Latest / Marketplace Businesses with Fexingo: Two-Sided Networks, Liquidity, and Take Rates / How Marketplaces Use Reviews to Solve Adverse Selection
Transcript
- Lucas: If these conversations are useful for what you're building or running, we want to mention something before we dive in. We deliberately don't run ads on these shows. If you want to support that choice, the link is buy me a coffee dot com slash fexingo. Low-key, no pressure — but it's what keeps us ad-free. Luna: Yeah, and it really does make a difference. Even a few coffees a month helps us keep the focus on the content, not the sponsors. Lucas: Exactly. So today, let's talk about adverse selection in marketplaces. It's this classic economics problem where bad products or bad actors can drive out good ones — before a transaction even happens. And the best marketplaces have learned to solve it with review systems. Luna: So it's not just about after the fact feedback. It's about preventing the lemons problem from ever taking root. Lucas: Right. Imagine you're listing a room on Airbnb. A potential guest messages you — no reviews, profile just created, no verified ID. Your first instinct might be to decline. That's adverse selection: the good guests get passed over because the bad ones look the same in the dark. Luna: And Airbnb's answer was the 'verified guest' feature, where guests can upload ID before booking. That shifts the signal from 'unknown' to 'trustworthy'. Lucas: Exactly. They also introduced the 'guest guarantee' — hosts get reimbursed if a guest damages property. That lowers the host's risk of accepting a bad guest, which increases supply. Both sides benefit. Luna: But let's talk about the review system itself. Uber uses a double-blind rating — neither driver nor passenger sees the other's rating until after the trip. That prevents retaliation. Lucas: Which is huge. If a driver knows a passenger is about to rate them, they might hold that feedback hostage. Uber's system encourages honest ratings because there's no immediate consequence. Luna: And they also use 'recent rating' rather than lifetime average. That means a few bad trips can drop your score fast, incentivising consistent quality. Lucas: eBay had a different approach early on. They used a simple positive/negative/neutral system with a net score. But the problem was that sellers could leave retaliatory negative feedback for buyers, so buyers were afraid to give honest negative reviews. Luna: Yeah, they eventually changed it so sellers can only leave positive feedback for buyers. That removed the fear factor. Lucas: And it made the feedback system much more useful for buyers. But it also meant sellers lost a signal about problematic buyers. So it's a trade-off. Luna: What about 'verified purchase' tags? Amazon uses those to indicate the reviewer actually bought the item. That's a direct attack on fake reviews. Lucas: It helps, but it's not bulletproof. There are still incentivised reviews — sellers send free products in exchange for honest reviews, but those reviews are often biased upward. Amazon has cracked down, but it's a cat and mouse game. Luna: Some marketplaces use 'review helpfulness' votes — like on Yelp or TripAdvisor — where users can vote on whether a review was helpful. That surfaces the most informative reviews. Lucas: But even that can be gamed. Competitors can downvote positive reviews. So platforms need to use machine learning to detect voting patterns that look coordinated. Luna: Let's talk about the economic angle. How do reviews affect take rates — the fee the marketplace charges? Lucas: Good question. A strong review system builds trust, which increases transaction volume. Higher volume means the marketplace can justify a higher take rate because the value is clearer. But if reviews are unreliable, users will transact off-platform to avoid fees. Luna: So a poor review system can actually force a marketplace to lower its take rate to compete with the black market. Lucas: Exactly. We've seen that with platforms like Craigslist — minimal review system, low trust, so they can't charge much. Meanwhile, a platform like Airbnb can charge 15% because the trust infrastructure is robust. Luna: There's also the problem of review inflation. On Uber, almost everyone has 4.8 stars or higher. That makes the rating useless for differentiation. Lucas: Right. When everyone has a high score, the system loses signal. Uber actually tried to nudge drivers by showing them that a 4.6 was considered 'low' even though it's still high in absolute terms. That's a design challenge. Luna: Some marketplaces use a 'median rating' instead of average to avoid distortion from outliers. Or they show the distribution of ratings — like 'X% gave 5 stars'. Lucas: Amazon does that. They show the bar chart of star ratings. That gives a more nuanced picture than just a single average. And they highlight the most recent reviews, which matters because product quality can change over time. Luna: So the key design principles are: make reviews honest, hard to game, and informative. What's the biggest mistake new marketplace founders make? Lucas: They launch with a review system that's too simple — just a star rating with no verification, no content, no double-blind. That invites gaming early, and once trust is broken, it's very hard to rebuild. Luna: Yeah, and they often don't invest in review moderation or fraud detection until it's too late. Lucas: Another mistake is not collecting enough structured data. A review that says 'great product' is less useful than one that says 'fits true to size, arrived on time, packaging was intact.' Luna: That's why some marketplaces use guided review forms — like asking 'Was this item as described? How was the shipping speed?' — to pull out specific signals. Lucas: And then they can aggregate those signals into a 'quality score' that's separate from the star rating. That gives both sides a richer picture. Luna: One more thing — cross-platform reputation. Some marketplaces allow users to import reviews from other platforms. That can help jumpstart liquidity for new users. Lucas: Right, like Airbnb lets you link LinkedIn or Facebook to establish credibility. But that also introduces privacy concerns. It's a balance. Luna: So what's the future of marketplace reviews? More ai generated summaries? Video reviews? Lucas: I think we'll see more use of natural language processing to extract sentiment and topics from text reviews, then present those as tags. Instead of reading 50 reviews, you see 'commonly mentioned: 'great value', 'fast shipping', 'sizing runs small'. That's already happening on some sites. Luna: And video reviews are becoming more common, but they're harder to moderate at scale. So we might see hybrid systems where text is primary and video is optional. Lucas: The bottom line: a marketplace's review system is not a feature — it's the foundation. If you get it wrong, the whole platform suffers from adverse selection. Get it right, and you build a moat. Luna: Well said. Thanks for the deep dive. Lucas: Thanks for listening. We'll be back next time.