Latest / Marketplace Businesses with Fexingo: Two-Sided Networks, Liquidity, and Take Rates / How Marketplaces Use Counterparty Ratings to Build Trust
Transcript
- Lucas: If you've ever bought something on eBay, you probably noticed the seller has a feedback score — a number that tells you how reliable they are. But did you know eBay also lets sellers rate buyers? Luna: I actually didn't. I always assumed it was one-way — buyer reviews the seller, end of story. Lucas: That's exactly the misconception. Most people think of marketplace reviews as a product rating — you rate the item, the service, the driver. But the best marketplaces also rate the customer. They call them counterparty ratings, and they're critical for liquidity. Luna: So we're talking about platforms that rate both sides — the buyer and the seller? Lucas: Exactly. Let's start with eBay. When eBay launched in 1995, it had a simple feedback system: buyers and sellers could leave positive, neutral, or negative comments about each other. By 1996, they added a numeric score. That score became the primary trust signal on the platform. Luna: Right. If you're buying from someone with 10,000 positive ratings versus someone with 10, you trust the first person more. Lucas: But here's the twist: eBay's feedback was originally symmetrical. Both parties could rate each other. And that mattered because it solved a specific problem — what happens when a buyer is difficult? If a buyer constantly returns items, files false claims, or pays late, sellers need to know. Luna: So the seller can choose not to sell to that buyer again. That protects the seller's liquidity. Lucas: Exactly. In fact, in 2014, eBay made a significant change: they started allowing sellers to leave detailed seller ratings for buyers, including things like communication and payment timeliness. That was a direct response to the fact that 5 percent of buyers were causing 80 percent of the headaches for sellers. Luna: That's a classic power-law distribution. A small number of bad actors can poison the well for everyone. Lucas: Right. And if sellers get burned enough, they leave the platform. That kills liquidity. So eBay had to create a mechanism to protect the supply side. That's what counterparty ratings do. Luna: Uber does something similar with driver and passenger ratings, right? Lucas: Absolutely. Uber launched its driver rating system in 2012. Passengers rate drivers one to five stars after every ride. But Uber also lets drivers rate passengers. If a passenger drops below a certain threshold — I think it's around 4.6 — they can lose access to the platform. Luna: Wait, a passenger can be banned? For what? Lucas: Yes. For things like being rude, leaving trash, asking the driver to speed, or just consistently rating drivers poorly. Uber's data shows that drivers who accept rides from high-rated passengers have higher satisfaction and lower churn. It's a two-way trust filter. Luna: That makes sense. If drivers only pick up low-rated passengers, they might have a bad experience and quit. So the rating system protects the driver pool. Lucas: Exactly. And there's a fascinating stat: a study from 2018 found that a half-star increase in a driver's rating led to a 20 percent increase in ride requests. That's huge. The rating directly impacts their income. So drivers have a strong incentive to be good. Luna: But what about rating inflation? Everyone gives five stars because they feel bad giving less. Lucas: That's a real problem. On Uber, the average driver rating is like 4.8 or 4.9. So the system loses its signal — everyone looks the same. Uber has tried to combat this by showing percentile ranks or using a 'star threshold' model where only ratings below a certain level count against you. Luna: And there's also the retaliation problem. If a buyer rates a seller poorly, the seller can rate the buyer poorly in return. That discourages honest feedback. Lucas: That's a classic issue in two-sided ratings. eBay actually used to show the feedback you left for others on your profile, so there was social pressure. But they also had a 'mutual feedback withdrawal' policy — if both parties agreed, they could remove negative feedback. That helped reduce retaliation. Luna: But doesn't that let bad actors off the hook? Lucas: It can. That's the trade-off. You have to balance trust with friction. If the system is too punitive, people stop using it. If it's too forgiving, the signal degrades. The best marketplaces iterate constantly. Luna: So what about Upwork? That's a marketplace for freelance services — the trust dynamics are even more complex. Lucas: Upwork uses a dual rating system: both the client and the freelancer rate each other after a project. But they also have a 'job success score' that factors in things like on-time delivery and repeat hires. It's a composite metric, not just a star average. Luna: And they've had to deal with rating fraud — fake reviews, collusion, that sort of thing. Lucas: Right. Upwork uses machine learning to detect patterns that suggest manipulation. For example, if a freelancer gets a perfect five-star rating from every client but only has one project, that's suspicious. They also require a minimum number of ratings before the score becomes visible. Luna: So there's a threshold for trust. You need enough data points before the system trusts the signal. Lucas: Exactly. And that's a key insight: counterparty ratings are only useful when there's volume. A new seller with 10 ratings is hard to evaluate. A seller with 500 ratings gives you a much clearer picture. Marketplaces often display the number of ratings alongside the average to help with that. Luna: I'm thinking about the design choice: should ratings be public or private? eBay's are public. Uber's driver rating is visible to the driver, but the passenger doesn't see their own rating unless they check. That's a different philosophy. Lucas: It's a trade-off between transparency and gaming. Public ratings create social accountability — you're less likely to be a jerk if everyone can see your score. But they also create incentives to game the system, like trading positive reviews. Luna: And there's the question of anonymity. On Airbnb, hosts and guests rate each other, but the ratings are only revealed after both have submitted. That prevents retaliation. Lucas: That's a smart design pattern — it's called 'double-blind' or 'simultaneous disclosure.' Airbnb introduced that around 2014. Before that, if a host saw a bad review from a guest, they could retaliate with a bad review of their own. The double-blind system solved that. Luna: So when you think about it, counterparty ratings aren't just a nice to have. They're a core liquidity tool. Without them, the market can collapse due to adverse selection. Lucas: Exactly. If only the bad buyers and bad sellers are left, the good ones leave. That's the 'market for lemons' problem. Counterparty ratings act as a filter that keeps the pool healthy. Luna: And it strikes me — this is all about trust, but trust is fragile. One bad experience can undo a hundred good ones. Lucas: That's why the best platforms invest heavily in their rating systems. It's not just code — it's psychology, game theory, and data science. And if you run a business that depends on two-sided trust, it's worth thinking about how you rate both sides. Lucas: Speaking of which — if these conversations are useful for what you're building or running, we keep this show ad-free thanks to listener support. If you find value in episodes like this, you can help us keep going at buy me a coffee dot com slash fexingo. No pressure, but it makes a difference. Luna: Yeah, it's a small way to keep the conversations honest and independent. We really appreciate everyone who chips in. Lucas: Alright, back to ratings. Let's talk about one more platform: TaskRabbit. TaskRabbit connects people who need tasks done with 'Taskers' who do them. They have a dual rating system, but with a twist — the ratings are tied to specific tasks, not just general reputation. Luna: So you can see that a particular Tasker has done 50 furniture assembly jobs with 4.9 stars, but only 2 cleaning jobs. That gives you more context. Lucas: Exactly. It's a move toward skill-specific ratings. And that's a trend we're seeing more of — moving from a single global score to multiple scores for different dimensions. Luna: Like what? Communication? Timeliness? Quality? Lucas: All of the above. On Fiverr, you can rate delivery time, communication, and service as described. On Airbnb, you rate cleanliness, accuracy, communication, location, check-in, and value. The more granular the feedback, the more useful it is. Luna: But there's a cost: more fields mean more friction. If you ask users to fill out six ratings after every transaction, they might just skip it. Lucas: Right. So platforms have to balance detail with completion rate. Some use a single overall rating and then allow additional detail optionally. Others use a default 'quick rating' and then prompt for more later. Luna: It's fascinating how much design thinking goes into something as simple as a star rating. Lucas: It really is. And the takeaway for anyone building a marketplace: don't just rate the product. Rate the people. Your marketplace's long-term health depends on it.