Nobody can quote you a real number, including us
Every comparison of payment costs that opens with a table of percentages is either out of date or was never accurate. Card rates depend on the card, the issuing country and your negotiated terms. Mobile money rates depend on volume and change. Network fees change by the hour with congestion. Any figure you read in an article is a starting point for a question, not an answer.
So this guide does something more useful than quoting numbers: it gives you the structure of each cost, and a method for filling in your own figures from quotes you actually hold. Do that once and you will make a better decision than any published comparison can give you.
The four layers of payment cost
Every rail has some combination of the same four costs, and providers compete by making one of them invisible.
The percentage. A share of the transaction value. Dominant on cards and mobile money, and the number most merchants think of as the fee.
The fixed component. A flat amount per transaction, which punishes small baskets. A twenty-shilling fixed fee is nothing on a ten-thousand-shilling order and brutal on a two-hundred-shilling one.
The currency spread. The margin taken when money changes currency, buried in the exchange rate rather than shown as a fee. Cross-border card payments and any conversion between KES and USD are where this hides.
The loss and labour layer. Chargebacks you lose, payments you never reconcile, orders abandoned because the checkout did not fit the customer, and staff time spent matching payments by hand. This layer is invisible on every rate card and is frequently the largest of the four.
Where the cost sits on each rail
Cards. A percentage plus a fixed fee, with the percentage rising for cross-border and premium cards. Add the currency spread if the card is not in your settlement currency. Then add chargebacks, which are not only the disputed amount but the fee to process the dispute and the staff time to fight it. Cards also carry an acceptance cost that never appears on any invoice: a Kenyan consumer without a card that works online cannot buy from you at all.
Mobile money. Familiar to essentially every Kenyan buyer, quick to confirm, and priced as a percentage that is usually competitive. The trap is not the fee, it is the labour layer. If M-Pesa payments arrive outside your store and a human matches them to orders, the true cost per order includes that person's time, plus the orders that get shipped twice or not at all. Properly integrated, mobile money is often the cheapest rail you have. Handled manually, it can be the most expensive.
Stablecoin transfers. The buyer pays a network fee to move the tokens, and on the networks E-Sarif watches that fee is typically a fraction of a dollar and, importantly, does not scale with the amount sent. The percentage the merchant pays is a matter of your commercial terms. What genuinely changes the arithmetic is the loss layer: settled transfers cannot be charged back, so an entire category of loss goes to zero. The costs that replace it are customer education and the occasional wrong-network incident.
A method for working out your own number
Take last month, not a hypothetical. Pull your actual order data and run each rail through the same five lines.
One: your real average order value, and the shape of the distribution. If half your orders are small, the fixed fee matters more than the percentage.
Two: the percentage and fixed fee from a written quote, per rail. Not from a website. Ask.
Three: any conversion cost, expressed as the gap between the rate you are given and the rate you can see quoted elsewhere on the same day.
Four: your loss rate. Chargebacks and disputes as a share of revenue for cards. Unreconciled or duplicated payments for manual mobile money. Wrong-network and support incidents for stablecoins, which you will have to estimate at first.
Five: the labour. Hours per week spent on payment reconciliation, multiplied by a realistic hourly cost, divided by orders. This line is why properly integrated payments pay for themselves, and it is the one merchants leave out.
Add the five lines per rail and divide by orders. That number is the cost per order, and it is frequently ranked differently from the percentages on the rate cards.
The cost that dwarfs all of them
Before optimising fees, look at your conversion rate. A checkout that offers a Kenyan shopper only a card form loses a meaningful share of buyers who would have paid by M-Pesa without a second thought. Those are not fees; they are the whole order, gone.
Run the arithmetic yourself with your own numbers. On any realistic order value, recovering even a small percentage of abandoned checkouts is worth more than shaving a fraction of a percent off your processing rate. This is why the useful question is almost never which rail is cheapest, but which set of rails covers every customer you have.
That is also why offering more than one method tends to beat optimising one. M-Pesa for local buyers, stablecoin for cross-border ones, and a card gateway if you need unattended renewals. Each covers customers the others lose.
What to ask a provider before you sign
Ask for the percentage and the fixed fee in writing, including the cross-border case. Ask what the settlement timing is and in which currency. Ask how the conversion rate is set, and what the spread is. Ask who bears the cost of a failed or reversed payment. Ask what a refund costs you.
Then ask the questions that reveal the labour layer: does the integration mark my orders paid automatically, can I reconcile against my own order references, and can I refund from my own admin. A provider that is cheap per transaction and leaves you matching payments by hand is not cheap.