Controlling Supply Quality in a Marketplace

By Brian Tr · Updated Oct 6, 2026

A marketplace doesn't make the thing its customers buy. Its suppliers do. That's the whole bargain, and it's also the whole risk: the operator is judged on quality it doesn't directly control.

Steven Tadelis, an economist who worked with eBay, opens one of his papers on this exact point. Platform markets, he and co-author Chris Nosko write, facilitate trade between anonymous buyers and sellers and don't control key variables such as inventory, price or transaction quality, which means the quality of sellers on the platform varies. A retailer picks what goes on its shelves. A marketplace has to build a system that picks, or filters, or corrects after the fact.

This article is about that system. It sits in the supply section of this collection, which starts from the chicken-and-egg problem. Getting suppliers in the door is covered in marketplace supply acquisition, and getting them live is covered in marketplace seller onboarding. Here we deal with what comes after: deciding who gets in at all, setting the bar, watching performance, rewarding the best, and removing the worst.

Why supply quality doesn't fix itself

George Akerlof's 1970 paper, "The Market for 'Lemons'" (Quarterly Journal of Economics, vol. 84, no. 3, pp. 488-500), is the standard starting point. His argument runs like this. When buyers can't tell good from bad before they buy, they pay a price that reflects the average. Sellers of good products then get less than they're worth and drop out, which lowers the average again. In his words, the "bad" cars tend to drive out the good. Economists call this adverse selection.

Akerlof also made a point that fits marketplaces closely. He notes that where buyers judge quality by a group statistic, there's an incentive for sellers to market poor quality, since the returns for good quality accrue mainly to the entire group whose statistic is affected rather than to the individual seller. Replace "group statistic" with "the marketplace's reputation" and you have the supply-quality problem in one sentence. One careless supplier costs the whole platform something, and the supplier doesn't pay for it.

Nosko and Tadelis give this an empirical form. Their abstract describes a reputational externality across sellers: buyers may draw conclusions about the quality of the platform from a single transaction. In their eBay data, a better experience with a seller made a buyer more likely to keep buying on the platform, and buyers with less history reacted more strongly than experienced ones. That's the cost of a bad supplier that doesn't show up in the supplier's own numbers.

Akerlof's closing section lists what societies do about it. He calls them counteracting institutions: guarantees, brand names, chains, and licensing. A marketplace is a modern collection of the same tools. The levers below map onto his list, with a few additions that only exist because a platform can see every transaction.

Lever 1: entry curation

The first decision is who gets in. It's useful to think of it as a spectrum with four positions, not a yes/no choice.

Toptal is a documented example of the vetted end. Its site says that of the thousands of applications it sees each month, typically fewer than 3% are accepted, and that the full screening process takes between three and eight weeks. A marketplace that works this way is making a deliberate bet: fewer suppliers, higher trust, and a customer willing to pay for the filter. Compare that with the open-listing model described in what is a marketplace business, where the operator provides search, payments and a review system and lets sellers sign up with little vetting.

The spectrum maps loosely onto the operator-involvement axis we cover in managed marketplace. The more you curate on entry, the more your marketplace behaves like a managed one, and the more each added supplier costs you.

One curation tool applies to almost everyone: verifying who the supplier is. It's cheap, it filters out bad actors, and in some markets it's the law, as the regulatory section below shows.

Lever 2: standards and policies

Curation decides who gets in. Standards decide what they have to do once they're there. Typical standards fall into a few categories:

• Listing standards: accurate descriptions, photos, prohibited items, required information.
• Service standards: response times, fulfilment or delivery windows, cancellation behavior.
• Conduct standards: no taking transactions off the platform, no fake reviews, no harassment.
• Safety and legal standards: licensing, insurance, product compliance.

Two rules make standards work. First, write them as measurable behavior, not attitudes. "Respond to messages within 24 hours" can be enforced. "Provide excellent service" can't. Second, publish them. A supplier who is penalized for a rule they never saw will blame the platform, not themselves, and the next supplier will notice.

Standards are also a form of the guarantee Akerlof describes. A buyer who knows the platform enforces a rule is paying for that assurance, even if they never read it.

Lever 3: performance monitoring

Standards need measurement. Marketplaces tend to track supplier behavior in four categories, whatever the specific metric names are:

Ratings are the last row for a reason. They're one input to supply quality, not the whole picture, and Nosko and Tadelis show why leaning on them alone is risky. In their eBay data, the standard "percent positive" seller measure has a mean of 99.3% and a median of 100%, which leaves almost no room to tell sellers apart. Out of more than 44 million U.S. transactions in October 2011, they report that only 0.39% had negative feedback, while more than 1% had an actual dispute ticket opened. Behavior and outcome data told the platform something the star rating didn't.

So build your monitoring around the first three categories, which come from what the supplier did, and treat ratings as a supporting signal. The first three also react faster. A cancellation shows up immediately. A review may never come.

Monitoring also needs to be fair to the supplier. Define each metric clearly, show suppliers their own numbers, and exclude events they couldn't control. Shopee's Singapore seller education hub, for example, defines the non-fulfilment rate as the percentage of orders cancelled or returned in the past 7 days, counting cancellations and returns only when the seller is at fault, and lets sellers appeal on certain metrics.

Lever 4: tiered status programs

Once you measure behavior, you can reward it. A status tier gives good suppliers a badge, better visibility, or other perks, and gives everyone else a target to aim at. It turns quality control from a punishment into a ladder.

Three companies publish the criteria on their own help pages. These are their stated thresholds as we retrieved them, and platforms revise them, so check the current page before relying on any number.

Airbnb Superhost. Airbnb's help page lists a 4.8 or higher overall rating, at least 10 reservations (or 3 reservations totalling at least 100 nights), a cancellation rate under 1% (with exceptions for major disruptive events or other valid reasons), and responding to 90% of new messages within 24 hours. Status is assessed quarterly, on January 1, April 1, July 1 and October 1, using the past 12 months of performance.

eBay seller levels. eBay's global seller performance policy describes three levels: Top Rated, Above Standard and Below Standard. For Top Rated, the transaction defect rate may be no more than 0.5% of transactions, associated with no more than 3 different buyers. A seller is evaluated as Below Standard when defects exceed 2% of transactions, and only if they're associated with more than 4 different buyers. That last condition is a good design detail: it stops a single difficult buyer from sinking a good seller.

Shopee Preferred Seller (Singapore). Shopee's seller education hub lists a set of onboarding targets: at least 10 unique buyers and 30 net orders in the last 30 days, a chat response rate of 85% or more, a shop rating of 4.6 or higher, and non-fulfilment and late shipment rates of 4.99% or less over the last 7 days, plus a penalty-points limit. Benefits include a badge, a dedicated search filter, higher listing limits and more free broadcast messages. Sellers who meet all requirements are onboarded weekly, and performance continues to be monitored afterwards, so status can be removed.

What do these programs have in common?

1. They use behavior metrics, not just ratings. Cancellation, response and defect rates appear in every one.
2. They have a volume floor. Airbnb and Shopee both require minimum activity, because a rate based on two transactions says nothing.
3. They re-evaluate on a schedule. Status is earned and can be lost, so it stays meaningful.
4. The reward is visibility. A badge, a search filter, higher limits. The platform pays in attention, which costs it little.

Lever 5: enforcement and offboarding

A standard nobody enforces is a suggestion. The enforcement ladder usually has several rungs:

1. Warning and coaching. Tell the supplier which metric slipped and how to fix it.
2. Reduced visibility. Rank them lower in search, remove a badge, or cap listings.
3. Restrictions. Hold payouts, limit new listings, or suspend certain features.
4. Suspension or removal. Reserved for repeated or serious violations.

eBay's policy page says the limitations on Below Standard sellers, such as additional final value fees, can vary by program. The pattern across platforms is consistent: start soft, escalate in steps, and keep the final step for clear cases.

Three practices keep enforcement credible:

• Give an appeal route. Mistakes happen, and a platform that can't correct them loses good suppliers along with bad ones.
• Explain the decision. Tell the supplier what rule they broke and what evidence the platform used.
• Apply rules to everyone. Exempting a large seller once teaches every other seller that the rules are negotiable.

Offboarding is the hardest part emotionally, because every supplier you remove is supply you worked to get. Do it anyway when the numbers justify it. The Akerlof logic says why: if bad suppliers stay, good ones leave or lower their effort, and the average falls.

The regulatory angle: what the EU requires

If your marketplace lets consumers in the European Union buy from traders, supply quality isn't just a business choice. The EU's Digital Services Act, Regulation (EU) 2022/2065, puts specific duties on online platforms that allow consumers to conclude distance contracts with traders. Several map directly onto the levers above.

• Entry curation (Article 30). Platforms must ensure traders can only use the service to offer products to consumers in the Union if the platform has first obtained certain information: name, address, phone and email; an identity document or electronic identification; payment account details; trade register details where applicable; and a self-certification committing to offer only compliant products. The platform must make best efforts to assess whether this information is reliable and complete. If it has reason to believe the information is inaccurate or out of date, it must ask the trader to correct it and swiftly suspend the trader until they comply.
• Compliance by design (Article 31). The platform's interface has to let traders provide the pre-contractual and product safety information the law requires.
• Right to information (Article 32). When the platform learns that an illegal product was offered, it must inform consumers who bought it, where it has their contact details, covering purchases made within the previous six months.
• Enforcement with due process (Articles 17 and 23). Restrictions such as suspending an account or payments require a clear statement of reasons, and platforms must warn before suspending users who frequently provide manifestly illegal content.

Article 30 also gives traders the right to lodge a complaint if they're refused or suspended, which echoes the appeal route recommended above. Micro and small enterprises are excluded from this section under Article 29, so size matters. Check how the rules apply to your company before you assume either way, and get legal advice for your specific case.

The core trade-off: growth versus quality, early on

Every lever above costs growth. Vetting slows signups. Standards lose suppliers who can't meet them. Enforcement removes supply you paid to acquire. And early in a marketplace's life, supply is the scarce thing. The chicken-and-egg problem says you need enough of both sides to make the platform useful, and strict quality control makes the supply side harder to fill.

The opposite error is just as costly. Open the doors to everyone and the first buyers meet the worst suppliers. Nosko and Tadelis's finding that one bad transaction can sour a buyer on the whole platform is the warning here: your earliest buyers are the ones with the least reason to forgive you. In their experiment, which changed eBay search results for a random sample of users to favor sellers with a higher estimated quality score, treated buyers were significantly more likely to return and purchase again. The measured gap in return probability was about 0.3 percentage points, which looks small per buyer but came from a sample of millions of searches.

A workable approach for an early marketplace:

1. Be strict where the stakes are high. If a bad supplier can hurt a buyer (safety, large payments, in-person service), gate entry early.
2. Be loose where mistakes are cheap. For low-stakes, easily reversible transactions, open entry plus monitoring is fine.
3. Seed the best suppliers by hand. A small number of excellent suppliers does more for early quality than a large number of average ones. Our guide to supply-first vs demand-first covers when this fits.
4. Instrument from day one. Even if you don't enforce yet, record cancellations, response times and defects, so you have the data when you need it.
5. Tighten as you scale. What works at 50 suppliers (a personal call) won't work at 5,000 (automated thresholds and tiers).

The same logic is why a minimum viable product approach works for supply quality: begin with the cheapest control that prevents the worst outcomes, and add controls as you see the failures that actually happen.

Which levers fit which stage

Key Facts: Supply Quality in Marketplaces

• Akerlof's 1970 paper argued that when buyers can't judge quality, bad products tend to drive out good ones (Quarterly Journal of Economics, vol. 84, no. 3).
• Nosko and Tadelis found that eBay's percent-positive seller measure had a mean of 99.3% and a median of 100%, while over 1% of transactions in October 2011 had a dispute ticket and only 0.39% had negative feedback.
• Toptal states that typically fewer than 3% of applicants are accepted after a screening process of three to eight weeks.
• Airbnb's help page lists a 4.8 rating, under 1% cancellation and 90% message response within 24 hours among its Superhost criteria, assessed quarterly.
• Shopee Singapore's Preferred Seller targets include a chat response rate of at least 85% and a shop rating of at least 4.6.
• The EU's Digital Services Act requires covered platforms to collect and make best efforts to check trader information before the trader can sell (Article 30).

Common mistakes

• Metrics without a volume floor. One cancellation out of two orders is a 50% rate. Require a minimum number of transactions before judging.
• Punishing outcomes the supplier can't control. Airbnb's page, for instance, exempts major disruptive events from its cancellation rate.
• Enforcing rules on small suppliers but not large ones. It's the fastest way to lose everyone else's trust.

Related Reading

• The Marketplace Chicken-and-Egg Problem
• Marketplace Supply Acquisition
• Marketplace Seller Onboarding
• Managed Marketplace
• What Is a Marketplace Business?
• Two-Sided Markets