Latest / Marketplace Businesses with Fexingo: Two-Sided Networks, Liquidity, and Take Rates / How Amazon Marketplace Solved the Empty Box Problem
Transcript
- Lucas: Let me set the scene. It's 1999. Amazon has just opened its marketplace to third-party sellers — about two hundred of them, mostly used-book dealers. Jeff Bezos stands in front of a whiteboard and says the goal is to offer 'every book, ever printed, in any language.' And it works. Selection explodes. But then something ugly happens. Luna: The empty box problem. Lucas: Exactly. Buyers start receiving packages that are just... empty. Or the wrong item. Or the item never ships. And there's no recourse — Amazon says 'you didn't buy this from us, you bought it from Bob's Used Books. Take it up with Bob.' But Bob doesn't answer. So the buyer leaves and never comes back. Luna: And that kills marketplace liquidity faster than anything. If buyers don't trust the transaction, they won't transact. And if they don't transact, sellers leave. The whole two-sided network collapses. Lucas: Right. So Amazon had to solve trust before it could scale its take rate. And the way they did it — the a to z Guarantee, payment escrow, seller performance metrics — became the template for every marketplace that followed. Let's walk through the mechanics. Luna: Let's start with the a to z Guarantee. That launched in 2000. What did it actually cover? Lucas: It said: if you buy an item from a third-party seller on Amazon, and the item doesn't arrive, arrives damaged, or isn't as described, Amazon will refund you. Up to $1,000 per claim. No questions asked — well, minimal questions. The key is that Amazon fronts the money. They don't wait to collect from the seller first. Luna: That's the trust mechanism. The buyer doesn't care who's at fault. They just want their money back. Lucas: Exactly. But Amazon can't just hand out refunds forever. So they paired the guarantee with payment escrow. When a buyer pays, Amazon holds the money. They don't release it to the seller until the buyer confirms delivery — or until a certain number of days pass without a claim. That gives Amazon leverage over the seller. Luna: And that's where the take rate comes in. Because once Amazon has the money in escrow, they can deduct their fee before the seller ever sees a dime. Lucas: Yes. The referral fee — typically 15% of the sale price, depending on category — comes out of escrow automatically. The seller never touches that money. So Amazon's take rate is essentially guaranteed. No invoicing, no chasing payments. The escrow system makes the fee collection frictionless. Luna: But the guarantee also gives Amazon the right to claw money back. If a claim is filed and Amazon decides the seller is at fault, they reverse the transfer from escrow. The seller is out the money, and Amazon keeps the fee. Lucas: Right. That's the teeth behind the trust promise. And it scales because it's automated. Amazon uses data — order defect rate, cancellation rate, late shipment rate — to decide which sellers are trustworthy. Sellers with good metrics get faster payments and lower scrutiny. Sellers with bad metrics get flagged, or suspended. Luna: So the performance data becomes a moat. A new marketplace can't replicate that overnight. They don't have the transaction history to build the trust algorithm. Lucas: Exactly. And that's why Amazon could raise its take rate over time. In 1999, the referral fee was around 6% on books. Today, the average across categories is 15%, with some categories like jewelry at 20%. Sellers accept it because the trust system drives conversion. A buyer is more likely to click 'buy' on Amazon than on some random site because they know they're protected. Luna: Let's put numbers on that. Amazon's third-party marketplace does about $400 billion in gross merchandise volume annually. If the average take rate is 15%, that's $60 billion in revenue. And the marginal cost of running the guarantee and escrow is tiny — it's mostly software. Lucas: Right. But here's the nuance: the guarantee has a $1,000 cap. Most transactions are well below that — the average selling price on Amazon is around $25. So the actual claim rate is low, maybe one to two percent of transactions. And when a claim does happen, Amazon often recovers from the seller. So the cost of the guarantee is a fraction of the fee revenue. Luna: It's a classic insurance model. You collect premiums — the referral fee — and pay out claims only when necessary. But the trust it builds increases the volume of insured transactions. It's a flywheel. Lucas: Exactly. And that's the insight for marketplace builders. If you want a high take rate, you can't just charge it. You have to earn it by solving a real problem for both sides. For buyers, the problem is risk. For sellers, the problem is access to buyers who trust the platform. Luna: If today was actually useful to you, the way these stay ad-free is listener support. You can find us at buy me a coffee dot com slash fexingo. Lucas: Yeah, it's a small way to keep the conversations going without any sponsor interruptions. Appreciate anyone who chips in. Luna: So let's talk about how other marketplaces copied — or failed to copy — this model. Lucas: Let's take eBay. eBay had a buyer protection program too, but it was different. eBay's guarantee was funded by a separate fee — not baked into the transaction fee. So sellers saw it as an upsell, not a feature. And eBay didn't hold escrow; the money went directly from buyer to seller. So when a claim happened, eBay had to try to pull money back from the seller, which was harder. Luna: And that's one reason eBay's take rate has stayed around 10-12%, while Amazon's is 15% and rising. The trust infrastructure enables the higher fee. Lucas: Right. And then look at newer marketplaces like Etsy. Etsy also uses escrow and a guarantee. Their take rate is about 6.5% — lower than Amazon. But Etsy's items are more idiosyncratic, handmade, often one-of-a-kind. The trust problem is different. A buyer on Etsy isn't just worried about non-delivery; they're worried about quality. So Etsy relies more on reviews and photos than on a financial guarantee. Luna: So the take rate isn't just a function of trust — it's a function of the type of trust needed. Commodity items like books or electronics: you need guarantee of delivery and authenticity. Unique items: you need quality assurance and seller story. Lucas: Exactly. And that's why marketplace design matters. If you're building a marketplace for used furniture, the empty box problem is real, but so is the 'this sofa is uglier than the photo' problem. You might need a refund policy that covers returns, not just non-delivery. That changes your cost structure and your take rate. Luna: Let's go back to Amazon's seller performance metrics. How did they evolve? Lucas: In the early 2000s, Amazon started tracking order defect rate — the percentage of orders that received negative feedback, a claim, or a chargeback. If your defect rate was above 1%, you were at risk of suspension. That created a strong incentive for sellers to ship fast and accurately. And it let Amazon offload enforcement to the algorithm. Luna: And the algorithm is constantly learning. It adjusts based on seasonality, category, even seller location. A seller in China might have a higher threshold for late shipments because of customs delays. But they also face higher scrutiny on counterfeit claims. Lucas: Right. And that's where the data moat becomes a competitive advantage. Amazon knows exactly which sellers are reliable in which contexts. A new marketplace can't replicate that without years of transaction data. So they have to start with a simpler trust model — maybe a flat guarantee with a lower cap, or manual claim review. Luna: Which is exactly what many early-stage marketplaces do. They offer buyer protection as a loss leader to build trust, then gradually raise fees as the data accumulates. Lucas: Yes. And eventually, if they get enough data, they can start offering premium services — like faster dispute resolution, or expedited payments for top sellers — and charge higher fees for those. That's how the take rate expands over time. Luna: So the playbook is: start with a simple guarantee, hold funds in escrow, collect a low fee, use that fee to fund the guarantee, and reinvest in data infrastructure. Then as trust grows, raise the fee. Lucas: Exactly. Amazon didn't start at 15%. They started at 6% on books, then gradually increased it as they proved the system worked. The guarantee was the foundation. Without it, they couldn't have built the marketplace at all. Luna: And now, with Amazon's marketplace generating over half of all e-commerce in the US, the trust moat is almost insurmountable. A competitor would need to process billions of transactions just to get the data to compete. Lucas: Right. And that's the real takeaway for this episode. Trust isn't a feature you bolt on. It's the architecture that determines whether your take rate is 2% or 20%. And the way you build it — guarantee, escrow, performance data — is the same whether you're selling books, rides, or freelance services. Luna: I think that's a good place to leave it. Next episode, we'll look at how marketplaces handle the opposite problem: too much supply and not enough demand. Lucas: That's a good one. See you then.