The Bazaar Without Walls

In September 1995 Pierre Omidyar, a software engineer in California, wrote a simple auction website over a long weekend and called it AuctionWeb. One of the first things sold on it was a broken laser pointer, for $14.83. Omidyar contacted the buyer to make sure he understood the pointer was broken. The buyer replied that he collected broken laser pointers.¹

The story is usually told as a joke about the strangeness of the internet. It is also a lesson about markets. A buyer wanted something no shop would stock. A seller had something no shop would buy. Neither knew the other existed, and neither could see the other or inspect the goods. AuctionWeb, renamed eBay in 1997, made the trade possible, but it raised a question that would shape every online marketplace that followed: why should strangers who cannot see each other trust each other enough to trade?

This essay follows how online marketplaces answered that question, from eBay’s feedback scores to Amazon’s warehouses, Taobao’s escrow, Flipkart’s cash on delivery and Airbnb’s guarantee. It is a story about trust, but its deeper subject is power. A marketplace begins as a meeting place. It becomes a ruler: setting fees, writing rules, settling disputes, ranking sellers and deciding who is seen. And in some cases it becomes a competitor to the very merchants it governs. To see why that matters, we need to begin with the oldest kind of market.

The bazaar and the problem of information

Markets are among the oldest human institutions. The agora of ancient Athens, the souks and bazaars of the Middle East and South Asia, and the market squares of medieval Europe gathered many sellers in one place, under rules set by custom, guilds or officials. Their problem, in every age, was information. A buyer did not know the quality of the goods, the fair price, or whether the seller was honest.

In 1978 the American anthropologist Clifford Geertz published a study of the bazaar in Sefrou, a town in Morocco, in which he argued that the bazaar should be understood as a system for searching for information that is scarce, unevenly distributed and costly to obtain. Prices were not posted; quality varied; goods were not standardised. Buyers and sellers coped in two main ways. They bargained intensively over each transaction, probing for information. And they formed lasting relationships, which Geertz called clientelisation: a buyer returned to the same trusted seller again and again, and the seller had a reason not to cheat a regular customer.² Reputation, built up over repeated dealings and spread by word of mouth, did the work that inspection could not.

Modern retailing replaced much of this with brands, fixed prices and the reputation of the shop itself. A buyer at a department store trusted the store to stand behind what it sold. The internet threatened to undo both solutions. It brought together millions of buyers and sellers who had never dealt with each other and had no shop to vouch for them. It allowed no bargaining in person, no inspection and, at first, no reputations at all.

Lemons

In 1970 the economist George Akerlof described what happens to such a market. In “The Market for ‘Lemons’,” he imagined a market for used cars in which sellers know whether a car is good or a “lemon,” and buyers do not. Buyers, unable to tell the difference, will pay only an average price. Owners of good cars, unwilling to sell at that price, withdraw them. The average quality falls, so buyers offer less, and more good cars leave. In the extreme, the market collapses, and only lemons remain. Akerlof argued that many institutions, such as guarantees, brand names, licensing and chains of shops, exist to counteract this problem of quality uncertainty.³ He shared the Nobel Prize in economics in 2001 for this work.

Early online trading was a lemons market waiting to happen. Every marketplace that succeeded did so by inventing a way to tell buyers which sellers and goods were good.

eBay: reputation as a number

AuctionWeb grew rapidly through 1996, and so did disputes. Omidyar, running the site largely alone, could not investigate them all. In February 1996 he introduced the Feedback Forum, which let each party to a trade leave a positive, neutral or negative rating of the other, with a short comment, displayed publicly with the seller’s or buyer’s name.¹ Omidyar’s stated hope was that most people were honest, and that a public record would let them prove it and expose those who were not.

The feedback score turned reputation, which in the bazaar lived in the memories of regular customers and in gossip, into a public number attached to a username. A seller with thousands of positive ratings had a valuable asset that one fraud could damage. A new seller had to earn trust with small transactions. The economist Paul Resnick and colleagues, studying eBay data from 1999, found that feedback was almost always positive, around 99 percent, and that reputation nevertheless had measurable value: established sellers could command somewhat higher prices.⁴ The system worked imperfectly, with fear of retaliation suppressing negative ratings, but it worked well enough for strangers to trade.

eBay’s model spread to almost every later marketplace: Amazon’s seller ratings and product reviews, Airbnb’s reviews of hosts and guests, Uber’s ratings of drivers and riders. Two problems followed it everywhere. Ratings inflated over time: a study by Apostolos Filippas, John Horton and Joseph Golden found that average ratings on several online marketplaces grew steadily more positive over the years, largely because raters felt social pressure not to harm the person they rated, so that ratings became less informative.⁵ And reviews could be faked, bought or suppressed. In August 2024 the US Federal Trade Commission adopted a rule banning fake reviews, paid reviews and the suppression of negative ones, with civil penalties for violations.⁶

Amazon: from shop to marketplace

Amazon began in 1995 as an online bookseller that bought, stored and sold its own stock. Its founder, Jeff Bezos, described its ambition as offering the largest selection on earth. That ambition was beyond what any retailer could stock. Amazon tried online auctions in 1999 and a section for small shops, zShops, without much success. In November 2000 it took a bolder step: it let independent sellers offer new and used goods directly on Amazon’s own product pages, alongside, and in competition with, Amazon’s own offer.⁷ A customer looking at a book could buy it from Amazon or, often more cheaply, from another seller.

The decision was controversial inside Amazon, because it invited competitors onto its most valuable pages. Bezos’s argument, often repeated in his letters to shareholders, was that what mattered was the customer’s experience: more selection and lower prices would bring more customers, which would attract more sellers, in a self-reinforcing loop. The 1997 letter to shareholders set out the philosophy of long-term customer focus behind such bets.

In 2006 Amazon launched Fulfillment by Amazon (FBA). Sellers could send their goods to Amazon’s warehouses; Amazon would store, pack and ship them, handle returns and customer service, and make them eligible for Prime, its fast-delivery membership. FBA solved the lemons problem from a different direction. A buyer did not need to trust an unknown seller to ship promptly or handle a return fairly, because Amazon did it. The seller borrowed Amazon’s reputation.

Third-party sellers became the core of Amazon’s retail business. By the last quarter of 2024 they accounted for a record 62 percent of the units sold on Amazon.⁸ Amazon earned revenue from them in several ways: commissions on each sale, fees for fulfilment and storage, and, increasingly, advertising, as sellers paid to appear at the top of search results within Amazon. The marketplace that began as a way to add selection had become a platform on which millions of businesses depended.

China: Taobao and the trust that eBay could not supply

In 2002 and 2003 eBay paid about $180 million to buy EachNet, the leading Chinese online auction site, which then held most of China’s market. In 2003 Alibaba, a Hangzhou company that ran a website connecting Chinese manufacturers with foreign buyers, launched Taobao, a consumer marketplace, and offered sellers free listings.⁹

Taobao’s advantage was not only price. It understood the specific trust problems of Chinese buyers. Few had credit cards, and many distrusted sellers they could not see. In 2004 Alibaba launched Alipay, which held the buyer’s payment in escrow until the buyer confirmed receipt. Taobao also built a messaging tool that let buyers chat with sellers before buying, restoring something like the bazaar’s conversation. eBay, meanwhile, moved EachNet onto its global technology platform, which made the site slower and less adapted to local habits, and charged listing fees.

By one estimate Taobao’s share of the Chinese market rose from 8 percent to 59 percent between 2003 and 2005, while eBay’s fell from 79 percent to 36 percent. eBay offered free listings in 2006, too late, and closed its Chinese site that year.⁹ Taobao became the largest online marketplace in the world, and Alibaba’s Singles’ Day sales on 11 November became the world’s largest shopping event. The lesson was not that free beats paid. It was that trust is local: the mechanisms that make strangers trustworthy depend on the habits, institutions and fears of the people trading.

India: paying on the doorstep

India posed a similar problem in a different form. Flipkart was founded in 2007 in Bangalore by Sachin Bansal and Binny Bansal, two former Amazon employees, as an online bookshop. Online shopping in India faced deep distrust: few people had credit cards, many feared fraud, and delivery was unreliable. Flipkart’s answer, introduced around 2010, was cash on delivery: the customer paid in cash when the goods arrived at the door, after seeing them.¹⁰ The marketplace, not the customer, carried the risk of the first transaction.

Cash on delivery was expensive. Couriers had to carry and reconcile cash, and customers sometimes refused parcels on arrival, which raised return costs. But it won millions of first-time online shoppers and became standard across Indian e-commerce. Flipkart also built its own logistics network, Ekart, because existing couriers could not deliver reliably.

India’s marketplaces were shaped heavily by regulation. Foreign direct investment in multi-brand retail was restricted, so foreign-funded e-commerce companies, including Amazon India and Flipkart, operated as marketplaces connecting independent sellers with buyers rather than as retailers holding their own stock. Rules announced at the end of 2018 and effective in February 2019 barred such marketplaces from selling products through sellers in which they held equity and from arranging exclusive deals, after complaints from traders that the marketplaces were favouring affiliated sellers.¹¹ In May 2018 Walmart bought 77 percent of Flipkart for $16 billion, its largest acquisition.¹² In 2022 the government began piloting the Open Network for Digital Commerce (ONDC), an attempt to apply the logic of UPI to commerce: an open protocol through which any buyer app could reach any seller app, so that no single platform would control the market. The essay on digital payments describes the payment system on which that idea was modelled.

Airbnb: strangers in your home

The sharing economy of the late 2000s pushed the marketplace model into more intimate territory. Airbnb, founded in San Francisco in 2008, let people rent rooms or homes to strangers; Uber, from 2009, let strangers ride in each other’s cars. Each relied on two-way ratings, on payment through the platform and on identity checks to persuade people to take risks far greater than buying a used laser pointer.

In June 2011 a host in San Francisco, writing under the name EJ, described returning home to find her apartment ransacked by an Airbnb guest. Airbnb’s initial response was widely criticised as slow. In August its chief executive, Brian Chesky, published an apology, saying the company had let her down, and announced a $50,000 guarantee for hosts against damage by guests, later expanded.¹³ The episode taught a lesson that every marketplace of strangers eventually learns. Ratings and identity checks reduce risk but cannot eliminate it, and when something goes badly wrong, the platform that profited from the trust is expected to stand behind it.

The marketplace as ruler

As marketplaces grew, they came to govern the businesses that sold on them. Amazon, Taobao and Flipkart each set the fees sellers paid, the rules they had to follow, the standards for delivery and returns, and the procedures for disputes. They decided which seller’s offer appeared first, through search rankings and, on Amazon, through the “Buy Box,” the button that determined which seller made the sale when a customer clicked “Add to Cart.” They could suspend sellers, sometimes with little explanation, which for a business dependent on the marketplace could be fatal. Millions of merchants lived under rules they did not write and could not appeal outside the platform.

The power grew with dependence. A seller who built a business on Amazon found that leaving meant losing most of its customers. Amazon’s fees, including commissions, fulfilment charges and advertising, rose as a share of sellers’ revenue over the 2010s, according to analyses by sellers and researchers, though Amazon disputed characterisations of its fees as excessive.

Referee and player

The most contested question was what happens when the marketplace also sells in its own market. Amazon was both the operator of the marketplace and its largest retailer, and it developed its own brands. It also had data on every seller’s sales. In 2020 reports in the press alleged that some Amazon employees had used data about individual third-party sellers to develop and price Amazon’s own products, which Amazon said violated its policies.

In July 2019 the European Commission opened an investigation into Amazon’s use of non-public data from independent sellers, and in 2020 a second into whether the rules for the Buy Box and Prime favoured Amazon’s own retail offers and sellers who used its logistics. In December 2022 the Commission accepted binding commitments from Amazon: not to use sellers’ non-public data for its own retail business, including its private-label products; to treat all sellers equally when choosing the Buy Box winner and to show a second competing offer; and to let Prime sellers choose their own delivery companies.¹⁴

In the United States, the FTC and seventeen states sued Amazon in September 2023, alleging that it illegally maintained monopoly power through practices that punished sellers for offering lower prices elsewhere and that effectively required sellers to use its fulfilment service to qualify for Prime. Amazon denied the allegations. The case is expected to go to trial in 2027. In a separate case, about how customers were enrolled in Prime and how hard it was to cancel, Amazon agreed in September 2025, a few days into trial, to pay a $1 billion civil penalty and $1.5 billion in refunds to customers, without admitting the allegations.¹⁵

What marketplaces are today

Online marketplaces are now a dominant form of retail in much of the world. They solve the lemons problem with a combination of tools: ratings and reviews, which make reputation public; escrow and guarantees, which move the risk of the first transaction to the platform; fulfilment, which lets unknown sellers borrow the platform’s reliability; and rules enforced by suspension. They are also governments over large economies of merchants, with the power to set fees, write rules and decide who is seen. And several of them compete with the merchants they govern.

How marketplaces changed buying and selling

Several shifts follow from the rise of online marketplaces.

From the shop’s reputation to the seller’s score

A shop vouched for what it sold. A marketplace vouches for the process, and lets the seller’s ratings vouch for the seller. Reputation became granular and public, which let small sellers compete with large ones and also made them vulnerable to a few bad reviews.

From relationship to ranking

The bazaar’s buyer returned to a trusted trader. The marketplace’s buyer searches, sorts by rating and price, and often buys from a different seller each time. Loyalty moved from the seller to the platform.

From a market to a platform

A market square belonged to the town; its rules were public. A marketplace belongs to a company; its rules are private, change without negotiation and are enforced by the company itself. Sellers gained access to vast numbers of customers and lost the independence of a shop with its own door.

From small choice to endless choice

A shop offers what fits on its shelves. A marketplace offers millions of items, many nearly identical. Search and ranking replaced the shopkeeper’s selection, and the burden of choosing moved to the buyer, guided by algorithms and reviews.

From neutral ground to a field with a home team

A traditional market square was run by authorities who did not themselves trade. Online marketplaces are run by companies that often do, which turns questions of fair competition into questions of governance.

What the history suggests

Marketplaces repeat patterns found across this site. The defences against fraud, ratings, escrow and fulfilment, worked best at scale, and so concentrated the market in a few platforms. Users decided what the platforms were for: eBay’s collectors, Taobao’s chatting buyers, Flipkart’s cash-paying customers. Local trust mattered more than global technology, as eBay learned in China. And incumbents absorbed what their sellers proved, as when marketplace operators developed their own versions of products that sold well.

But marketplaces also show something more particular. They solved Akerlof’s problem not by making buyers better at judging quality but by building a private institution that judged it for them, through scores, guarantees and rules. That institution became powerful in proportion to its success. The question the history leaves is the one regulators are now trying to answer: what obligations come with running a market that millions of others depend on? That is where the philosophy begins.

Understanding the philosophy

The rest of this essay examines the ideas beneath online marketplaces, one at a time. Each is explained in plain terms, connected to the tradition of thought that has examined it, traced in the history above, tested against its strongest objection and related to how people and organisations decide today. The connections to philosophers, economists and anthropologists are interpretations offered to make the ideas clearer. The founders of marketplaces were solving practical problems and did not, as a rule, set out to answer these thinkers.

1. When buyers cannot judge quality

The idea in plain terms

If buyers cannot tell good goods from bad, they will pay only an average price, sellers of good goods will leave, and the market will fill with bad ones. Every market that works has found some way around this problem. Online marketplaces had to solve it from scratch, among strangers who could not see each other or the goods.

The tradition

Akerlof’s 1970 paper made the problem precise and showed how widely it applied: to used cars, to insurance, to credit in developing countries, to employment. His insight was that the problem is not that sellers are dishonest but that information is unequal. Even honest sellers of good goods suffer, because buyers cannot distinguish them.

Economists then studied the solutions. In 1973 Michael Spence described signalling: a party with private information can take a costly action that only those with good quality would find worthwhile, such as a seller offering a long warranty, which would be ruinous for a seller of bad goods. Joseph Stiglitz studied screening, in which the uninformed party designs choices that lead the informed to reveal themselves. Spence and Stiglitz shared the 2001 Nobel Prize with Akerlof. Akerlof himself pointed to institutions that counteract quality uncertainty: guarantees, brand names, chains, licensing.

Where it shows up

Each successful marketplace invented its own counter to the lemons problem. eBay’s feedback created a reputation that sellers would not risk for a single fraud. Amazon’s fulfilment let unknown sellers borrow Amazon’s reliability, and its generous return policy acted as a warranty. Taobao’s escrow meant a buyer paid only after receiving the goods. Flipkart’s cash on delivery went further, letting the buyer see the goods before paying. Airbnb’s guarantee insured hosts against the worst guests. In Spence’s terms, sellers who accepted these arrangements signalled that they expected to deliver; in Akerlof’s, the platforms built the institutions that counteract quality uncertainty.

The strongest objection

The objection is that online markets have less of a lemons problem than offline ones, because buyers can compare prices and read reviews instantly, and return goods easily. Information is more abundant, not less. That is partly true, and it is the result of the solutions above, not evidence that the problem was absent. Without them, early online trading was indeed plagued by fraud, as PayPal and eBay found. And new forms of the problem keep appearing: counterfeit goods, fake reviews and products whose quality cannot be judged until long after purchase.

What it means today

Any business that sells to strangers is solving a lemons problem, whether it knows it or not. The question to ask is which mechanism is doing the work: the seller’s reputation, the platform’s guarantee, the return policy, the brand. Each has costs and weaknesses. A business selling through a marketplace is borrowing the marketplace’s solution and paying for it, in fees and in dependence.

2. Reputation as memory

The idea in plain terms

People behave better when they expect to deal with each other again, and when others will hear about how they behaved. Reputation is a kind of social memory that makes the future shape the present. Online marketplaces turned that memory into a number, which made it portable, comparable and permanent, and also made it something to manage.

The tradition

In 1984 the political scientist Robert Axelrod published The Evolution of Cooperation, which reported a series of computer tournaments in which strategies played a repeated prisoner’s dilemma, a game in which two players each do better by betraying the other in a single round but both do better by cooperating over many rounds. The winning strategy, tit for tat, cooperated first and then copied the other player’s last move. Axelrod emphasised what he called the shadow of the future: cooperation is possible when the future matters enough relative to the present. Repeated dealings make honesty rational.

Philosophers had seen this long before. David Hume argued in the eighteenth century that people keep promises partly because they know that breaking them would cost them the trust of others in future dealings, so that a merchant’s reputation for honesty was an asset worth more than any single gain from cheating.

Economists studying medieval trade made the same point with evidence. The economic historian Avner Greif showed that eleventh-century Maghribi traders in the Mediterranean enforced honesty among agents through a coalition that shared information about cheaters and refused to deal with them, a reputation system without courts.

Where it shows up

eBay’s Feedback Forum was a reputation system in exactly this tradition, extended to strangers who might trade only once by making each trader’s history visible to all future partners. The shadow of the future was cast not by repeated dealings with the same partner but by the public score that every future partner would see. Amazon’s seller ratings, Airbnb’s two-way reviews and Uber’s ratings all apply the same principle.

The weaknesses of numerical reputation also showed up. Because ratings were public and reciprocal, people feared retaliation and avoided negative ratings, so scores clustered near the top. Over time ratings inflated, as Filippas, Horton and Golden documented, and a score of 4.8 out of 5 came to signal mediocrity. And because reputation had value, it attracted fraud: fake reviews, bought ratings, review suppression. The FTC’s 2024 rule is an attempt to protect the memory from forgery.

The strongest objection

The objection is that numerical reputation is a poor substitute for the rich, contextual judgement that bazaar relationships produced, and that ratings can be cruel, as when a few bad reviews destroy a small business or a driver is deactivated for a low score caused by things beyond their control. A number cannot capture why someone was rated badly. These criticisms are fair. But they apply to a system that made trade between strangers possible at a scale the bazaar never reached. The question is how to design reputation so that it remains informative and fair, not whether to have it.

What it means today

Ratings are useful, and less informative than they look. When using them, it is worth reading the distribution and the comments rather than the average, and asking who rated and why. When designing them, it is worth knowing that reciprocal, public ratings inflate, and that private or blind ratings, where neither side sees the other’s until both have rated, give more honest signals. Inside organisations, performance ratings face the same pressures, and inflate for the same reasons.

3. Getting both sides at once

The idea in plain terms

A marketplace is useful to buyers only if there are many sellers, and useful to sellers only if there are many buyers. A new marketplace has neither, and each side waits for the other. This is the chicken-and-egg problem. Once solved, it reverses: the largest marketplace attracts more buyers and sellers, and becomes harder to compete with.

The tradition

The economist Jeffrey Rohlfs analysed in 1974 how the value of a telephone service depends on how many others subscribe, and showed that such services face a critical mass below which they cannot sustain themselves. In 1985 the economists Michael Katz and Carl Shapiro formalised network effects: when the value of a product to each user rises with the number of other users, markets tend to tip towards a single winner, and expectations about who will win become self-fulfilling. The theory of two-sided markets, developed by Rochet and Tirole in the early 2000s, extended this to platforms serving two groups whose value to each other drives participation.

Where it shows up

Every marketplace in this history had to cross the threshold. eBay began with collectors, a community that already wanted to trade with each other. Amazon brought its existing customers to independent sellers, so that the buyer side was already present. Taobao attracted sellers with free listings, solving the seller side first. Airbnb was famously small for its first years, and its founders photographed hosts’ homes themselves to make listings attractive. Once each crossed the threshold, it grew rapidly, and in most countries a few marketplaces came to dominate. The network reversal pattern describes the opposite movement, when a key group leaves, as eBay’s sellers did in China.

The strongest objection

The objection is that network effects in marketplaces are weaker than they appear. Sellers and buyers can use several marketplaces at once, a practice economists call multi-homing, so no marketplace has a lock on either side. New marketplaces keep appearing: Etsy for handmade goods, Shopify enabling independent shops, Meesho in India for small resellers, Temu and Shein in recent years. This is a fair point, and it explains why marketplace competition has not ended. But the largest marketplaces combine network effects with logistics, data and advertising, which reinforce one another, and their positions have proved durable.

What it means today

For anyone building a marketplace, the hardest problem is the first one: which side to attract first, and how. The history suggests starting with a group that already wants to trade, or bringing one side from an existing business. For sellers, the network effect that makes the largest marketplace attractive is the same force that makes leaving it difficult. And for regulators, markets with strong network effects tend towards concentration, which is why competition authorities treat the largest marketplaces differently from shops.

4. The market as private government

The idea in plain terms

A marketplace writes rules, enforces them, settles disputes, sets the equivalent of taxes in its fees, and can exclude people from economic life on its territory. These are functions of government. But a marketplace is a private company, accountable to its shareholders, not to the merchants who live under its rules. When a large share of an economy runs through such companies, a significant part of economic life is governed privately.

The tradition

In 2017 the American philosopher Elizabeth Anderson published Private Government, an argument about the workplace. She pointed out that employers exercise authority over employees that would be considered a form of government if a state did it: regulating their speech, dress, movements and time, often with little accountability and few rights of appeal. Anderson argued that political philosophy, focused on the state, had neglected these private governments, and that the ideal of free markets, as originally imagined by thinkers such as Adam Smith, assumed a world of independent producers rather than large organisations commanding many people.

The idea extends naturally to platforms. The legal scholar Kate Klonick and others have described content moderation on social media as a form of private governance. Marketplaces govern economic conduct: what can be sold, at what price, under what conditions, with what penalties.

Older traditions offer criteria for good governance. The ideal of the rule of law, articulated by thinkers from Aristotle to the British jurist A. V. Dicey in the nineteenth century, holds that power should be exercised according to general, public, stable rules, applied equally, with fair procedures for those affected. The principle of natural justice holds that people should be heard before being punished and that decisions should give reasons.

Where it shows up

Every large marketplace in this history became a governor. Amazon’s seller policies, performance metrics and suspension procedures, Taobao’s rules on counterfeits, Flipkart’s standards for delivery and returns, and Airbnb’s rules for hosts and guests all function as law for millions of businesses and people. Their fees act as taxes. Their search rankings and the Buy Box decide who prospers. Sellers suspended without clear reasons, and with limited means of appeal, have long complained of arbitrary rule. Regulation increasingly imposes rule-of-law requirements on platforms: the European Union’s Platform to Business Regulation of 2019 required online intermediaries to publish clear terms, explain ranking parameters and give reasons for suspensions.

The strongest objection

The objection is that marketplaces are not governments, because sellers choose to join them and can leave, whereas citizens cannot easily leave a state. Marketplace rules are terms of a voluntary contract, and competition among platforms disciplines them. Moreover, platforms’ enforcement against fraud and counterfeits protects buyers and honest sellers, which is the main reason marketplaces work at all. These points are important. But the voluntariness is limited when a marketplace controls most of a seller’s customers, and the power to set terms is very unequal. Anderson’s argument about employment, that the right to quit does not make an employer’s authority accountable, applies here too.

What it means today

For sellers, a marketplace should be treated as a jurisdiction: understand its rules, its enforcement and its appeals, and limit dependence on any one. For those running platforms, the legitimacy of their rules matters for the business, not only for regulators: clear rules, consistent enforcement and fair procedures retain sellers and reduce the pressure for outside regulation. The question every large platform faces is whether it will govern well voluntarily or be made to.

5. Referee and player

The idea in plain terms

When the company that runs the market also sells in it, it has two roles with conflicting interests. As referee, it should treat all sellers fairly. As player, it wants its own products to win. It also has information about every rival’s sales. The temptation to use the referee’s power for the player’s benefit is obvious, and difficult to detect from outside.

The tradition

The legal principle nemo iudex in causa sua, no one should be a judge in their own cause, is among the oldest in Western law, recorded in Roman law and central to English common law’s idea of natural justice. It holds that a decision is tainted if the person making it has an interest in the outcome, whether or not they actually decide unfairly, because fairness must be seen as well as done.

The law of fiduciary duties expresses a related idea. Those who act for others, such as trustees, agents and company directors, must not put themselves in a position where their own interests conflict with those of the people they serve, and must not profit secretly from their position.

In competition law, these ideas take the form of concerns about vertical integration and self-preferencing. In 2017 Lina Khan, then a law student and later chair of the FTC, argued in “Amazon’s Antitrust Paradox” that Amazon’s dual role as marketplace and seller, and the data it collected, gave it advantages that the prevailing antitrust focus on consumer prices failed to capture. The search essay describes the parallel case of Google placing its own shopping service in its search results.

Where it shows up

Amazon’s dual role was the subject of the European Commission’s investigations, which ended with Amazon’s commitments not to use sellers’ non-public data for its retail business and to treat offers equally in the Buy Box. The FTC’s case raises related claims about how Amazon treats sellers and its own fulfilment business. India’s rules barring foreign-owned marketplaces from selling through affiliated sellers were a blunter response to the same concern: they separated the referee from the player by law.

The strongest objection

The objection is that integration benefits customers. Amazon’s own retail offers and private-label products often have lower prices; its logistics are faster; competing with sellers keeps prices down. Supermarkets have sold their own brands next to national brands for a century without anyone calling them judges in their own cause. Forbidding platforms from selling would raise prices and reduce quality. This is a serious argument, and the commitments the EU accepted did not ban Amazon from selling. The answer of the tradition is not that a referee may never play but that it must not use the referee’s powers, the data and the rules, to win.

What it means today

For sellers on a marketplace that competes with them, data about their products is visible to a rival; strategies should assume so. For companies running platforms, the safest course is to build walls between the marketplace and the retail business, and to be able to show that the same rules apply to both. For regulators, the history suggests that conduct rules, such as data separation and equal treatment, can address the conflict without breaking up businesses, if they are enforced.

6. Does commerce make people gentler?

The idea in plain terms

One of the oldest hopes about markets is that trade civilises. When people depend on each other for exchange, they have reasons to be honest, polite and peaceful. Online marketplaces put this hope to a large test: they connect strangers across the world, asking them to trust each other for mutual gain.

The tradition

In The Spirit of the Laws (1748) the French philosopher Montesquieu wrote that commerce cures destructive prejudices and that wherever there is commerce, manners are gentle. This idea, known as doux commerce (gentle commerce), held that trade replaces violent passions with calm interests and binds nations together through mutual need. Adam Smith shared a version of it, arguing that commerce introduced order and good government, and with them the liberty and security of individuals. In 1977 the economist Albert Hirschman traced the history of this idea in The Passions and the Interests, showing how early modern thinkers came to see the pursuit of interest as a check on more dangerous passions.

The opposite view has an equally long history. Critics from Rousseau to Marx argued that commerce corrodes virtue, turning human relations into calculations. In What Money Can’t Buy (2012) the philosopher Michael Sandel argued that markets can crowd out non-market norms, so that putting a price on something can change its meaning. The economist Samuel Bowles, in The Moral Economy (2016), reviewed experimental evidence that incentives sometimes undermine the motives they are meant to support.

Where it shows up

Online marketplaces offer evidence for both views. eBay’s founding hope was that most people were honest, and the overwhelming majority of transactions between strangers on marketplaces succeed. Buyers in one country trust sellers in another they will never meet. Airbnb hosts admit strangers into their homes. In that sense, marketplaces have extended the gentle commerce Montesquieu imagined to a global scale. But the trust is supported by ratings, escrow, guarantees and enforcement, and fraud, fake reviews and counterfeit goods remain large problems. The honesty of marketplace trading is less a sign that commerce has made people gentle than that institutions have made honesty pay.

The strongest objection

The objection to the gentle-commerce reading is that it confuses institutional enforcement with moral improvement. People behave well on marketplaces because they are watched and rated, not because trade has changed them. The objection to the corrosion reading is that it romanticises the communities that markets replaced, which were often closed, unequal and hostile to strangers. Both objections have force. The history of marketplaces suggests that the relationship between trade and virtue runs through institutions: well-designed markets reward honesty, and badly designed ones reward deception.

What it means today

For those building marketplaces, the design of incentives determines the character of the community: what behaviour is rewarded, what is punished, what is seen. For participants, the trust that marketplaces make possible is real, and it depends on institutions that can fail. And for societies, the question is not whether commerce is good or bad for character, but which rules make it so.

7. The endless aisle and the burden of choice

The idea in plain terms

A shop offers a few hundred or a few thousand items. A large marketplace offers hundreds of millions. That is wonderful for anyone looking for something unusual, like a broken laser pointer. It also means that buyers face more options than they can compare, many nearly identical, and must rely on rankings, reviews and recommendations to choose. Choice was once limited by shelf space. Now it is limited by attention.

The tradition

In 2003 the economists Erik Brynjolfsson, Yu Hu and Michael Smith estimated that the greater variety of books available online gave consumers a benefit several times larger than the benefit of lower prices, because people could find titles that physical shops did not stock. In 2004 Chris Anderson, editor of Wired, called this the long tail: online retailers could profit from selling small quantities of very many items.

On the other side, the psychologist Barry Schwartz argued in The Paradox of Choice (2004) that more options can make people less satisfied, more anxious and more prone to regret. A widely cited experiment by Sheena Iyengar and Mark Lepper in 2000 found that shoppers offered 24 jams were less likely to buy than those offered six. Later research tempered this: a 2010 meta-analysis by Benjamin Scheibehenne and colleagues found that the average effect of choice overload across many studies was close to zero, with large effects only under certain conditions, such as when options are hard to compare. Herbert Simon’s idea of satisficing, choosing the first option that is good enough rather than searching for the best, describes how people actually cope, an idea explored in Bounded Rationality.

Where it shows up

The endless aisle is the defining feature of online marketplaces. eBay made collectibles of every kind findable. Amazon’s third-party sellers expanded its catalogue far beyond what it could stock. Taobao offered almost anything made in China. The burden of choosing moved to search ranking, sponsored listings and reviews. Marketplaces filled with near-identical products from many sellers, sometimes with interchangeable brand names, and buyers relied on ratings, which inflated, and on rankings, which sellers could pay to influence through advertising.

The strongest objection

The objection is that choice is mostly good, and the evidence for choice overload is weak, as the meta-analysis showed. People use filters, reviews and brands to narrow choices, and they value being able to find exactly what they want. This is largely right. The concern is not choice itself but who shapes it: when the tools that narrow the choice are controlled by the marketplace, which earns money from sellers’ advertising, the choice is guided by interests the buyer does not see.

What it means today

For buyers, the useful habit is to decide what matters before searching, and to treat rankings and sponsored results as advertisements, which many are. For sellers, visibility in an endless aisle has become a cost to be paid, through advertising and fees, not a reward for being good. For organisations, the same dynamic applies to procurement and hiring platforms: abundant options shift power to whoever controls the filter.

8. Who owns a reputation?

The idea in plain terms

A seller who has spent ten years earning thousands of good reviews on a marketplace has built something valuable. But the reviews belong, in practice, to the marketplace. The seller cannot take them to another platform or to its own website. If the seller is suspended, or leaves, the reputation disappears. The thing that most makes a small business trustworthy online is held by someone else.

The tradition

John Locke argued in his Second Treatise of Government (1689) that people acquire property in things by mixing their labour with them: whoever cultivates a field makes it theirs, because the labour was theirs. On a Lockean view, a seller’s reputation, earned by years of honest dealing, looks like the seller’s property. Against this, the marketplace can argue that it built the system that made the reputation possible and meaningful, and that the reviews are also the work of the buyers who wrote them, so that no single party owns them.

Economists have studied what happens when an asset that one party creates is locked to another’s platform. Williamson’s idea of asset specificity, described in the essay on subscription software, applies: an investment that has value only within one relationship leaves its owner exposed to the other party. Reputation earned on one marketplace is among the most specific assets a seller can have.

Data protection law has begun to treat some data as portable. The European Union’s General Data Protection Regulation of 2018 gave individuals a right to receive their personal data and move it to another provider. The Digital Markets Act extended portability obligations to the largest platforms. Researchers and policymakers have debated whether reputation, such as ratings earned by gig workers and sellers, should also be portable, so that people could carry their record from one platform to the next.

Where it shows up

eBay’s feedback score, Amazon’s seller ratings and Airbnb’s reviews are among each platform’s most valuable assets, because they are what make strangers trustworthy and because they keep sellers and hosts from leaving. A seller who considers moving to a rival marketplace must start again with no reviews, facing the same lemons problem as a newcomer. Suspension destroys the reputation outright. The network reversal pattern shows how quickly a marketplace can lose sellers when they do leave; the lock on reputation is one reason they rarely do.

The strongest objection

The objection is that portable reputation would be easy to manipulate and hard to interpret. Ratings mean different things on different platforms, with different scales, standards and fraud controls; a score imported from one platform could not be trusted on another. Marketplaces invest heavily in detecting fake reviews, and that investment would be undermined if reputations could be carried from less careful platforms. These are real difficulties. But they argue for careful design, such as verified, standardised summaries, rather than for leaving the asset entirely with the platform.

What it means today

For sellers, reputation on a marketplace is borrowed capital. It is worth building reputation in channels the seller controls as well: its own website, direct relationships with repeat customers, a brand that buyers search for by name. For marketplaces, holding reputations hostage keeps sellers but breeds resentment and invites regulation. And for policymakers, the question of who owns a reputation is becoming one of the central questions of platform fairness.

Bringing the threads together

Online marketplaces solved Akerlof’s problem among strangers by building institutions that made quality visible: reputation as public memory, escrow and guarantees that moved risk to the platform, and fulfilment that let strangers borrow a trusted name. They crossed the chicken-and-egg threshold and then tipped towards a few winners. They became private governments over millions of merchants, and in some cases players in the games they refereed. They extended the old hope of gentle commerce, by making honesty pay rather than by changing people. They replaced the limits of the shelf with the limits of attention, shaped by rankings the platforms control. And they kept the reputations their sellers earned, which made those sellers both trustworthy and tied.

The bazaar without walls

The buyer of the broken laser pointer could never have found it in a shop. eBay found it for him, and in doing so showed what an online marketplace could do: connect strangers with improbable wants across any distance. The institutions that made this possible, ratings, escrow, guarantees, fulfilment, are among the most useful inventions of the internet era, and they have opened markets to millions of small sellers who could never have reached customers on their own.

The history also shows what was traded away. The bazaar was a public space with many owners; the online marketplace is a private one with one. Its rules are written by a company that may also compete in it. The working principles that follow are not complicated. Understand what mechanism is making a stranger trustworthy, and what it costs. Read reputations critically, because they inflate. Treat a marketplace you depend on as a jurisdiction, and keep a way to reach customers that does not run through it. If you run a market, keep the referee separate from the player, and govern as if the merchants could leave, even if they cannot. A marketplace’s first achievement is to make strangers trust each other. Its lasting test is whether the strangers can trust the marketplace.

Selected bibliography

• Locke, John. Second Treatise of Government. 1689.
• Montesquieu. The Spirit of the Laws. 1748.
• Hume, David. A Treatise of Human Nature, Book III. 1740.
• Smith, Adam. An Inquiry into the Nature and Causes of the Wealth of Nations. 1776.
• Dicey, A. V. Introduction to the Study of the Law of the Constitution. 1885.
• Akerlof, George A. “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism.” Quarterly Journal of Economics 84, no. 3 (1970): 488–500.
• Spence, Michael. “Job Market Signaling.” Quarterly Journal of Economics 87, no. 3 (1973): 355–374.
• Rohlfs, Jeffrey. “A Theory of Interdependent Demand for a Communications Service.” Bell Journal of Economics and Management Science 5, no. 1 (1974): 16–37.
• Hirschman, Albert O. The Passions and the Interests. Princeton University Press, 1977.
• Geertz, Clifford. “The Bazaar Economy: Information and Search in Peasant Marketing.” American Economic Review 68, no. 2 (1978): 28–32.
• Axelrod, Robert. The Evolution of Cooperation. Basic Books, 1984.
• Katz, Michael L., and Carl Shapiro. “Network Externalities, Competition, and Compatibility.” American Economic Review 75, no. 3 (1985): 424–440.
• Greif, Avner. “Reputation and Coalitions in Medieval Trade: Evidence on the Maghribi Traders.” Journal of Economic History 49, no. 4 (1989): 857–882.
• Iyengar, Sheena S., and Mark R. Lepper. “When Choice Is Demotivating.” Journal of Personality and Social Psychology 79, no. 6 (2000): 995–1006.
• Resnick, Paul, and Richard Zeckhauser. “Trust Among Strangers in Internet Transactions: Empirical Analysis of eBay’s Reputation System.” In The Economics of the Internet and E-Commerce, ed. Michael R. Baye. Elsevier, 2002.
• Brynjolfsson, Erik, Yu (Jeffrey) Hu and Michael D. Smith. “Consumer Surplus in the Digital Economy.” Management Science 49, no. 11 (2003): 1580–1596.
• Rochet, Jean-Charles, and Jean Tirole. “Platform Competition in Two-Sided Markets.” Journal of the European Economic Association 1, no. 4 (2003): 990–1029.
• Schwartz, Barry. The Paradox of Choice. Ecco, 2004.
• Anderson, Chris. The Long Tail. Hyperion, 2006.
• Scheibehenne, Benjamin, Rainer Greifeneder and Peter M. Todd. “Can There Ever Be Too Many Options? A Meta-Analytic Review of Choice Overload.” Journal of Consumer Research 37, no. 3 (2010): 409–425.
• Sandel, Michael J. What Money Can’t Buy. Farrar, Straus and Giroux, 2012.
• Bowles, Samuel. The Moral Economy. Yale University Press, 2016.
• Anderson, Elizabeth. Private Government. Princeton University Press, 2017.
• Khan, Lina M. “Amazon’s Antitrust Paradox.” Yale Law Journal 126, no. 3 (2017): 710–805.
• Filippas, Apostolos, John J. Horton and Joseph M. Golden. “Reputation Inflation.” Marketing Science 41, no. 4 (2022): 733–745.

Research limitations

The story of the broken laser pointer is Omidyar’s own and is widely repeated; the date of AuctionWeb’s launch varies by a few days across sources. The figure of about 99 percent positive feedback comes from Resnick and Zeckhauser’s analysis of 1999 data as reported secondarily. The market-share figures for Taobao and eBay in China are from one analysis and should be treated as approximate. The date when Flipkart introduced cash on delivery is reported as around 2010; accounts differ. The description of Amazon’s fees as rising over the 2010s relies on analyses by sellers and researchers that Amazon disputes, and no figures are given. The 2020 press reports about Amazon employees’ use of seller data are described as allegations; Amazon said the practice violated its policies. The FTC’s antitrust case against Amazon has not been decided, and its claims are allegations; the Prime settlement involved no admission. The account of India’s marketplace rules is simplified. The philosophical connections to Geertz, Akerlof, Locke, Williamson, Spence, Stiglitz, Axelrod, Hume, Greif, Rohlfs, Katz and Shapiro, Rochet and Tirole, Elizabeth Anderson, Klonick, Dicey, Khan, Montesquieu, Smith, Hirschman, Sandel, Bowles, Brynjolfsson, Chris Anderson, Schwartz, Iyengar and Lepper, Scheibehenne and Simon are the author’s interpretation; their works are paraphrased apart from short, attributed phrases. The decisions in the decisions box are judged by the author, separating what could be known at the time (process) from what happened (call).

Footnotes

1. Wikipedia, “eBay”; Academy of Achievement, “Pierre Omidyar”. ↩ ↩²

2. Clifford Geertz, “The Bazaar Economy: Information and Search in Peasant Marketing,” American Economic Review 68, no. 2 (1978). ↩

3. George A. Akerlof, “The Market for ‘Lemons’: Quality Uncertainty and the Market Mechanism,” Quarterly Journal of Economics 84 (1970). ↩

4. Paul Resnick and Richard Zeckhauser, “Trust Among Strangers in Internet Transactions” (2002); see also the Harvard Gazette, February 2003. ↩

5. Apostolos Filippas, John J. Horton and Joseph M. Golden, “Reputation Inflation”, Marketing Science (2022); NBER Working Paper 25857. ↩

6. “FTC issues final rule banning fake and misleading consumer reviews and testimonials,” Wilson Sonsini, August 2024. ↩

7. Wikipedia, “Amazon Marketplace”. ↩

8. “Amazon steers third-party seller share to all-time high,” Marketplace Pulse; “Amazon is regaining 1P unit share,” Marketplace Pulse. ↩

9. Wikipedia, “Taobao”; ICMR, “eBay’s problems in China”; Knowledge at Wharton. ↩ ↩²

10. “Walmart–Flipkart deal: a timeline,” Quartz. ↩

11. “Walmart pumps in $1.2 bn to fund Flipkart’s operations,” Inc42, on India’s rules for foreign-owned marketplaces. ↩

12. “Walmart buys $16B controlling stake in India’s Flipkart,” Associated Press via News 3 LV, May 2018. ↩

13. “Airbnb offers unconditional apology, and $50,000 insurance guarantee,” TechCrunch, 1 August 2011. ↩

14. “European Commission accepts commitments by Amazon barring it from using marketplace seller data,” Society for Computers and Law, December 2022; the 2020 allegations were first reported by the Wall Street Journal in April 2020. ↩

15. “Amazon reaches $2.5 billion settlement with FTC over Prime program,” CNBC, 25 September 2025; NPR; on the antitrust case, Retail Dive. ↩