Marketplace Business Model

What formulas and metrics should a marketplace track?#

Transaction volume and GMV#

Completed transactions = qualified requests × match rate × completion rate

GMV = completed transactions × average transaction value

Use GMV or GMS for the value of transactions conducted through the marketplace, with a written policy for taxes, tips, shipping, cancellations, and refunds. GMV is market activity; it is usually not company revenue.

Revenue and take rate#

Transaction revenue = GMV × transaction take rate + fixed transaction fees

Total marketplace revenue = transaction revenue + subscriptions + listing fees + ads + other services − refunds and credits

Blended take rate = total marketplace revenue ÷ GMV

Always disclose a transaction take rate that excludes subscriptions, ads, and services alongside the blended figure. Otherwise a rising blended take rate may reflect a new revenue stream rather than higher fees on transactions — the Etsy pattern above.

Keep the numerator and denominator consistent. If revenue includes payment fees charged on shipping but GMV excludes shipping, or GMV is net of refunds while revenue is not, the ratio can move for the wrong reason.

Contribution economics#

Marketplace contribution = marketplace revenue − payment processing − variable support − trust and safety − unrecovered refunds and disputes − transaction incentives − variable infrastructure − transaction-linked insurance or fulfillment

Contribution margin = marketplace contribution ÷ marketplace revenue

Contribution per completed transaction = marketplace contribution ÷ completed transactions

Define which costs are included and keep the definition stable. Payment fees may sit in cost of revenue while trust and support staff sit in operating expenses; for internal unit economics, include every cost that rises because more transactions occur, regardless of statement classification. Etsy's cost of revenue — $817.8M in 2025 against $2,883.5M of revenue — includes payments fees, hosting and bandwidth, chargebacks, refunds it cannot recover from sellers, and third-party support. (Etsy 2025 Form 10-K) Those are precisely the costs a founder misses by modeling revenue as GMV × take rate and stopping there. Track contribution margin explicitly.

Liquidity metrics#

Liquidity means a participant with qualified intent can reliably find an acceptable counterparty and complete an exchange. Track by market cell and cohort:

• match rate, fulfillment rate, and cancellation rate;
• median and 90th-percentile time to match;
• supplier response rate and quote-to-book rate;
• repeat transaction rate and supplier earnings per available hour or listing;
• share of transactions completed entirely on-platform.

A global average hides dead zones. Show the distribution — the percentage of market cells above a target fulfillment rate, and the share of GMV those cells represent.

Two acquisition equations#

Demand CAC = demand acquisition spend ÷ newly activated transacting buyers

Supply CAC = supply acquisition and onboarding spend ÷ newly activated transacting suppliers

Do not divide by registrations. An activated supplier completes a defined first action — accepting and fulfilling a transaction — within a specified window. Measure retention separately for buyers and suppliers, but connect both to transactions. Track customer acquisition cost by side.

Worked example: an equipment-repair marketplace#

A marketplace connects small businesses with independent equipment-repair technicians in one metro area. In one month: 1,500 buyers submit qualified repair requests, 80% receive at least one acceptable match, 85% of matched requests become completed jobs, and the average completed job is $240.

Matched requests = 1,500 × 80% = 1,200

Completed jobs = 1,200 × 85% = 1,020

Fulfillment rate = 1,020 ÷ 1,500 = 68%

GMV = 1,020 × $240 = $244,800

The 68% fulfillment rate is more decision-useful than "1,500 leads." It shows 480 qualified buyers did not transact.

Revenue. The marketplace charges technicians 12% and buyers 3% (a 15% transaction take rate) and collects $99/month from 40 technician businesses.

Transaction revenue = $244,800 × 15% = $36,720

Subscription revenue = 40 × $99 = $3,960

Total marketplace revenue = $40,680

Blended take rate = $40,680 ÷ $244,800 = 16.6%

The dashboard should show both 15.0% transaction and 16.6% blended. The second is not evidence that transaction pricing increased.

Contribution.

Marketplace contribution = $40,680 − $18,596.20 = $22,083.80

Contribution margin = 54.3% · Contribution per job = $21.65 · Contribution ÷ GMV = 9.0%

The marketplace charges a 15% transaction take rate but retains only 9.0% of GMV as contribution.

Trust-loss stress test. Hold volume and revenue constant and let a fraud and quality problem raise unrecovered losses from 1% to 5% of GMV ($12,240), support from $3 to $7 per job ($7,140), and incentives from $4,000 to $8,000. Total variable cost rises to $36,468.20, contribution falls to $4,211.80, and contribution margin collapses from 54.3% to 10.4%. GMV, completion count, and take rate all look stable while the economics disappear.

Treat both scenarios as illustrations of structure, not benchmarks. The percentages depend entirely on the assumed cost rates, and a single month of 1,020 jobs is far too small a sample for the rates themselves to be stable — real cohorts show wide variance in dispute and cancellation rates.

What the team should investigate next. The 32% unfulfilled demand is the largest operating gap. Split it into: no relevant technician available; technician did not respond; buyer rejected price or timing; match accepted but canceled; and job moved off-platform. Each cause implies a different product, supply, pricing, or policy response. "Acquire more users" is not a diagnosis.

How do founders apply the model in practice?#

Design the initial wedge#

Choose a market cell where demand has a painful recurring coordination problem, supply is fragmented and reachable, the transaction is valuable enough to fund trust and payment costs, quality can be screened or signaled, and the platform can add value after the first introduction. A wedge is not simply a small TAM — it is a place where the marketplace can become meaningfully more liquid than the alternatives.

Sequence go-to-market around the constrained side#

If qualified supply is scarce, demand marketing worsens the experience. If demand is scarce, onboarding more suppliers reduces provider earnings and response rates. A workable sequence: manually recruit enough relevant supply for one cell → generate a controlled amount of qualified demand → concierge-match to learn failure reasons → standardize the repeated workflow → automate only after exceptions are understood → expand to adjacent cells against explicit entry thresholds.

Set marketplace pricing#

Start with four questions: which side receives the more measurable economic value; which side is scarcer or more price-sensitive; which side creates more value for the other by joining; and which costs and liabilities the platform assumes per transaction.

Then test fee allocation, not just fee level. A buyer fee reduces checkout conversion; a seller fee raises offered prices, reduces availability, or increases leakage. Subscriptions work for high-frequency suppliers but deter long-tail supply. Listing fees screen low-intent inventory but shrink variety. Treat pricing experiments as market experiments: measure both sides, completed volume, repeat, leakage, and contribution — not only revenue per transaction.

Design payments and the merchant model#

Map the flow of funds before implementation. Who is the seller to the buyer? Who sets price and terms? Who is merchant of record? Who collects identity and tax information? Who pays payment and payout fees? Who bears refunds, disputes, fraud, and negative balances? When is payout released? Which jurisdiction's tax, licensing, employment, insurance, and consumer rules apply?

Stripe's documentation makes the dependency explicit: charge type and negative-balance responsibility determine who responds to a dispute and which account is debited for the chargeback and fees, and platforms using indirect charges should take negative-balance responsibility themselves. (Stripe, Risk and liability management with Connect) A payment provider supplies infrastructure; it does not make the commercial and legal allocation decisions disappear.

Plan trust and safety proportionally#

Match controls to expected harm: identity and business verification; credential, license, or insurance checks; listing review; device, payment, and behavioral fraud signals; transaction holds or rolling reserves; reviews tied to verified transactions; and dispute intake with evidence, appeal, and enforcement. Every control creates friction — measure false positives, onboarding loss, appeal outcomes, and time to resolution alongside prevented loss.

Report in three dashboard layers#

Use the model in the pitch#

A strong marketplace traction slide shows the atomic transaction and initial cell; qualified demand, effective supply, and completed transactions; match rate, completion rate, and median time to match; buyer repeat and supplier active retention; GMV, transaction revenue, take rate, and contribution; and evidence that liquidity or economics improve for mature cohorts. A graph of registered users says almost nothing about whether a market exists.

How do investors evaluate a marketplace?#

Is the market actually liquid? Investors request cohort and market-cell data, not consolidated growth. They want to see that a buyer can reliably complete a transaction and that qualified suppliers earn enough to stay available.

Is there evidence of a reinforcing loop? Denser supply raising fulfillment and buyer repeat; more qualified demand raising supplier utilization and retention; older cells needing less incentive spend for the same service level. Correlation is not proof — brand spend, seasonality, or manual operations may drive both sides.

Are the unit economics real? Investors reconstruct contribution from payment statements, support cost, incentives, refunds, guarantees, insurance, and fulfillment. A marketplace that calls incentives "growth investment" forever may be hiding the price required to clear the market.

Does the platform keep adding value? The question is not "can users go off-platform" but "what workflow, protection, reputation, or demand access do they lose by leaving?" Evaluate switching costs and lock-in directly.

Is GMV being confused with revenue? Under Topic 606, gross versus net presentation turns on whether the entity controls the specified good or service before transfer, not on which number looks larger. (FASB ASU 2016-08)

What are the common mistakes?#

• Counting users instead of market outcomes. Downloads, sign-ups, and listings are inputs; the outcome is a completed, satisfactory interaction. Always connect participant counts to availability, transactions, and retention.
• Treating all supply as interchangeable. A thousand providers in the wrong category, geography, price band, or time window do not make a market liquid. Use effective supply and market cells.
• Reporting only blended averages. Overall fulfillment can rise because volume shifts toward an already-dense city while new cities fail. Report cohorts and the distribution of cell-level outcomes.
• Confusing GMV with revenue or cash. GMV is value exchanged in the market; revenue is what the company earns under its contracts and accounting policy; cash held before payout may be payable to others. Present all three distinctly.
• Calling subsidies network effects. Discounts can coordinate a cold start, but volume that disappears when incentives stop is not an organic loop. Show contribution and retention after normalizing incentives.
• Ignoring the second CAC. Supplier recruiting, verification, onboarding, and activation cost money. Treating provider acquisition as a free community motion overstates unit economics.

When does the marketplace model break?#

• The market is too thin or infrequent. Rare transactions, highly specific needs, and fragmented geography can make relevant supply uneconomic to maintain. Human brokerage may remain necessary, making the business a service with software support.
• Quality cannot be assessed before harm occurs. For medical, financial, childcare, or infrastructure services, reviews are slow or inadequate, and verification, supervision, insurance, and remediation can overwhelm the take rate.
• Participants multi-home and switching is easy. If offers are portable and prices transparent, participants compare platforms or route around them. Scale alone will not prevent it, and a one-time match with no continuing workflow is especially vulnerable to disintermediation.
• Trust costs rise faster than revenue. Fraudsters adapt, higher transaction values raise expected loss, and expansion adds language, verification, and support complexity — the stress test above showed stable GMV with contribution margin falling from 54.3% to 10.4%.
• The fee ceiling sits below the cost floor. If participants can transact elsewhere cheaply, the marketplace may be unable to charge enough to cover payments, support, verification, losses, insurance, and acquisition. A large-GMV business can remain structurally uneconomic.

Two further failure modes deserve separate mention. Governance creates conflicts: when the operator ranks participants and also sells its own products, it has an incentive to favor itself — a conflict the 2023 Merger Guidelines specifically recognize for platform operators that are also participants, which can both damage trust and attract competition scrutiny. (2023 Merger Guidelines, Guideline 9) And regulation changes the operating system: obligations run well beyond terms of service. In the United States, the INFORM Consumers Act imposes collection, verification, disclosure, reporting-mechanism, and seller-suspension duties on covered consumer-product marketplaces. (FTC INFORM Consumers Act guidance) In the EU, DSA Articles 29–32 add trader traceability, compliance-by-design, and information obligations for platforms letting consumers contract with traders, subject to the micro/small-enterprise exclusion. (Regulation (EU) 2022/2065, Articles 29–32) Employment classification, licensing, tax, sanctions, consumer protection, privacy, product safety, and insurance rules may also apply. Legal design should precede geographic or category expansion, not follow volume.

Founder decision checklist#

• What are the distinct sides, and what is the atomic transaction?
• Which terms do participants control, and which does the platform control?
• What is the first market cell, and which side is constrained today?
• What counts as qualified demand and effective supply, and what match, completion, and repeat levels indicate liquidity?
• Why will participants stay on-platform after meeting?
• What is the fee structure on each side; who is merchant of record; and who bears refunds, fraud, guarantees, and negative balances?
• What is contribution per completed transaction after all variable marketplace costs, and how much demand CAC and supply CAC does that support?
• Which trust controls and regulatory obligations must be in place before launch?
• Under what conditions would a reseller, SaaS, lead-generation, or managed-service model work better?

If the answers depend on "scale will fix it," the mechanism is incomplete.