Latest / Marketplace Businesses with Fexingo: Two-Sided Networks, Liquidity, and Take Rates / How the Best Marketplaces Use Search to Drive Liquidity
Transcript
- Lucas: Search is probably the most underrated lever in a two-sided marketplace. Most founders obsess over liquidity — getting enough buyers and sellers in the door — but once that initial balance exists, the way you surface inventory determines whether the marketplace compounds or stalls. Luna: That makes sense. I think of search as the invisible hand that decides which supply meets which demand. And if it's tuned wrong, you can have plenty of both sides but still get zero transactions. Lucas: Exactly. There's a great example from Etsy. Back in 2020, they overhauled their search algorithm to favor recency and listing quality over pure relevance. On paper, that sounds like a small tweak. But it boosted conversion rates by about 15 percent within a few months. Luna: Fifteen percent just from changing how results are ranked? That's huge. What was the old algorithm doing wrong? Lucas: The old system was basically a keyword match — if you searched 'vintage ceramic mug', it'd show you every listing with those exact words, oldest first. That meant stale inventory from sellers who'd listed once and never updated. The new algorithm introduced what they called 'listing quality score' — it weighted things like recent sales, shipping speed, and whether the photo was high-resolution. Luna: So they prioritized signals that predicted a good buyer experience, not just textual relevance. That's smart, but it must have pissed off some sellers who had been ranking well on keywords alone. Lucas: It did. There was a backlash from a subset of sellers who saw their traffic drop. But Etsy was willing to sacrifice short-term seller satisfaction for long-term buyer trust. And the data supported it — repeat purchase rates went up. Luna: That trade-off is interesting. A marketplace can optimize search for conversion in the moment, or for long-term liquidity. Sometimes those are in tension. Lucas: Right. And the really good marketplaces design search to solve the chicken and egg problem even after the marketplace is mature. Take Zillow. When you search for homes, they don't just show you the most likely match — they intentionally sprinkle in a few 'just listed' or 'price reduced' listings that might not perfectly match your query. Luna: Because those listings drive urgency and engagement. If you see a house that just dropped by 20 grand, you're more likely to click and maybe even tour it, even if it's not exactly what you wanted. Lucas: Exactly. Zillow's search isn't optimized for perfect precision — it's optimized for discovery. That's a critical distinction. In a marketplace, search isn't just a utility; it's a liquidity tool. Luna: And that's very different from, say, Google Search, where the goal is to get you to the right page and then leave. In a marketplace, you want the user to stay, browse, and transact. Lucas: Absolutely. Uber is another case. When you open the app, the search is basically 'where are you going?' But the algorithm doesn't just match you to the nearest driver. It also considers driver earnings, surge pricing, and estimated wait times to balance supply and demand across the city. Luna: So search is actually a marketplace optimization problem in disguise. You're not just finding a ride — you're allocating scarce driver resources. Lucas: Precisely. And that's why the best marketplaces invest heavily in search early, even when they have few listings. Because the algorithm you build when you have 100 listings becomes the foundation when you have a million. If you train it to maximize short-term revenue, it's very hard to retrain it for long-term health. Luna: That reminds me of a common mistake I see: marketplaces that rank listings by the platform's take rate or the seller's advertising spend. So the highest-paying listings appear first, regardless of quality. Lucas: It's tempting because it boosts revenue immediately. But it kills the buyer experience. If buyers don't find what they want, they leave. And then sellers have no reason to advertise because there are no buyers. Luna: Yeah, it's a classic short-term gain, long-term pain. I think of Amazon's early search — they famously resisted paid placement for years, even though it left money on the table, because they wanted search to be purely relevance-based to build trust. Lucas: And now Amazon's search is heavily monetized, but they built that trust first. They earned the right to introduce ads because the baseline relevance was already strong. That's the sequence: relevance first, monetization later. Luna: So what about newer marketplaces? How should a founder think about search when they have, say, 50 listings? Lucas: The honest thing is, with 50 listings, search barely matters — you can manually curate. But you should still design the data model as if you'll have 50,000. Capture attributes like condition, location, price, and seller rating from day one. Because the moment you hit a thousand listings, you need those signals to rank. Luna: It's like infrastructure debt. If you skip it early, you pay interest later with bad search results. Lucas: Exactly. And one more thing: search also communicates to sellers what the marketplace values. If you rank by 'newness' or 'price', sellers optimize for that. If you rank by 'seller rating', they focus on customer service. So search is also a cultural signal. Luna: That's a powerful point. The algorithm is a mirror of the marketplace's priorities. Lucas: Look, we talk a lot about network effects and liquidity on this show, but search is often the mechanism that makes liquidity real. Without good search, you have two sides that exist but never meet. Luna: And that's why we keep coming back to it. It's not glamorous, but it's foundational. Lucas: Yeah, and conversations like this — about the nuts and bolts of building something that works — are exactly why we do this show. A quick honest thing: this podcast stays ad-free, and that's possible because a handful of listeners chip in monthly at buy me a coffee dot com slash fexingo. If these deep dives are useful for something you're building or running, that's where you can keep them coming. Luna: Yeah, it's a small group already, and it genuinely funds the research and production for all these episodes. No pressure, but if it's valuable to you, that's the way to support it. Lucas: Alright, back to search. One marketplace that I think nails this is Airbnb. Their search algorithm doesn't just match location and price — it also personalizes results based on past bookings, wish lists, and even how long you linger on a listing. Luna: Interesting. So they're using behavioral data to infer intent. If you click on a lot of treehouses, they'll surface more unique stays, even if you searched 'apartment in Paris'. Lucas: Exactly. And that increases the chance of a booking because it feels serendipitous. But it also means the algorithm has to be careful not to narrow your view too much. You don't want to create a filter bubble where you only see what you've already seen. Luna: That's a real risk. Over-personalization can reduce liquidity by hiding supply that a buyer might actually like if they saw it. Lucas: Right. So the best search algorithms balance relevance and discovery. They show you what you asked for, but also a few things you didn't know you wanted. That's the magic. Luna: And that balance is hard to get right. It requires constant experimentation. Lucas: It does. But marketplaces that invest in search as a core competency — not just a feature — end up with a durable competitive advantage. Because search data is a moat: the more transactions happen, the better the algorithm gets, and the harder it is for a competitor to replicate. Luna: So search is not just a utility. It's a network effect in its own right. Lucas: Exactly. And that's the takeaway I want listeners to hold: if you're building a marketplace, treat search as a liquidity engine from day one, not a feature you bolt on later. Luna: Great place to leave it. Thanks, Lucas. Lucas: Thanks, Luna. See you next time.