Marketplace Unit Economics: Who Earns the Margin

Marketplace volume is not your margin

Marketplace unit economics asks one useful question. After one completed transaction, who gets the money and what does the platform keep after the costs caused by that transaction?

Start with the transaction, not the market size. Trace what the buyer pays, what the seller receives, what the platform charges, and what it spends to make that order happen safely. Then repeat the work by buyer cohort and seller cohort. A marketplace can move a large amount of money and still lose money on the activity that creates it.

This is where gross marketplace volume becomes dangerous. It looks like scale. It is not revenue, contribution, or cash available to run the company. It is the top of a bridge. The economics live underneath it.

Keep the operating view separate from the accounts

The operating model and the financial statements answer different questions.

The operating model asks what changes when another order completes. The accounts decide how revenue and costs must be reported under the applicable standards and contracts.

That distinction matters for a marketplace. The IFRS Foundation's review of IFRS 15 explains that principal versus agent treatment turns on the nature of the promise and whether the company controls the good or service before transfer. An agent generally recognizes its fee or commission as revenue. A principal generally recognizes the gross consideration for the promised good or service.

Do not settle that question in a spreadsheet. Contract terms, control, responsibility, inventory risk, and pricing discretion can change the answer. A qualified accountant should decide the reporting treatment. A lawyer should confirm what the contracts actually make the platform responsible for.

For operating work, use neutral labels such as transaction value, platform fee pool, and order contribution. Reconcile them to the accounts after the model is useful.

Build three ledgers for one order

A two-sided marketplace needs three ledgers.

The buyer ledger shows the item or service price, buyer fees, taxes, credits, refunds, and final cash paid.

The seller ledger shows the transaction value, platform charges, adjustments, refunds, penalties, and final payout.

The platform ledger shows the fees it is entitled to keep, then every variable cost required to acquire, complete, protect, and support the order.

If the ledgers do not reconcile, stop. A take rate calculated from one side and a payout calculated from the other can hide money in the gap. Taxes, tips, shipping, discounts, and refunds are common reasons. None belongs in the model by accident.

Public marketplace metrics make the same point. Etsy's annual filing for 2025 defines gross merchandise sales as item value excluding shipping and net of refunds. It states that this measure is not Etsy revenue. Uber's annual filing for 2025 defines Gross Bookings differently. Its definition includes applicable taxes, tolls, and fees and does not adjust for several payments, discounts, and refunds.

The lesson is not about either company. It is that labels do not make denominators comparable. Read the definition every time.

Walk one hypothetical order from payment to contribution

Take a hypothetical marketplace order with a listed transaction value of $100.

Assume the seller contract charges a $12 platform fee. Assume the buyer pays a separate $4 service fee. The buyer pays $104 before any tax that the model may need to treat separately. The seller receives $88 before any other seller adjustment. The platform fee pool is $16 before accounting treatment.

Now list the variable costs caused by this completed order:

• Payment processing and payout cost of $3.20
• Expected refund and chargeback loss of $0.80
• Support handling cost of $1.20
• Identity, fraud, and transaction review cost of $0.70
• A seller activation credit allocated to the order of $1.10

Those hypothetical costs total $7.00. The arithmetic is $3.20 plus $0.80 plus $1.20 plus $0.70 plus $1.10. Order contribution is the $16 fee pool minus $7.00 of variable costs, which equals $9.00.

The contribution rate on the fee pool is $9.00 divided by $16, or 56.25 percent. The fee pool as a share of the listed transaction value is $16 divided by $100, or 16 percent.

Every figure in this example is a hypothetical input. None is a benchmark, forecast, recommendation, or claim about a real marketplace. The value is the bridge, not the answer.

Change the model to match the real contract. If the platform funds a buyer discount, record who pays for it. If a refund also reverses the seller payout, show that. If the platform absorbs the refund, show that instead. If support is handled by salaried staff, estimate the cost from recorded handling time and loaded employment cost, then label the estimate.

Put costs in the right layer

Not every cost belongs against one order. Split the cost base into three layers.

Variable costs move with activity. Payment processing, payouts, per-check verification, shipping labels, transaction support, refunds, chargebacks, and order-linked incentives often sit here.

Step costs arrive when volume crosses a capacity limit. A new support shift, another review team, a higher software tier, or a regional operations lead may support a band of transactions rather than one order. Divide the cost across the relevant band, but keep the capacity trigger visible.

Fixed costs exist even if no order completes. Core product work, finance, legal, security, insurance, and management often begin here. They still matter. They just answer the company economics question after the order economics question.

Do not push fixed costs into an invented per-order precision too early. First prove that completed orders create positive contribution before central overhead. Then ask how much contribution the current volume produces and whether it can cover the next real cost layer.

Use completed transactions as the first denominator

Created listings are not completed transactions. Neither are searches, matches, accepted quotes, bookings, or authorized payments.

Build the marketplace funnel from recorded states:

• Qualified buyer request
• Eligible supply available
• Match or listing viewed
• Purchase or booking made
• Transaction completed
• Refund or dispute resolved
• Buyer repeats within a defined window
• Seller remains active within a defined window

The exact states depend on the model. A labor marketplace, resale marketplace, and delivery marketplace do not complete in the same way. Write one completion definition and use it across product, finance, support, and operations.

Then calculate contribution at each important failure point. A booking that cancels may still create processor cost, support work, and a seller payment. An order that looks like zero revenue can carry a real negative contribution.

Measure both sides by cohort

Blended averages hide which side is getting worse.

Group buyers by the period and channel that first brought them in. Track qualified requests, first completed transactions, repeat completions, refunds, support use, and contribution over a fixed window.

Group sellers by activation period and source. Track time to first eligible listing, time to first completed transaction, completion rate, cancellations, disputes, support use, payout, and retained activity over the same kind of window.

Now the marketplace can see whether growth comes from better matching or heavier subsidy. It can also see whether one buyer cohort creates orders that sellers do not want, or whether one seller cohort creates support and refund costs that the fee pool cannot carry.

Acquisition cost belongs beside these cohorts. Do not divide all marketing spend by all new accounts. Separate paid acquisition, referrals, sales work, partnerships, and organic discovery. Use the cost and the qualified cohort each route actually produced.

The useful question is not lifetime value. It is more concrete. How much observed contribution did this cohort create during the measured window, and what did it cost to acquire and serve?

Find the subsidy before it becomes strategy

Many marketplaces subsidize one side. That can be rational. It can also become a permanent leak with a good story attached.

List every credit, guarantee, fee waiver, free service, minimum payment, and make-good. Assign each one to the buyer, seller, market, and period it supported. Then ask what behavior changed.

A seller credit may improve supply in a thin market. A buyer discount may bring the first qualified request. A guarantee may reduce uncertainty enough to close a transaction. None has earned permanence merely because transactions rose while it existed.

Run the cohort again without mixing subsidized and unsubsidized activity. If the order contribution disappears when the incentive is removed, the incentive is part of the price. Treat it that way.

Stress the constraint that can reverse the answer

A marketplace model does not need a polished forecast first. It needs a reversal test.

Change one input at a time. Lower completion. Raise refund loss. Add the real support minutes. Remove a temporary fee waiver. Increase seller churn. Include the next capacity hire. Use the contract's actual payout timing. Show which change turns positive order contribution negative or prevents total contribution from covering the next cost layer.

Then test that input in the operation. If completion is decisive, inspect why qualified transactions fail. If support cost is decisive, classify contacts by cause. If seller retention is decisive, measure whether good sellers leave because demand is weak, jobs are poor, or the operating burden is high.

This is the same discipline we use in our working method. Find the assumption carrying the answer, set a kill condition, and test it before scaling the whole machine.

Write the decision in one page

The final unit economics memo should fit on one page before its evidence appendix.

State the transaction and completion definition. Show the three ledgers. Name the fee pool, variable cost, and order contribution. Show buyer and seller cohorts. Separate variable, step, and fixed costs. Identify every subsidy. Name the input most likely to reverse the result. Assign the next test and the person who owns it.

End with a decision. Continue the market, change the transaction, change the buyer or seller segment, change the fee structure, or stop. A marketplace that cannot explain one completed order is not ready for a larger market-size slide.

If you need to map the economics before committing to a build, what we do explains the research work and work with us explains how a first engagement starts.

This article is general business education. This is not financial advice. It is not investment, legal, tax, or accounting advice. Every numerical example is hypothetical and uses stated inputs. A qualified accountant and lawyer should confirm reporting and contract treatment. Investment decisions should be handled with a licensed financial adviser.

Sources

• IFRS Foundation: Post-implementation Review of IFRS 15. Principal and agent assessment, control, and gross or fee-based revenue recognition. Accessed 6 September 2026.
• Etsy 2025 Annual Report on Form 10-K. Definitions of gross merchandise sales, revenue take rate, and cost categories. Accessed 6 September 2026.
• Uber 2025 Annual Report on Form 10-K. Definitions of Gross Bookings and agent versus principal revenue presentation. Accessed 6 September 2026.