Latest / Marketplace Businesses with Fexingo: Two-Sided Networks, Liquidity, and Take Rates / How Marketplaces Use Search Relevance to Optimize Discovery
Transcript
- Lucas: If you've ever typed 'vintage leather jacket' into Etsy and gotten back a page of results that actually make you want to buy something, you've experienced search relevance at work. Luna: And if you've typed the same thing and gotten a bunch of cheap polyester replicas, you've experienced the failure of it. Lucas: Exactly. Search relevance is arguably the single most important algorithm in a two-sided marketplace, especially one like Etsy where the catalog is massive and highly heterogeneous. We're talking over 100 million active listings, many of them unique. Luna: So how does Etsy decide which listings show up first when I search for something specific? Is it just keyword matching? Lucas: It used to be much more keyword-driven. Sellers would stuff their titles and tags with every possible search term, and the algorithm would surface whatever had the highest text match. But that led to a lot of irrelevant results. Around 2019, Etsy overhauled its search algorithm to prioritize what they call 'conversion probability.' Luna: Conversion probability — meaning the likelihood that a given search leads to a sale? Lucas: Right. Instead of just matching keywords, the algorithm now scores each listing based on how likely it is that a buyer will click on it and then purchase it. That score is a blend of signals: listing quality, seller history, pricing competitiveness, shipping speed, and of course, textual relevance. Luna: So a seller with a stellar track record and fast shipping gets a boost even if their title isn't perfectly optimized? Lucas: Exactly. It's a move away from pure SEO gymnastics and toward rewarding what actually drives buyer satisfaction. Etsy published a study saying that after this change, buyer satisfaction scores improved by about 15 percent, and conversion rates went up. Luna: Fifteen percent — that's a huge lift for a platform already doing billions in GMV. Lucas: It is. And it demonstrates why search relevance is such a powerful lever for liquidity. If buyers find what they want quickly, they buy more often. If sellers get fair exposure, they stay on the platform. It's a classic network effect reinforcement. Luna: But there's a tension here. Sellers have historically relied on SEO tactics to drive traffic. If Etsy changes the algorithm, it can feel like the rug is pulled out from under them. Lucas: That's the big trade-off. Sellers invest time in keyword research and tagging, and sudden algorithm changes can tank their visibility. Etsy has tried to mitigate that by giving sellers more data in their analytics — showing them what search terms are driving traffic and how their listing quality score compares. Luna: So transparency is part of the solution. But what about the discovery side — the 'I don't know exactly what I want' searches? Lucas: That's where personalization comes in. Etsy uses collaborative filtering and browsing history to surface items similar to what you've looked at before. If you've been browsing mid-century modern furniture, a search for 'desk lamp' might prioritize that style over others. Luna: Which is great for the buyer, but it also narrows the discovery funnel. You might miss out on something outside your usual pattern. Lucas: That's the classic filter bubble problem. Etsy balances it by occasionally injecting exploratory results — items from new sellers or different categories — to keep the ecosystem healthy. They call it 'serendipity' in their search docs. Luna: Speaking of healthy ecosystems, a quick honest thing — we keep this show ad-free, and it's funded entirely by a small group of listeners who chip in a few bucks a month through buy me a coffee dot com slash fexingo. Lucas: Yeah, and that support is what lets us keep drilling into topics like search algorithms without worrying about sponsors or ad breaks. It's a simple model, and we're grateful it works. Luna: So back to relevance — how do other marketplaces handle this differently? Take Airbnb, for example. Lucas: Airbnb's search is fascinating because the inventory is even more differentiated. Every listing is unique — location, price, amenities, host quality. Their algorithm uses a concept called 'booking probability,' similar to Etsy's conversion probability, but with more weight on location relevance and price relative to comparable listings. Luna: And they also have the 'superhost' badge, which presumably boosts ranking. Lucas: Exactly. Superhosts get a visibility boost because they have a proven track record of great experiences. That's a trust signal baked directly into the ranking algorithm. Luna: What about a marketplace like Uber? Search there is trivial — you type a destination, and it shows available cars. But the relevance is all about proximity and ETA. Lucas: Right, Uber's search is almost pure proximity-based because the use case is time-sensitive. But they still have a relevance layer: they might show a slightly longer ETA if it's an UberX vs a Comfort, to give you options. Luna: So search relevance is really a spectrum from 'give me exactly what I typed' to 'show me what I might like.' Lucas: Exactly. And the best marketplaces optimize for the specific type of decision the buyer is making. On Etsy, it's high-consideration, emotional buying. On Uber, it's low-consideration, utilitarian. The algorithm has to match the psychology. Luna: Let's talk about the technical side for a second. How do marketplaces actually measure search relevance? Lucas: They use metrics like precision at K — how many of the top results are relevant — and mean reciprocal rank, which measures how high the first relevant result appears. But more importantly, they look at downstream metrics like click-through rate, add to cart rate, and ultimately conversion rate. Luna: So it's not just about showing the right thing, it's about showing it in the right order. Lucas: Exactly. And that ordering has huge implications for seller behavior. If a seller knows that shipping speed affects ranking, they'll offer faster shipping. If they know that high-quality photos matter, they'll invest in better photography. The algorithm shapes the entire marketplace. Luna: That's a powerful lever — but also a responsibility. A bad algorithm can drive away both buyers and sellers. Lucas: Absolutely. That's why marketplaces like Etsy and Airbnb run extensive A/B tests before rolling out changes. They want to avoid the kind of backlash that happened when eBay changed its Best Match algorithm in 2018 and many sellers saw their traffic drop dramatically. Luna: What's the next frontier for search relevance? Are we going to see more ai driven personalization? Lucas: Definitely. We're already seeing marketplaces use large language models to interpret search queries more naturally — so a search for 'a gift for my mom who likes gardening' can return relevant results without the buyer having to guess the right keywords. Luna: That would be a game-changer for discovery. But it also raises the bar for sellers to tag their listings properly. Lucas: True. But if the algorithm is good enough, it can infer the intent even from messy tags. That's where machine learning really shines — it can learn patterns that explicit keyword matching misses. Luna: So the takeaway for marketplace builders is: invest in search relevance early, measure it carefully, and remember that it's a two-way mirror reflecting the health of your entire ecosystem. Lucas: Well said. And if you want to see how these principles play out in a real marketplace, just open Etsy and search for something quirky. Pay attention to what shows up first, and ask yourself why.