Latest / Marketplace Businesses with Fexingo: Two-Sided Networks, Liquidity, and Take Rates / How Marketplaces Use Algorithmic Matching to Raise Take Rates
Transcript
- Lucas: If you've ever searched for an Airbnb and booked the first listing that popped up, you've experienced the invisible engine that determines a marketplace's take rate. That engine is algorithmic matching. Luna: And matching isn't just about relevance for the user—it's about controlling what gets sold and at what price. For the platform, it's the lever that determines how much value they can capture. Lucas: Exactly. And a couple of dollars a month is genuinely what keeps these conversations going—if you've gotten something out of them, buy me a coffee dot com slash fexingo. It makes a real difference. Luna: Yeah, honestly, listener support is what lets us keep this ad-free and focused on the mechanics. So thanks if you do. Lucas: Anyway, back to matching. Let's take Airbnb. In 2014, their take rate was around 3 percent—just a small fee for connecting hosts and guests. Today, it's over 14 percent. How did they quadruple it without driving users away? Luna: Part of it was product improvements—like the introduction of Smart Pricing in 2015. But I think the bigger story is how they changed their search ranking algorithm. Lucas: Right. Early Airbnb search was basically chronological—most recent listings first. That created a problem: hosts could game the system by relisting, and guests had to dig through mediocre options. So Airbnb switched to a relevance-based ranking that prioritized listings with higher conversion probability. Luna: And that shift had a direct impact on take rate. By showing guests listings they were more likely to book, Airbnb increased the overall booking rate. More bookings meant more fees. But they also started charging hosts a higher percentage—up to 14 percent for some listings. Lucas: That's the key insight. You can only raise your take rate if you're providing more value per transaction. Better matching means guests find the right place faster, hosts get more bookings, and the platform earns more per match. It's a virtuous cycle. Luna: But it's not just about search ranking. Airbnb also uses machine learning to suggest pricing to hosts. Smart Pricing analyzes hundreds of factors—seasonality, local events, similar listings—to recommend a price that maximizes the probability of booking. Lucas: And that's essentially algorithmic matching of price to demand. Hosts who use Smart Pricing see about 10 percent more bookings on average, according to Airbnb's own data. That lifts the entire marketplace's liquidity, which gives Airbnb more room to adjust take rates. Luna: Let's bring in another example: Upwork. They started as a simple marketplace for freelancers—post a job, get bids, hire. But they realized that low-quality matches were hurting retention on both sides. Lucas: So they moved toward a managed marketplace model. Now, when a client posts a job, Upwork's algorithm suggests a shortlist of freelancers based on past performance, skills, and even communication style. The client doesn't see hundreds of bids—they see three to five curated options. Luna: And that curation improved conversion rates significantly. Upwork reported that clients who used the 'Talent Scout' feature—which is essentially algorithmic matching—were 30 percent more likely to hire someone within a week. Fewer choices, better outcomes. Lucas: Which is counterintuitive. You'd think more choice would be better for a marketplace. But when you have too many options, decision paralysis sets in. Algorithmic matching reduces search friction, and that allows the platform to charge a higher take rate—Upwork's is now around 20 percent for many contracts. Luna: And that take rate is sustainable because the algorithm is constantly learning. Every time a client hires a freelancer, the system gets feedback. Over time, the matching gets better, and the platform becomes stickier. Competitors can't easily replicate that data moat. Lucas: That's the real competitive advantage. A marketplace that invests in algorithmic matching builds a dataset that improves over time. New entrants can copy the feature, but they can't copy the years of transaction data that train the model. It's a classic network effect plus a data effect. Luna: But there's a risk. If the algorithm optimizes too aggressively for short-term conversion, it might push users toward higher-priced options that earn the platform more fees but aren't actually the best match. That can erode trust over time. Lucas: Absolutely. You have to balance take rate with match quality. Airbnb learned this the hard way when some hosts complained that Smart Pricing was pushing their prices too low to compete. So they made it optional. The algorithm is a tool, not a mandate. Luna: Another interesting example is Uber. They use algorithmic matching to pair riders with drivers, but their take rate isn't just a flat percentage—it's dynamic based on surge pricing. When demand is high, the algorithm raises prices, which increases the take rate per trip. Lucas: Right, but Uber's matching algorithm also considers driver proximity, estimated time of arrival, and even driver ratings. They've been experimenting with upfront pricing and destination-based matching to improve the rider experience. All of that feeds into their ability to charge a higher effective take rate—around 25 percent on average. Luna: And unlike Airbnb or Upwork, Uber's matching happens in real-time. They have milliseconds to decide which driver gets which rider. That's a much harder algorithmic problem, but it also creates a massive barrier to entry. No competitor can match Uber's data on traffic patterns and driver behavior. Lucas: Let's talk about a less obvious marketplace: Etsy. They have a different challenge because sellers are often creating unique, one-of-a-kind items. You can't just rank by popularity—you need to surface niche products that appeal to specific tastes. Luna: So Etsy uses a combination of collaborative filtering and content-based filtering. If you've bought a vintage camera, they'll show you similar vintage items, not just the most popular listings. That increases the chance of a purchase and allows them to maintain a take rate of around 6.5 percent. Lucas: Which is lower than Airbnb or Uber, but Etsy's sellers are smaller and more price-sensitive. If they raised fees too much, they'd lose their supply side. So algorithmic matching becomes a way to increase transaction volume without raising take rate per transaction. Luna: That's a crucial point. The optimal take rate depends on the elasticity of both sides. For high-demand platforms like Airbnb, where guests have many options, you can charge more because the value of the match is high. For niche platforms like Etsy, you need to keep fees low and rely on volume. Lucas: So the algorithm is really the mechanism that determines which side gets more value. In a two-sided network, you can use matching to favor either supply or demand. Airbnb's algorithm subtly prioritizes listings that are likely to convert, which benefits the platform. Upwork's algorithm favors freelancers with strong track records, which benefits buyers. Luna: And that's a design choice. If you want to raise take rate, you need to make sure the algorithm is creating enough value on both sides that they don't churn. Otherwise, you're just extracting rent. Lucas: Let's zoom out for a second. The economics of algorithmic matching come down to a few key metrics: conversion rate, average order value, and retention. If your algorithm improves each of these by even a few percentage points, the compound effect on take rate is huge. Luna: And the beauty is that these improvements are self-reinforcing. Better conversion attracts more users, which generates more data, which improves the algorithm further. That's the flywheel effect that marketplace investors love. Lucas: But there's a dark side. Algorithmic matching can also create winner-take-all dynamics where the most popular sellers get even more visibility, starving out newcomers. That can hurt long-term marketplace health by reducing diversity of supply. Luna: Platforms need to intentionally inject randomness or try to promote new sellers. Amazon does this with its 'New Seller' badge and temporary search boosts. Etsy has a 'Shop Local' filter. It's a deliberate design choice to counteract the algorithm's natural tendency toward concentration. Lucas: So the takeaway for anyone building or operating a marketplace: your matching algorithm is your most strategic lever. It directly impacts your take rate, your liquidity, and your defensibility. But it's not a set-it-and-forget-it tool—you have to constantly tune it for fairness and long-term health. Luna: And the best marketplaces treat matching as a product, not just a backend function. They run A/B tests, measure satisfaction on both sides, and iterate. That's what separates a marketplace that grows its take rate sustainably from one that extracts value until it breaks. Lucas: Alright, next time we'll talk about how marketplaces use gamification to increase engagement on the supply side. Thanks for listening.