather than a gap.
The first is the share of a sale a business does not get to keep, and every floor and every ceilingThe upper limit placed on something by whoever is entitled to place one. placed on it. The second is what must be disclosed to a business, and to a person, about what they are being charged. The third is what a payment system may charge the participants inside it, on card rails and on the account to account rails alike. All three are set by the Reserve Bank of India, the rails themselves are described by NPCI, all three move, and all three are drawn below as rows with nothing in them.
A calculator that carried a level would go wrong quietly on the morning that level changed, and this one cannot. The blank rows are a feature of the tool rather than an admission about it. An empty cell carrying a label and an address is really a question with directions attached to it. A row filled in last year and left sitting there quietly stops being true, and a reader would believe it anyway.
Three conditions and where each of them is set
| What is set | The value here | Who sets it |
|---|---|---|
| Every charge in a payment chain, together with any floor and any ceiling placed on it | Not stated here | Reserve Bank of India at rbi.org.in |
| What must be disclosed to a business, and to a person, about what they are charged | Not stated here | Reserve Bank of India at rbi.org.in |
| What a payment system may charge the participants inside it, on card rails and on the account to account rails alike | Not stated here | Reserve Bank of India at rbi.org.in, with the rails themselves described by NPCI at npci.org.in |
Copy the middle column out and go and complete it from the address on its right. Empty, this table still does work: which limits exist, and whose signature sits under them, changes far more slowly than the limits themselves do.
Who sets every charge in a payment chain?
The three conditions are drawn again below, under a working calculator.
A reader who has just computed something is in exactly the frame of mind to believe the next number they see. The next thing below is therefore a blank with an address in it. The blank is not modesty. A blank is the only design that survives the day a level changes, and that design decision is the teaching as much as the arithmetic above it is.
The habit worth keeping from all of this: naming one thing these screens will never give, and saying where the answer would be sought.
Where the three blank rows on these screens get filled in
| What this guide points at instead of stating | Who settles it | Site | Checked |
|---|---|---|---|
| Every charge in a payment chain, together with any floor and any ceiling placed on it | Reserve Bank of India | rbi.org.in | 25 August 2026 |
| What has to be disclosed to a business, and to a person, about what they are charged | Reserve Bank of India | rbi.org.in | 25 August 2026 |
| What a payment system may charge the participants inside it, and the rails it runs them on | Reserve Bank of India, and NPCI on the rails | rbi.org.in, npci.org.in | 25 August 2026 |
Pushkar Tiffin Counter and Setu Payments Limited are invented.
Educational material. Not advice on any investment, tax, budget or market position.
