Latest / Marketplace Businesses with Fexingo: Two-Sided Networks, Liquidity, and Take Rates / How Marketplaces Use Dynamic Pricing to Balance Supply and Demand
Transcript
- Lucas: If you use a ride-hailing app in Latin America — say Cabify in Mexico City — you might notice the price to go across town feels consistent, even at 3 PM on a Tuesday. Luna: Not the surge pricing you'd expect from Uber at 2 AM on a Saturday, right? Lucas: Exactly. And that's not an accident. Cabify runs dynamic pricing that's about balancing supply and demand across the entire day — not just reacting to spikes. They use machine learning to set price floors and ceilings that keep drivers online even when demand is low, and riders booking even when demand is high. Luna: So they're managing liquidity, not just maximising revenue per trip. Lucas: Right. And this is a really useful case for understanding how marketplaces can use pricing as a tool to build liquidity, not just capture it. Most people think dynamic pricing is just surge pricing — raise prices when demand exceeds supply. But the real art is setting lower bounds too. If you don't have a price floor, drivers during off-peak hours might just go home. And if you don't have a price ceiling, riders might get angry and switch to public transit. Luna: Quick honest thing — a handful of listeners chip in monthly through buy me a coffee dot com slash fexingo, and that's literally what funds making this many of these episodes. If these conversations are useful for what you're building or running, that's the only reason they keep coming out. Lucas: Yeah, it's a small group, and it keeps the whole thing ad-free and independent. Appreciate every one of them. Anyway — so Cabify's model is particularly interesting because they operate in markets where drivers have lower margins than in the US or Europe. A $2 trip in Mexico City might be the difference between a driver staying online or turning off the app. So the marketplace can't afford to let prices drop below a certain threshold. Luna: What's that threshold based on? Lucas: Cabify uses a cost-plus approach. They estimate the driver's cost per mile — fuel, maintenance, depreciation — and set the price floor at a small margin above that. So even during the slowest hours, a driver knows they'll at least cover costs. That keeps supply on the platform. And then on the demand side, they set a price ceiling at roughly what a rider would pay for a taxi — so the rider never feels gouged. That keeps demand coming back. Luna: So they're essentially capping their own take rate in certain situations? Lucas: Exactly. During off-peak hours, Cabify might only take 10% commission, or even 5%, just to keep the transaction happening. During peak hours, they can take 25% or 30%. But the blended take rate across all hours might be 18% — which is lower than Uber's typical 25% in developed markets. But Cabify's argument is that a lower blended take rate with higher liquidity is more valuable than a high take rate with thin supply. Luna: And that trade-off is the core of marketplace pricing, isn't it? Lucas: It is. And it's not just ride-hailing. Let's take Airbnb. They experimented with dynamic pricing through their Smart Pricing tool. The idea was to help hosts set prices based on demand — higher for events, lower for Tuesdays in February. But they had to be careful not to force it. Hosts who felt their pricing autonomy was taken away would leave. So Airbnb made it opt-in. Luna: So they gave hosts the tool but let them choose. That's a different approach from Cabify, which sets the price algorithmically but transparently. Lucas: Right. And the difference comes down to the nature of supply. In ride-hailing, supply is drivers who are online right now. They have low switching costs — they can turn off the app and go home. So the marketplace needs to guarantee them a minimum income per hour. In short-term rentals, supply is a physical asset that's already committed. The host has already cleaned the apartment and handed over the keys. They can't just 'go home' at 3 PM. So the marketplace has more flexibility to experiment with pricing on the margin. Luna: But both face the same tension: if you raise the take rate too high, you lose either supply or demand. Lucas: Yes, and dynamic pricing is one way to manage that tension without having a single take rate that alienates everyone. The real innovation in marketplace pricing over the last few years has been moving from a flat commission to a variable take rate that adjusts with market conditions. Luna: Can you give a specific example of a marketplace that does variable take rate well? Lucas: Sure — let's look at OpenTable. They charge restaurants a per-diner fee that varies by day and time. A Saturday night reservation might cost the restaurant $2 per diner, but a Tuesday lunch might cost only $0.50. That's dynamic pricing on the take rate itself. Restaurants are happier to pay when they know they'll fill seats that would otherwise be empty. And OpenTable keeps volume high during off-peak times. Luna: So they're using the take rate as a lever to shape supply behaviour — not just capturing value from demand. Lucas: Exactly. And that's the shift in thinking. For years, marketplaces thought of take rate as a fixed percentage they were entitled to for facilitating the transaction. But more sophisticated marketplaces now see take rate as a variable tool they can adjust to optimise the total value exchanged on the platform. Luna: Does dynamic pricing ever backfire? I mean, what about the backlash against Uber surge pricing during emergencies? Lucas: It absolutely can. The most famous case is Uber during the 2014 Sydney hostage crisis. Surge pricing kicked in because demand spiked near the scene, and the public outcry was huge. Uber refunded all fares and apologised. But the lesson is that dynamic pricing can't be purely algorithmic — it has to have judgment. Cabify, for example, caps surge at 2x normal prices, even if demand would justify 4x. They'd rather have longer wait times than brand damage. Luna: So there's a reputational constraint on how high you can push prices. Lucas: Right. And that constraint is asymmetric. You can lower prices without backlash, but raising them — especially in visible, emotional situations — can destroy trust. That's why many marketplaces now use 'surge lite' — small increases rather than big jumps. Or they use non-price rationing, like wait times or waitlists, to avoid the optics of price gouging. Luna: What about marketplaces where supply is more elastic? Like freelance platforms? Lucas: Good question. On Upwork, freelancers set their own rates, and the marketplace takes a sliding fee that decreases as the freelancer earns more with a client. That's a different kind of dynamic pricing — not adjusting prices in real time, but adjusting the take rate based on relationship depth. The logic is that repeat transactions have lower marginal cost for the platform, and they want to encourage long-term relationships. Luna: So the take rate becomes a retention tool. Lucas: Exactly. And that's actually one of the most powerful applications of dynamic pricing in marketplaces — not just adjusting prices to balance supply and demand in the moment, but adjusting the platform's cut over time to encourage the behaviours you want. Lower take rate for repeat transactions, higher for one-off matches. It aligns incentives. Luna: I want to go back to the Cabify example. How did they actually implement this? Did they build their own ML models or use something off the shelf? Lucas: They built their own. According to a talk their VP of Data gave last year, they started with a simple rule-based system: if demand exceeds supply by 20%, raise prices by 10%. But they found that was too reactive. Riders would see a price surge and open another app. So they moved to a predictive model that forecasts demand 30 minutes ahead and adjusts prices proactively. The model uses historical data, weather, events, and even public transit schedules. Luna: Proactive rather than reactive — that's a big shift. Lucas: Huge. Because the goal isn't just to clear the market at the current moment — it's to keep the market liquid over the next hour. If you raise prices reactively, you might get a surge of drivers, but they take 10 minutes to arrive. By then, demand might have dropped. So you end up with too many drivers and then they leave. A reactive system can create oscillations. A predictive system smooths that out. Luna: And that's better for everyone — riders get consistent pricing, drivers get stable income. Lucas: Right. And the marketplace gets higher total transaction volume. Cabify found that after they implemented predictive dynamic pricing, their gross merchandise volume grew by 12% without changing the number of active users. They were just matching supply and demand more efficiently. Luna: Is there a case where dynamic pricing reduces total GMV? Lucas: Yes — if you set the price floor too high. Early on, some marketplaces thought they could guarantee high earnings for suppliers by setting a high floor. But that priced out demand. For example, a luxury goods marketplace that sets a minimum price of $500 might lose all the casual buyers who would have bought a $300 item. The total GMV drops. The floor has to be low enough to keep demand alive. Luna: So it's a balancing act. Too low, and suppliers leave. Too high, and buyers leave. Lucas: Exactly. And the optimal price floor isn't static — it changes with the competitive landscape. If a competitor launches with lower prices, you might need to lower your floor to keep buyers. If your supply becomes scarce, you might raise your floor to attract more suppliers. Dynamic pricing means the floor and ceiling are both moving targets. Luna: What about marketplaces that have physical goods? Like Amazon Marketplace? Lucas: Amazon uses dynamic pricing extensively, but it's more about pricing the products themselves than the take rate. Their algorithm reprices items constantly based on competitor prices, demand, and inventory levels. But their take rate is mostly fixed — 15% for most categories. They don't adjust the commission based on time of day. That's because physical goods have a different cost structure — storage, shipping — that don't vary much hour to hour. Luna: So dynamic pricing on the take rate is more suited to service marketplaces where supply is perishable — a car seat, a hotel room, a driver's hour. Lucas: Exactly. Perishable supply is the key. If the supply can be sold again tomorrow, you don't need to discount it today. But if it's gone forever — like an empty restaurant table at 8 PM — you want to discount aggressively to fill it. That's why OpenTable and hotel booking sites are the most aggressive with dynamic pricing. Luna: And that's also why ticket marketplaces like StubHub use dynamic pricing — a concert seat has zero value after the show starts. Lucas: Right. StubHub actually encourages sellers to lower prices as the event approaches, and they adjust their own take rate to incentivise that. They might take a 10% commission on a last-minute sale versus 20% on a sale two months out. That's dynamic take rate to encourage fast clearance. Luna: So the future of marketplace pricing is probably even more granular — real-time, personalised, context-aware. Lucas: I think so. We're already seeing marketplaces experiment with personalised pricing — showing different prices to different users based on their willingness to pay. That's controversial, but the technology is there. The question is whether users accept it. For now, most marketplaces stick to transparent, rule-based dynamic pricing that everyone can see. That builds trust. Luna: And trust is the foundation of liquidity. Lucas: Exactly. Without trust, no amount of fancy pricing will keep both sides on the platform. So the best dynamic pricing is the one that's fair, predictable, and aligned with what users expect. Cabify's model works because drivers know the floor will cover their costs, and riders know the ceiling won't exceed a taxi. That mutual predictability is what keeps the network liquid. Luna: So if you're building a marketplace today, what's the one pricing principle you'd take away? Lucas: I'd say: design your take rate to be a lever for liquidity, not just a revenue maximiser. Start with a fixed commission, but build the ability to adjust it based on supply-demand conditions. Use price floors and ceilings to keep both sides in the market. And always, always test the impact on trust — because a price that feels unfair will kill your marketplace faster than a low take rate ever will.