Platform commission: where it comes out, and who holds the processor fee
In short
A platform's application fee is not a percentage but a deduction at a named point in the money flow: the charge arrives, the processor takes its fee, the platform takes its commission, and the seller is credited what is left. Where that deduction happens decides what your ledger can answer once the order reverses.
Key takeaways
- One charge produces four amounts, and each must be booked against a named party to reconcile.
- The percentage is a commercial decision; the point of deduction is an engineering one.
- Who holds the processor's fee is decided by your charge model, not by your commission rate.
- Most processors do not return their own fee on a refund — confirm this against your processor's current terms.
The phrase application fee comes from payment platform APIs — Stripe's Connect documentation uses application_fee on charges created on a connected account and application_fee_amount on charges created on the platform. Read as a number it is uninteresting. Read as a movement it is the whole design: at some instant an amount stops belonging to the seller and starts belonging to the platform, and every question about refunds and reconciliation is a question about that instant.
What the percentage should be is a commercial decision for the business running the market. Modelling it so a rate change is not a migration is argued in modelling commission so it can change later.
One charge, four amounts, four parties
| Amount | Value | Booked against | Who decides it |
|---|---|---|---|
| Charge total | 10,000 | The buyer | Listing price plus whatever the platform adds |
| Processor fee | 290 | Whichever account settles the charge | The processor's contract with you |
| Platform commission | 1,500 | The platform | The platform's commercial terms with sellers |
| Seller net | 8,210 or 8,500 | The seller | Falls out of the three above, and out of who holds the processor fee |
The seller net has two possible values from identical inputs, and the difference is not commission. It is whether the 290 came out of the seller's share or the platform's. That is the number sellers feel, and the one most platforms cannot explain, because it was settled implicitly inside a payments integration months earlier.
Deducted where the charge lands, or deducted when the money moves
There are two moments commission can be taken: the charge arrives already split, or the full amount lands in one balance and a later movement credits the seller. Which model you use is a rails decision, compared in charge then transfer, or pay the seller directly. What matters here is what each leaves for the ledger.
| Question | Deducted at the charge | Deducted at the transfer |
|---|---|---|
| Commission recognised | At settlement, atomically with the charge | When the transfer is made, possibly days later |
| The seller's statement | One line: gross, fee, net | Two lines that must be joined to make sense |
| Cancelled before release | Already recognised, so reverse it | Nothing to reverse — just do not transfer |
| Daily reconciliation | Charge, fee and payout per order | Charges plus a transfer stream keyed to orders |
Deducting at the transfer is more forgiving because the amount is computed later, when you know whether the buyer complained. That is why platforms holding funds before release end up here — though holding money is not escrow, as escrow is usually a delayed payout sets out.
The refund, where somebody has to be out of pocket
Reverse a payment and three amounts move in different directions. The buyer gets the charge total back. Whether commission returns to the seller is a policy decision with its own article — who eats the commission on a refund. The processor's fee is the one people forget: many processors do not return it on a refund, so it stays gone and somebody's balance is short. Confirm this against your processor's terms, because it varies by processor and country.
- Decide the incidence in writing before launch. Platform absorbs, seller absorbs, or shared — all three are defensible, and no answer is not.
- Book the absorbed amount as its own ledger line. Processor fee borne by platform is measurable; a quietly smaller commission is a mystery.
- Model a dispute separately from a refund. A dispute adds a fee on top of the reversed amount, landing on whichever account is liable.
- Show the seller their net. The gap between what a seller sees and what the processor will pay is covered in two balances a seller can see.
The commission percentage is a negotiation. Who holds the processor's fee when an order reverses is an architecture decision, and it is made once.
Take rate and application fee are not the same measurement
The application fee is what you deducted on one transaction. Take rate is what the platform retained across all of them, after refunds, absorbed processor fees, discounts, chargebacks and goodwill credits. The gap is the cost of everything that went wrong, expressed as a percentage, and a platform that cannot compute both from its ledger is not measuring its margin. That needs the deduction reproducible by a job every day, matched against what the processor says happened. Where reconciliation finds a difference it should stop and escalate rather than adjust its way to agreement — the discipline in AI agents in production, and the work we do under AI agents and automation.
Sellers ask what they will actually receive long before they ask about your rate, and one who cannot see a worked net during onboarding stalls — a delay measured by the activation clock between signup and first live listing. When the money arrives is a separate promise, kept or broken by the payout schedule. This silo sits under split payments and the money ledger, in our marketplace practice.
Frequently asked questions
Short answers to the follow-ups this page tends to raise.
What is an application fee in marketplace payments?
It is the platform's commission on a payment it facilitated, deducted as part of the money movement rather than invoiced afterwards. The term comes from payment platform APIs, where the platform names an amount to retain from a charge made on a seller's behalf. That instant of deduction, not the percentage, decides how refunds behave.
Who pays the payment processor's fee in a marketplace, the platform or the seller?
Whichever account settles the charge, unless you decided otherwise and implemented that decision. Where the charge is created on the platform, the platform's balance carries the processing fee and normally the dispute amount; where the charge sits on the seller's account, the seller's balance does. Write the choice into the seller terms.
Is take rate the same as the commission percentage?
No. Commission is what you deducted on a transaction; take rate is what you retained across all of them after refunds, absorbed processor fees, discounts and chargebacks. The gap is the cost of everything that reversed, and it moves while a headline rate never does.
- commission
- split payments
- ledger
- payouts
The work behind this page
Builds from our portfolio that this page draws on.
Read next
- A handful of sellers are taking everything: is it the ranker or the market?Concentration is only a defect if your ranker built it. Re-rank the same queries with every history-derived signal removed, and the share distribution tells you which answer you have.diagnostic
- Acceptance rate: what it measures, and when it measures the platformAccepted over offered sounds like a measure of a seller. The platform chooses the denominator, so an uncorrected acceptance rate is partly a measure of your own targeting.definition
- Availability: a set of intervals, not a grid of day cellsAvailability is not a stored fact. It is the answer to a question, computed from recurring rules, exceptions and what has already been consumed — and a day-cell table is a cache of that answer.definition
- Available in search, gone at checkout: locating the stale layerRe-ask the source of truth at the moment of failure. If it agrees with the index, your two paths apply different rules; if it disagrees, something is holding a stale copy.diagnostic
- Dispatch radius: a travel-time boundary that only looks like a circleThe circle on the map is a proxy. The real boundary is however far a provider can travel in an acceptable time, which is different in every direction and at every hour.definition
- Inventory hold: the row that exists so the second buyer is refusedA hold is a record, not a screen state. It names one unit of supply, one claimant, one expiry and one reason — and while it lives, nothing else may overlap that unit.definition
Working on something in this space?
Tell us where you are in a sentence or two. We'll tell you honestly whether we're the right team, and what a sensible first slice of the work looks like.
Start the conversation