Two-Sided Markets

What exactly is a two-sided market?#

A two-sided market connects two distinct groups that need each other to complete a valuable interaction, while the platform sets the rules, access, discovery, trust, payments, or other infrastructure around it.

The defining feature is indirect, or cross-side, network effects: more relevant sellers make the platform more useful to buyers; more qualified buyers make it more useful to sellers; more developers make an operating system more useful to users, and vice versa. The two effects need not be equal — one side may care intensely about the other while the reverse effect is weak. The 2023 US Merger Guidelines note explicitly that indirect network effects can be asymmetric and heterogeneous across sides and segments.

Rochet and Tirole's foundational research framed the business-model problem as getting both sides on board while choosing how to court each and still earn money overall. Their sharper economic test is price-structure non-neutrality: holding the platform's total price constant, shifting price from one side to the other can change participation or transaction volume.

The practical founder test#

1. Distinct sides: Are there separate participant groups with different jobs, constraints, and alternatives?
2. Interaction: Does the product enable a search, match, transaction, or complementary-product relationship between them?
3. Cross-side effect: Does more relevant and available participation on one side measurably improve the other side's conversion, retention, willingness to pay, or outcomes?
4. Price-structure effect: Would shifting the same total fee or subsidy between sides change participation or transaction volume?

If the last two answers are no, the business may still be attractive — but "two-sided market" is probably not the mechanism doing the work.

Two-sided market versus adjacent models#

A pipeline business can add a marketplace, and a marketplace can integrate vertically into supply. The label should follow actual control and risk: if the company sets inventory, performs the service, controls fulfillment, and bears delivery risk, investors will analyze it partly as an operator rather than a capital-light platform.

Why do two-sided markets matter to founders?#

Growth on one side can improve acquisition and retention on the other#

In a working loop, relevant supply improves demand conversion, which generates supplier earnings, which improves supply retention and availability, which further improves demand outcomes. That can lower acquisition cost or raise repeat use without proportional paid marketing.

The feedback must operate where users actually make choices. A ride network is local by place and time; a labor marketplace may be local by skill, seniority, rate, time zone, and contract type. Uber's 2025 Form 10-K makes the constraint explicit: balancing supply and demand in a given area at a given time can matter more to service quality than the network's absolute size, and insufficient participation reduces network liquidity.

The platform has two go-to-market motions, not one#

Each side can have a different acquisition channel, activation event, sales cycle, willingness to pay, retention driver, and density requirement. "Acquire users" is too vague for a marketplace plan. Founders must decide which side is harder to acquire, which side creates more cross-side value, which side is seeded first, whether one side should be free or guaranteed, and what density threshold admits more demand.

The cheapest participant is not necessarily the right side to acquire first. Ten available specialists may create more buyer value than 10,000 low-intent profiles — and buyer demand that rarely converts wastes supplier time and causes supply churn.

Pricing is a system, not a single take rate#

A platform can charge access fees, listing fees, subscriptions, buyer service fees, seller commissions, payment fees, advertising, fulfillment fees, or premium tools. It can also pay referral rewards, guarantees, promotions, rebates, or minimum earnings. The economic question is not "what take rate can we charge?" but which combination of prices, subsidies, access rules, and service levels maximizes durable interaction volume and contribution across both sides.

The side with more elastic participation or greater cross-side value may receive the lower price. A subsidized side is not irrational if its participation generates enough value elsewhere. It becomes irrational when the subsidy is permanent, competitors can copy it cheaply, or the incremental activity never produces contribution.

Liquidity is the product before network effects become a moat#

Early marketplaces often describe future defensibility before proving present utility. Users care first about whether the platform solves the job now: can the buyer find an acceptable option, can the supplier earn enough, how long does matching take, and how often do accepted interactions cancel or move off-platform?

Airbnb's metric architecture is the transferable lesson rather than its scale. It counts Nights and Seats Booked net of cancellations and alterations and treats Gross Booking Value as a leading indicator of revenue — 533.0 million nights and seats and $91.3 billion of GBV in 2025, against $12.24 billion of revenue. Count realized interactions and transaction value, not accounts.

Governance can create or destroy platform value#

The platform chooses who may join, what may be listed, how ranking works, who bears loss, how disputes resolve, and whether the operator competes with participants. More participants can also create negative effects: fraud, spam, low-quality supply, congestion, price undercutting, or excessive advertising. Airbnb notes that trust-and-safety measures are important but add cost and friction and can reduce listings or bookings.

Regulators treat platform structure as economically important too. The EU Digital Markets Act identifies strong network effects, multi-sidedness, dependence, lock-in, lack of multi-homing, vertical integration, and data advantages as features that weaken contestability for large gatekeepers. Its obligations do not apply to a startup, but the mechanisms matter long before that scale.

The mechanism: cross-side utility, liquidity, and price structure#

1. Model the value to each side#

U_A = V_A + α_A × Q_B - P_A - F_A U_B = V_B + α_B × Q_A - P_B - F_B

where V is standalone value, Q is quality-adjusted serviceable participation on the other side, α is the strength of the cross-side effect, P is price net of incentives, and F is non-price friction — onboarding, waiting, risk, search, integration, or switching cost. Participants join or transact when expected utility beats their next-best alternative.

This is not a forecasting equation until its terms are estimated; its job is to force the team to specify the mechanism. Use quality-adjusted, serviceable participation, not raw accounts: a tutor unavailable for the requested subject and time contributes nothing to that match, and a stale listing makes search worse.

2. Define the market cell in which matching must clear#

Market cell = geography × category × time window × participant constraints

Rides within 10 minutes in a specific zone and hour. A licensed nurse for a night shift within commuting distance. A spare part with an exact specification and delivery deadline. Liquidity measured globally hides dead cells — report cohorts and distributions: median and 90th-percentile time-to-match, fulfillment by neighborhood or category, utilization by day or skill.

3. Track liquidity from both sides#

Demand fulfillment rate = completed matches / qualified demand events Supply utilization = completed matches / serviceable supply opportunities Match acceptance rate = accepted matches / matches offered Repeat rate = participants with another completed interaction / eligible participants

No single metric suffices. High fulfillment with very low utilization can mean oversupply and supplier churn; high utilization with low fulfillment means shortages, long waits, and poor buyer retention. Track time-to-match, cancellation, refund, dispute, and off-platform leakage alongside them.

4. Test price structure, not only total price#

Total platform fee = buyer fee + seller fee

If moving one dollar of fee from sellers to buyers changes participation, availability, conversion, or transactions, the split matters. Test the split within comparable market cells and allow enough time to observe retention effects on both sides. Do not assume the higher-willingness-to-pay side should pay more; weigh participation elasticity, the cross-side value each additional participant creates, pass-through ability, multi-homing options, and the fraud, credit, insurance, and support cost each side carries.

5. Reconcile GMV, revenue, and contribution#

GMV = completed transaction count × average transaction value Effective take rate = platform revenue / GMV Transaction contribution = platform revenue - transaction-variable costs - incentives

Variable costs include payment processing, fraud and chargebacks, insurance, fulfillment, verification, support, refunds, credits, and side-specific guarantees. Compare take rates across companies only after reconciling what each includes in GMV, whether revenue is gross or net, and where incentives are presented — two platforms with the same economic fee can report different revenue ratios. Uber, for example, reports "Revenue Margin" as revenue divided by Gross Bookings while separately describing driver incentives, consumer discounts, payments, insurance, and network costs. A revenue ratio is not a profit margin.

6. Measure the cross-side effect directly#

Cross-side elasticity A←B = % change in A outcome / % change in serviceable B participation

Estimate it with staged geographic launches, randomized ranking or availability interventions, waitlist releases, or credible natural experiments. Do not simply correlate large markets with good outcomes — attractive markets cause both supply and demand to be high, which is the classic confound in every marketplace deck.

Worked example: a local tutoring marketplace#

One city, one subject category, one month. All figures are hypothetical.

Step 1 — measure liquidity. Qualified lesson requests 800; completed bookings 560; active tutors 140; serviceable slots per tutor 8, so total serviceable slots 140 × 8 = 1,120.

Demand fulfillment = 560 / 800 = 70% Supply utilization = 560 / 1,120 = 50%

The aggregate does not prove oversupply. Monday-evening algebra could be short while weekend calculus is oversupplied. Segment time-to-match, fulfillment, and utilization by subject, location, time, price band, and tutor quality before acting.

Step 2 — GMV and fee revenue. Lesson price before platform fees $90; buyer service fee 5%; tutor commission 15%.

GMV = 560 × $90 = $50,400 Buyer-fee revenue = $50,400 × 5% = $2,520 Tutor-commission revenue = $50,400 × 15% = $7,560 Platform revenue = $10,080, an effective take rate of $10,080 / $50,400 = 20%

The buyer pays $90 × 1.05 = $94.50; the tutor receives $90 × 0.85 = $76.50 before their own taxes and expenses; the platform keeps $18 before its costs.

Step 3 — transaction contribution. Blended payment cost 3% of the buyer charge; variable verification, support, refund, and dispute cost $4 per booking; demand-side credits and referral rewards $2 per booking.

Contribution per booking = $5,132.40 / 560 = $9.17 Contribution margin on platform revenue = 50.9%

If the month also acquired 90 first-time transacting buyers at $24 each and activated 15 new tutors at $140 each, acquisition and activation cash is (90 × $24) + (15 × $140) = $4,260, leaving $872.40. That does not prove the new cohorts are profitable — it is one month of a payback that depends on repeat transactions and side-specific retention.

Step 4 — why the fee split matters. Keep the total fee at 20% but change the split to a 15% buyer fee and 5% tutor commission. The buyer now pays $90 × 1.15 = $103.50, the tutor receives $90 × 0.95 = $85.50, and platform revenue per completed lesson is unchanged at $18.

Assume an experiment indicates the higher buyer fee reduces completed demand by 12% while better tutor economics add 10% serviceable supply. Demand is the binding side:

Completed bookings = 560 × 0.88 ≈ 493 GMV = 493 × $90 = $44,370 Platform revenue = $44,370 × 20% = $8,874 Payment cost = 493 × $103.50 × 3% = $1,530.77 Other variable cost = 493 × $4 = $1,972; Demand incentives = 493 × $2 = $986 Transaction contribution = $4,385.23

The total fee per completed transaction never changed, but contribution fell from $5,132.40 to $4,385.23 because the reallocation hurt demand more than it relieved the binding constraint. A different market could produce the opposite result — and the -12%/+10% response is an assumption, not a finding. The point is to measure the joint response rather than assume a universally "correct" side to charge.

Related concepts#

• Transaction-Based Pricing: connect GMV, take rate, transaction fees, and contribution.
• Pricing Metric and Value Metric: charge for access, listing, transaction value, or completed outcomes.
• Customer Segments: define distinct sides and the within-side segments that value the network differently.
• Willingness to Pay: estimate participation and price response on each side.
• Penetration Strategy: evaluate temporary pricing used to build density without making subsidies permanent.
• Hybrid Pricing: combine access, subscription, transaction, and premium-service revenue.
• TAM, SAM, and SOM: size the market using serviceable market cells rather than global participant counts.
• Ideal Customer Profile: prioritize the participants most likely to create repeatable liquidity.
• Data moats: separate cross-side network effects from data-learning advantages.
• Managed Services: distinguish a platform's scalable interaction layer from human-operated delivery.

Sources#

1. Rochet, J.-C. and Tirole, J., "Platform Competition in Two-Sided Markets," Journal of the European Economic Association 1(4), 2003
2. Rochet, J.-C. and Tirole, J., "Two-Sided Markets: A Progress Report," RAND Journal of Economics 37(3), 2006
3. Airbnb, Inc., 2025 Form 10-K
4. Airbnb, Q4 and Full Year 2025 Financial Results
5. Uber Technologies, Inc., 2025 Form 10-K
6. U.S. DOJ and FTC, 2023 Merger Guidelines, Guideline 9 (multi-sided platforms)
7. FTC, "How Loyalty Discounts Between Firms Harm Competition When There Are Network Effects: FTC v. Surescripts" (February 2026)
8. Regulation (EU) 2022/1925, Digital Markets Act

---

This page is an educational and operating explanation, not legal or financial advice. Exclusivity, parity and most-favored-nation clauses, loyalty pricing, worker classification, payments, and consumer-protection duties vary by jurisdiction — obtain qualified advice before adopting them.