A buyer opens your marketplace, searches once, finds nothing relevant, and never comes back. A seller lists, waits a week, gets no orders, and quietly leaves. Neither did anything wrong, there simply weren't enough people on the other side. That failure has a name, and getting past it is the whole game.

The quick version

  • Liquidity is the ease with which a buyer and a seller find the right match, a16z calls it the most critical thing a marketplace has, and most marketplaces that fail simply never reach it.
  • Two-sided businesses face a chicken-and-egg problem: buyers won't come without sellers, and sellers won't come without buyers. You break the loop by starting small and dense, not big and thin.
  • Get past it and two-sided network effects take over: more of one side attracts more of the other, which is the moat, but only after liquidity exists, never before.
  • The moves: find the hard side and grow it first, constrain to one geography or category until it's liquid, subsidise the side that's harder to attract, and measure the match (fill rate / search-to-fill), not vanity totals.

The idea in depth

A normal business has customers. A marketplace has two sets of customers who need each other, and it makes money only when it introduces them successfully. That structural difference changes everything about how you grow it, and almost every marketplace post-mortem traces back to the same root cause.

Liquidity is the number that matters, and it isn't users

Andreessen Horowitz, in its marketplace glossary, defines liquidity as "the ease with which buyers and sellers can find the right counterpart" and calls it "the most critical aspect of a marketplace; without it, a marketplace isn't valuable." The firm's blunt observation is that most marketplace failures come down to never achieving, or never maintaining, enough of it.

The reason total user count misleads you is that a marketplace can have plenty of both sides and still fail to match them. A million sellers and a million buyers who never transact is not a liquid marketplace; it's two crowds in separate rooms. Liquidity is measured at the level of the transaction: from the seller's side, the odds of making a sale in a reasonable window; from the buyer's side, the odds of finding what they came for. Investor Bill Gurley, in his much-cited 2012 essay "All Markets Are Not Created Equal," argues that a marketplace needs genuine "pull" on both sides, aggregating supply is necessary but nowhere near sufficient without organically aggregated demand.

So the move is to pick a liquidity metric and govern by it. Buyer-side: search-to-fill or fill rate (what fraction of real demand gets satisfied). Seller-side: the share of active listings that transact in a set period. If your headline dashboard is registrations or GMV alone, you are measuring the size of the crowd, not whether the introductions are working, and a marketplace that doesn't match is just an expensive directory.

The chicken-and-egg problem, and why you break it by going small

Because each side's value depends on the other already being there, a new marketplace starts at zero on both axes simultaneously. This is the chicken-and-egg (or "cold start") problem, and it is genuinely hard: the rational first buyer should wait for sellers, and the rational first seller should wait for buyers, so left alone the thing stays empty.

The economics here are not folklore. Jean-Charles Rochet and Jean Tirole formalised two-sided markets in their 2003 paper "Platform Competition in Two-Sided Markets" (work that contributed to Tirole's 2014 Nobel Prize), showing that a platform's job is to "get both sides on board," and that how you split the price between the two sides, not just the total price, determines whether the market forms at all. One side often has to be subsidised so the other side becomes worth showing up for.

The practical answer to chicken-and-egg is counter-intuitive: don't try to fill the whole market, fill a tiny corner of it completely. Andrew Chen, who led growth at Uber, builds his 2021 book The Cold Start Problem around the idea of the atomic network, the smallest network dense enough to stand on its own. Uber didn't launch "ride-sharing"; it launched a few dozen black cars in one slice of San Francisco, dense enough that a rider got a car in minutes. Liquidity in one small place beats thinness everywhere.

flowchart LR
    A(["No sellers
→ no reason for buyers"]) --> B(["No buyers
→ no reason for sellers"])
    B --> A
    A -.break the loop.-> C(["Seed one dense corner
(atomic network)"])
    C --> D(["Local liquidity: matches happen"])
    D --> E(["Both sides have a reason
to stay → expand"])
The chicken-and-egg loop, and the way out: don't fill the whole market, saturate one small corner first. Leaders Loop

So the move is to constrain ruthlessly at launch. One city, one category, one campus, one use case, narrow enough that the two sides actually collide. Concentrate every dollar of supply and demand acquisition inside that boundary until the corner is liquid, then copy the playbook to the next corner. Spreading thin to "look big" is the most common self-inflicted death.

Find the hard side, and grow it first

The two sides are rarely equally difficult to acquire. One is usually scarcer, pickier or slower to win, the hard side, and it tends to be the one that creates the value the other side comes for. Lenny Rachitsky, who led marketplace growth at Airbnb, surveyed marketplace founders and found the strong majority attacked supply first; his guidance is to work out which side is hardest and most valuable, and grow that side first. In Platform Revolution, Parker, Van Alstyne and Choudary describe the same instinct as seeding: deliberately manufacturing early supply (sometimes the platform itself plays one side) so the other side has a reason to arrive.

Which side is "hard" is an empirical question for your market, not a rule. For most labour and services marketplaces it's supply (good providers are scarce); for some content or rental marketplaces, demand is the bottleneck because supply is eager and abundant. The point is to diagnose it honestly and concentrate there. Reading the under-served side's real need is usually how you find the lever.

So the move is to name your hard side out loud and over-invest in it. Subsidise it, hand-hold it, and accept worse unit economics there for a while, because the easy side will follow it for free, and pouring acquisition into the easy side first just creates frustrated visitors who find nothing.

Liquidity is the cause; network effects are the reward. Chase the reward before you've earned the cause and you build a beautiful directory nobody transacts on.

The honest limitation: network effects are oversold, and liquidity is local

Here's where the standard story breaks down. "Network effects" gets used as a magic word, as if scale alone produces a moat. It often doesn't. Many marketplaces have local network effects, liquidity in San Francisco does nothing for a buyer in Sydney, so a national user count can hide a dozen thin, vulnerable city markets. Gurley's essay is largely a catalogue of why some markets simply make worse marketplaces than others, regardless of execution.

Two more cautions. Multi-homing erodes the moat: if sellers and buyers can cheaply use a competitor at the same time (drivers running Uber and a rival simultaneously), being bigger buys you less defensibility than the textbook implies. And the academic two-sided-market models are powerful for pricing logic but are simplifications, Rochet and Tirole themselves are modelling idealised platforms, not predicting that any given startup will tip. Treat network effects as something you must engineer and re-earn locally, not a force that switches on once you cross some user count.

So the move is to measure liquidity per market, not in aggregate, and to ask continuously what would make each side not switch to a rival, better matching, lower friction, reputation that doesn't travel. A moat you can't point to at the level of a single city is a moat you don't have.

A worked example

The figures below are illustrative, chosen to show the mechanics, not to report a real company.

Picture a marketplace for freelance video editors. The founders launch nationally: 5,000 editors signed up, 4,000 businesses browsing. It looks healthy. But a business in Leeds searches and sees three editors, all busy; an editor in Bristol gets one nibble a month. Spread across the whole country, neither side is dense enough to match. Both churn. This is high user count and near-zero liquidity, the classic trap.

They restart with a constraint: one city, one niche, short-form social video for Manchester e-commerce brands. They diagnose the hard side as supply (good editors are scarce and choosy), so they recruit 40 strong editors by hand and guarantee them a minimum of paid test briefs, a deliberate subsidy of the hard side. Now demand has something worth showing up for. Within that corner, the fill rate (briefs that get matched to an editor within 48 hours) climbs from 18% to 71%.

flowchart TD
    A(["National launch
5,000 editors, 4,000 buyers
fill rate 18%"]) --> B(["Thin everywhere
both sides churn"])
    B --> C(["Restart: 1 city, 1 niche
Manchester short-form video"])
    C --> D(["Hand-recruit 40 editors
+ guaranteed paid briefs
(subsidise hard side)"])
    D --> E(["Local liquidity
fill rate 18% → 71%"])
    E --> F(["Editors earn → refer peers
buyers return → copy to city #2"])
From thin-everywhere to dense-somewhere: the fix is constraint plus a subsidised hard side, then replication. Illustrative figures. Leaders Loop

Once that corner works, the two-sided loop turns: editors who earn reliably refer other editors, buyers who get matched come back and tell peers, and the founders lift the constraint one market at a time, Manchester, then Birmingham, then Glasgow, running the same dense-corner playbook in each. The national footprint they tried to buy on day one, they now build one liquid city at a time. The lesson is the same one the whole field keeps relearning: dense beats big.

Frequently asked questions

What exactly is marketplace liquidity?

It's the probability that the two sides successfully match: that a seller makes a sale, and a buyer finds what they came for, within a reasonable window. a16z calls it the most critical aspect of a marketplace. Crucially it's measured at the transaction level (fill rate, search success), not by counting registered users, a marketplace can be huge and still illiquid if the two sides never connect.

Should I grow supply or demand first?

Grow whichever side is the hard side, the scarcer, harder-to-win, more value-creating side, first. For most marketplaces that's supply, and Lenny Rachitsky's survey of founders found the majority attacked supply first. But it's an empirical question for your market: if supply is eager and abundant and demand is the bottleneck, invest there. Diagnose honestly rather than following the default.

How do I actually break the chicken-and-egg problem?

Stop trying to fill the whole market. Constrain ruthlessly, one city, one category, one campus, until that small corner is genuinely liquid (Andrew Chen's "atomic network"). Subsidise or hand-recruit the hard side so the other side has a reason to show up, prove the loop in one place, then replicate. Going small and dense beats going big and thin almost every time.

Are network effects a guaranteed moat once I'm big?

No, that's the most expensive misconception in the category. Many network effects are local (liquidity in one city does nothing for another), and multi-homing, both sides cheaply using a rival at the same time, erodes defensibility. Treat the moat as something you re-earn market by market through better matching and lower friction, not a switch that flips at some user count.

What metrics should a marketplace leader watch?

Lead with liquidity per market (fill rate, search-to-fill, share of listings that transact), then layer growth and quality on top. Watch the hard side's health specifically, its churn and earnings/outcomes, because that's the side that, if it leaves, collapses the loop. Aggregate GMV and total users are lagging, easily-flattering numbers; the matching rate is the leading one.

Related in the Toolkit

Where to go next