Over the past few weeks, a growing narrative has emerged arguing that AI could destabilize the economy and disrupt entire business models. Last week, I published my view that AI is far more likely to be a productivity revolution than an economic collapse.

But what does this mean specifically for marketplaces?

Many founders are asking:

• Will LLMs capture discovery?
• Will AI compress take rates?
• Will traffic shift away from platforms?
• How defensible are marketplaces in an AI-native world?

In this episode, I break down:

• Why most fears around AI disintermediating marketplaces are overstated.
• Where AI does threaten marketplace margins.
• The structural advantages marketplaces retain.
• The immediate opportunities AI creates for liquidity, cross-border trade, and profitability.
• What founders should do now.

If you’re building, investing in, or operating a marketplace, this episode is for you!

For your reference I am including the slides I used during the episode.

If you prefer, you can listen to the episode in the embedded podcast player.

In addition to the above YouTube video and embedded podcast player, you can also listen to the podcast on iTunes and Spotify.

Transcript

 Hi everyone. I hope you’re having a wonderful week. Basically the last few weeks there’s been a lot of brouhaha and concerns that like AI is going to take over the world. There’s going to be 90% unemployment, grade, the great depression, whatever, and I so fundamentally disagree with that thesis and perspective.

That last week, I spent the time to write a blog post on. The impact of AI on and the fact that it’s actually more likely to lead to productivity revolution than to collapse. Now, the corollary question that people in tech and in marketplaces have been asking is, what is the impact of AI on marketplaces?

And so what I’ve been focusing on rethinking about is, okay, in a world where everyone’s worried and focused in the LLMs, and actually people are worried that they will replace the top funnel, et cetera, what is the actual impact? And I’ve realized that. My thesis perspective and also what I’m seeing on the ground every day is so profoundly different from what the worst case scenarios that people have in their mind is.

And so wanted to share the impact of AI in marketplaces. So without any further ado, let’s get started.

Welcome to Episode 52, Marketplaces and the Age of  AI.

So let me start by showing my presentation so you get a sense of yeah what’s been going on? And okay. So let’s start with a sense of where the market is at. And clearly we’re in the middle of an AI bubble where everything has been ai, all AI all the time. So if you take a step back and you look at where the market is venture dollars of recovering.

And of increase from the lows of let’s say, 22. Not, but mostly in the US and pretty much all of it in  AI. So see, everything has been going up, like ramp sizes, valuations, et cetera, but driven by  AI. So if you look at the first nine month of the year of last year, 75% of the dollars invested went into AI startups.

So it’s been pretty crazy. And at a global level, it’s been about 50% of the funding has been in  AI and that massive increase. And if you look at YC, I think 95% of the startups at YC last year were AI related companies. Now what’s interesting is most of the capital has been going to very few companies, so the anthropic of the world and the Open AI, but also Cursor, Lovable, et cetera.

So the larger models over half the capital has been going to round over 500 million. So it’s been all AI all the time with the largest companies capturing the most value or the largest amount of funding. Obviously there was just a huge ran for OpenAI. There was a round in the works right now for philanthropic.

So it continues to be highly concentrated. The. Yeah, mostly the fin foundational models.

Now things we’re seeing from a trend perspective in terms of what are people investing in outside of the foundational models, so lovable and cursive for X. So vibe coding on a vertical level or quasi no code coding for sites on a vertical level is coming to is coming to the fore.

Using AI to improve the productivity of existing industries is becoming bigger and bigger. So you imagine like AI companies to help the workflows and construction where the GCs and the subcontractors can be on the same page and see exactly what, who’s doing what and simplifying all work processes.

Agents managing other agents and then like general things in compliance, rust trust, et cetera. And the biggest actually trend in the last few weeks, month has been OpenClaw. OpenClaw, this open source local, it could be on a local computer, it could be on a virtual, private server.

Agent that basically runs as your personal assistant is super capable and can do a whole bunch of other things. Still reasonably hard to set up and takes a fair amount of training. And also there’s pretty fundamental security concerns, but the founder of OpenClaw just hired by OpenAI.

And I’m pretty sure that all of the core foundational models are going to have an OpenClaw type equivalent, where you have your super smart Jarvis equivalent assistant that’s going to be at your disposal in the coming weeks, month, et cetera.

Exits are recovering in the market and obviously we’re expecting more m and a and more IPOs especially with SpaceX coming possibly OpenAI and others.

So marketing conditions are improving in ventures and exits as a whole, but frankly only in one sub-sector. Which is AI which actually has been not great for other companies, including marketplaces because people were seeing in AI companies go from zero to a hundred million revenues to a billion of revenues in record time and your marketplace startups, that’s going from a couple million to 10 million to 30 million doesn’t sound so exciting anymore. And beyond that, people are funnily worried that AI is going to disrupt the marketplace fundamentally. So it’s been hard to raise in marketplaces. As much as we’ve been contrarian and we’ve been very selective, so we’ve been investing in applied AI and I’ll talk about what that means for us.

I actually want the AI bubble to continue because I worry that should it implode the baby will be thrown out by the bath water. And even the people that have remained disciplined, the companies that have great union economics growing nicely, that are already having a hard time raising will find it even harder to raise in the future.

So I said lots of IPOs in that are coming to the core so market conditions look reasonably positive. Now, this is a bubble. Very unclear and hard to tell when the bubble will end. And, so we shall see I hope it lasts for many years to come. If only because it’s laying the foundations of a productivity revolution where I can imagine that things will keep getting cheaper, better, faster, as they have over the last, two centuries allowing us to have an even higher quality of life, fewer hours worked on a go forward basis.

So the same way that the bubble of the late nineties, late, the foundations with like fiber, et cetera, that led to the internet revolution in the 2000s. I’m hoping this one lasts long enough that we have access to, in in a way subsidized AI knowledge.

Because right now, they’re they most of these companies have like negative gross margin such that we can build amazing companies on a go forward level.

Yeah, secondary markets been have also started taking off, in fact, it’s leading a lot of interesting businesses because there’ve been very few exits outside of AI.

One of the interesting trends in in, in financing and in venture is people who, knowing they are buying secondaries in different companies, and especially the top, the short tail companies, the sweets and anthropic, but you have an entire asset class being created where you have LP saying, or leader investors saying, “Hey! We have LPs in these venture funds that have been in it for 10 years, 12 years, that haven’t had many exits yet. They want liquidity and so they’re willing to sell at a 20, 30, 40% discount in nav.” And so a lot of investors were not coming in to buying late stage LP stakes in venture, which actually is an interesting asset class because I suspect that you can get a pretty good discount.

And at the same time, liquidity is about to happen given the M&A markets are opening up, given the IPO markets are opening up. So interesting asset class.

So now let’s talk about the impact of AI in marketplaces. So the first big fear that people have is AI will capture the top funnel.

Everyone’s going to go to ChatGPT, or Gemini or Claw, and they’re going to say, I want to buy this, and the transaction will fully happen there. And you will no longer visit eBay, Amazon, DoorDash, Uber booking, et cetera. And I suspect that is, first of all, wrong. The top of funnel will not move to the LLMs.

And let me explain to you why, when you think through actual user behavior, like why are people going and visiting these different sites and what is the what is their thinking pattern? And so when people are going to these marketplaces, there’s typically three approaches for how and when and why they go there.

So if you go to a site like Vinted people don’t go there knowing what they want to buy. It’s more like shopping as entertainment. It’s like I’m going to walk on Broadway in SoHo and I’m going to go and enter their shops not having a real clear sense of what I’m looking for, and if something resonates, I’ll go buy it.

And so you see the engagement in these sites is like 20 pages per visit. People say 10, 20 minutes, 30 minutes on the site every time they visit, and they come multiple times a month. And because the LLMs are all about efficiency and giving you the one thing you want. This is not. There’s no risk at all of this being disrupted.

Like in the top thousand priorities at OpenAI. There’s not let’s analyze the shopping pattern of individual X and create a browsing feed of things that could entertain him to just look at with very low purchase rate. That’s not even in the consideration setting. Sites like Vinted, I see have zero risk of being disrupted because people are not there to be efficient.

They were there to browse and see what’s available. So as long as you have a very long tail lots of different items that, that you can look at and that people find compelling, I don’t see the top funnel changing in any way, shape, or form.

The other second big pattern were where people buy on or search or look for transactions in marketplace is search.

So if you know exactly what you’re looking for, a lot of people just go to Amazon and type what they’re looking for and they type LG C3 65 EVO TV. Poof! One result, they buy it. And they don’t even go to the search engines typically, even they go directly to an Amazon or an eBay.

Now, even if you didn’t start there and you started on an LLIM, or you started on Google, because these marketplaces of high market share, the results you would get actually would still come from the underlying marketplaces. So if you go to Google today and you type in the name of a specific product, almost all the results come from eBay and Amazon because.

Combined, they have a 43% market share of e-commerce. And so even if you went an LLM and you said, I want to buy LG C3 65 inch Evo new or used most of the results would most likely come from eBay and Amazon. So regardless, maybe there’s a little bit of the value capture at the top, but should be no different from the value capture than a Google has.

When people were buying their brand. And so because at the end of the day, OpenAI is not going to be doing customer care, fulfillment, shipping, payments, returns, financing, et cetera. So first of all, I don’t think most of the traffic, if you know exactly what you’re looking for, there’s really zero reason you should be going in LLM. You could go directly to Amazon, eBay, or a vertical site type where you’re looking for it. Poof, get it. There’s no reason go Googled either for that matter. That said this shows that LLMs in general, or a existential threat to Google because instead of getting a lot of results, you give one result on is better.

So if I was Google, I’d be worried. Funnily worried about the impact of LLMs, which is why they’re posting in Gemini. But if I’m whatever, eBay, I wouldn’t be that worried because the at the end of the day the problem that they’re trying to fix and the value they bring is pretty friend, fundamentally different.

Now the third behavior pattern has a little bit more risk. So the other big behavior pattern that you have on for search is something called a consider purchase. You are looking, you want to buy something, but you’re not sure what exactly. So there have been a number of sites that have humid advisors, so there’s four for travel, for, there’s a curated, if you want to buy, like it used to be high-end ski equipment, but it’s broadened.

Or and you have Stitch Fix where you have an a, a fashion advisor telling you what you want. But beyond that, you can imagine there are considerate purchases if you want to buy a car, if you want to buy a house. And there you can make the argument that the LLMs that know you very well actually will play a pretty profound role in advising you as to what is the best neighborhood for you to live in, the best car relative to your needs, et cetera.

Now, that’s part of the reason that some of these sites like Curated, I think they sold for 300 million, but they’d raised 200 million, so it wasn’t a very good asset. But even then, it’s not completely guaranteed that it moves to the LLMs. You can make the argument that the AI implementation that you have within your site, like the Instacart recipe recommendation or Rufus for Amazon.

Because they’re specialized in this category. And you can see in in AI recommendation engine being built by Zillow or being, and Trulia and building, being built by Carvana. That might be as good, if not better than the ones in the LM. So much more risk of disruption here. But then again, consider purchases or a small percentage of the overall shopping pattern that people have in marketplaces.

First concern, will the entire top of funnel move to the LLMs? I think the answer is no. I think maybe if a little bit of it moves up there, but even if it did, I don’t think it would capture a lot of value. Number one, yeah, I don’t think the top funnel moves to the LLMs.

Number two, if it moves to the LLMs. So let’s assume the worst case scenario. What is the. How impacted would the marketplace be? There, I think there’s a lot of nuance depending on who the marketplace is and what it does and how much value it really, truly fundamentally provides. So first of all is how much work is the marketplace doing?

If the marketplace is purely matching a buyer and a seller, so Angie’s List or Zillow or Thumbtack. It’s not doing a lot of work. In fact, you, the user have to do a lot of work because you’re looking at the listings, you’re picking the right ones. You’re, or if you’re putting a if you’re, You go to Thumbtack and you’re asking for people to bid on a job, then you have 20 bids. You have to pick one. You are doing a lot of work.

So in that case, where the level of work of work or management done in the marketplace is low. There’s much more risk in a way of disruption. That’s why the commissions in general of these marketplaces a bit lower. But if you’re doing inventory management last mile, picking in, picking and packing, last mile delivery.

Financing, payments, returns, et cetera. Much, much less risk of disruption. So even at the top of funnel move there, I don’t see companies like DoorDash, Uber, or Amazon being at any risk whatsoever given the amount of work that they do. So ma, the amount of management done matters. And by the way, the trend in marketplaces for the last 25 years have been the more modern, newer marketplaces do more and more. And in fact, you can use AI to do more and to do things that was not possible before. So the more managed the marketplace, the less risk of value captured by the top of funnel LLM should the traffic move there. And as I said, I don’t expect much of it will move there.

So number two is how much work are you doing from a supply perspective? Now of course if you’re Expedia and you’re doing travel and most of the airline, there are five airlines that have most of the volume and they’re already not paying you high commissions, it’s very easy to replicate. So I could see how you could go to ChatGPT and say, “Hey, book me a fly from New York to Salt Lake City.”

And it can do that pretty effectively because there are only five airlines that has to look at to a lesser extent, but still possible. Something like booking.com for hotels. There’s many long tail hotels, which in fact has been the strength of booking. But the big chains where people are pretty loyal to Hilton, Hyatt, have a fair amount of market share.

And as a result, if to the extent you’re booking in a places where you have loyalty points, À la Hyatt. Or Hilton you’re, they can actually reasonably replicate that. So you can actually go to your LLM and say, Hey, book me a flight to Salt Lake City and then book me at the Hyatt in Salt Lake City.

And it can do that reasonably effectively, or at least it will be able to do that reasonably effectively so it doesn’t feel, so in that case the amount of supply is not all that unique. Now, if you take another extreme example, the other end of that, like an Airbnb or a DoorDash, where there’s.

Thousands, probably hundreds of thousands of small mom and pop restaurants and like people’s lists, individual listings, where the supply is very unique, very disaggregated, very long tail. This is not a job that the LMS want to be doing in any way, shape, or form. So much more productive, protected same thing for, for Amazon.

Amazon is really marketplace, by the way. Most of the. There’s thousands and thousands of suppliers on Amazon or in Etsy. And in Uber right now the supply integration, the drivers are, there’s a fair amount of drivers. But again, in a self-driving world, perhaps that changes.

So if you think of like where you’re going to be protected, the more work you do. And the more individual, unique disaggregated, fragmented supply you have, the more protected you are. And so that’s why I’m not at all worried, frankly for the DoorDash’s and Airbnb’s, et cetera, or Amazons of the world, and more so for let’s say the Expedia’s and TripAdvisor’s of the world.

Less. Next thing to think about is like how much work do you do relative to the consumer and if to the extent the transaction is a one-off, you’re buying a car only every five years, you’re buying a house only seven every seven, eight years. Then having a considered purchase going to the LLMs, it’s interacting with them probably makes a lot of sense.

But to the extent you’re doing, you’re using something like Uber every day, the LMS do not want to be dealing with customer care and the fact that the user forgot their phone in the car or they were dropped off at the wrong place, or DoorDash, the wrong food was delivered and people were ordering these things multiple times a week often, and definitely multiple times a month.

And so the more often the more low value, average order value is low, the more there’s a high frequency. So the lower the price and the more conceived transaction, the less the LLMs are going to want to deal with that. And so that’s why, again I think the Uber, is DoorDash, Uber East Amazons of the world are very protected.

Because it’s both high frequency and reasonably low price. Versus a Zillow, let’s say or frankly even flying a plane, which people, most people don’t do on a pretty regular basis. And first argument as I as I said was, I don’t think the top of funnel moves to the LLMs, but even if it did, there’s a set of marketplaces and companies that are reasonably protected because they have unique disaggregated supply. They have they. They do a lot of work, so they’re a more managed marketplace. They have high frequency, low average order value, which creates an incentive for people not to do it.

So most of these companies I don’t see at risk of disintermediation or not dis remediation, margin compression. So what should marketplaces do? And this is pretty different from the tactics of what I think this should implement from an ad perspective, which I will be covering soon, it’s more okay. So if the LLMs or not could be doing things like picking and packing and last mile and aggregating long-term, long tail supply and doing financing and financing guarantees, et cetera, do that. Right?

Make sure that your supply is unique and different and differentiated, which frankly you want to do anyway when you’re building a marketplace, right? Like you don’t want concentrated non-differentiated supply. You should build your own AI. So I gave the example of Rufus for Amazon. So if I was Carvana, I’d build my own AI recommendation engine.

And by the way, in the long term, if you think of a UX UI perspective, today, you have one search box where people type in what they’re looking for, and a separate search box for the L long form questions that are LLME, I’m not sure that makes sense. I would probably have one search box. And if it’s a long question you answer, with the AI type answer.

And if it’s a short question like, LGC 5 65, C3 65 inch poof, you give the search result, leverage, obviously your strength, which usually you have a high market share. And so think through. I would probably index myself in the LMS to get the free traffic and I’ll talk about how much free traffic that is, but I would not let them use you for training data.

So there’s a nuance. Index, but don’t be used for training data. Control the customer experience, which is two things. A, have amazing high NPS, but B. Don’t over monetize, don’t raise prices, right? Be fair in terms of how much you’re monetizing. And, think through that pro perhaps the customer acquisition costs might go up or actually will change and will move from SEM, perhaps SEO in, in, into things like LLMs.

So lots of things you can do in order to protect yourself. Now many people that are worried have said, oh, I’m not going index myself in LLMs and eBay recently made the decision to not index themselves in the LMS while which is a big classified side of France. So leading classified in France did the opposite and fully integrated.

And the argument I would make is you should index yourself. It’s no different from indexing yourself in Google. If you’re actually indexing yourself in Google, there’s no reason you should not index yourself in the in, in the LLMs. Now, if you told me you have 99% market share in your category, you’re the dominant player, and you don’t want people to start the search in this, their journey anywhere else in your site because you control that experience.

Of course, don’t index yourself in Google or the LLMs, but on a global level. The percentage of startups that commend so much market share and control that they that they can afford not to be indexed and have free traffic from Google or LLMs, I think is very low. So for 99% of marketplaces out there, the recommendation is index yourself in the LLMs. Go for it.

Probably warrant fixing or protecting or remission. So a lot of people have been saying, oh, search traffic in is going down. So first of all, that’s not true. It’s basically flat. So the SEO traffic you’re getting is flat. And so you continue working in SEO, don’t ignore it. But don’t ignore the LLMs.

Right now, the traffic from AI is about 34, so a third of the size. So it’s huge and growing very quickly. So if you are not indexing yourself, you’re keeping all of this. Incremental traffic away from your site. So I think eBay made a mistake, in other words. And by the way, normally, is it a lot bigger?

It’s mostly mobile. And good news or not, or not necessarily good news for them. Probably most incumbents don’t move very quickly, have not been super smart. But Google actually has been very strong and they’ve realized that search sorry. LLMs and AI is an existential threat for them.

And so they’ve actually started including the snippets and like showing AI results first, and they’ve been willing to decrease the number of how high the links the sponsored links show up. So Google is definitely evolving to be AI first, so you should definitely, as I said, index yourself there.

But to this day, despite all the noise around, oh, Claude is better, et cetera. The Claude thing is really more of a B2B which is funnily important by the way play than a consumer play right now on the consumer side. ChatGPT still has an 86% market share, so yes, it was a hundred percent, that’s 86%, and Gemini and Claude are gaining share, but from a reasonably low base, I have a hard time seeing that changing unless something pretty profoundly bad happens at OpenAI. They run outta funding for what, whatever reason, which I don’t see happening. Or somehow there’s a true revolution in one of the LLMs but don’t really see it happening. And part of the reason I find that the market shares are sticky, by the way, is to the extent that you have a hundred percent of your conversations in history with one LLM, they have so much knowledge about you and who you are, what you want. That changing anywhere, even in the model is better would lead to worse outcomes. And so at this point in my case, I have so much history with ChatGPT.

It’s very hard for me to move anywhere else because the quality of the answers, the nuance, et cetera, is pretty profoundly different. That said, there are things you use for different things, right? So Claude is better at coding right now, so you’re coding use a combination of Claude or Cursor. I actually love the Sora video generation with your face on iOS from ChatGPT.

So for video I actually use  ChatGPT. and then of course for images. It’s interesting, I used to be a hundred percent a Midjourney. And more and more I’m using GPT in addition to Midjourney. So we’ll see how that plays out. But like video, and of course I’m not a professional videographer. I used to play with Runway and now I’ve a hundred percent switched to Source.

So also interesting how these things evolve over time as the capabilities change. But from the core answering questions perspective hard to imagine people moving to the extent they’ve used it a lot. Now, that said. Most people haven’t used it a lot. I. In my post last week on the usage of AI, we’re still in the very early innings.

I think 80% of the world population is not used AI in any way, shape or form. And most of the rest are free users on one of the different LLMs, which is reasonably low quality usage output, et cetera. So the usage level is a lot lower than people think it is. The thing is, we in tech, or people in finance are the early adopters and super power users.

But that’s not true. So Max says, Hey, you can explore the memory as well and do full data report from OpenAI, which you can. But that requires certain level of technical sophistication that normies don’t have, right? Like the same thing like setting up your OpenClaw where you have to go and in its soul file and define all of oh the personalities and approaches and the way you wanted it to behave and then connect to the backend, like again.

If you’re a normie, there’s, this is not happening. No Normie should set up OpenClaw. I think but it’ll come. And though I suspect that our friends at Open AI and others might not let you export the full memory in the future because that’s their lock-in. But TBD we’ll see how it plays out.

If they retain this level of market share, I think they won’t do it. If they don’t, they probably will. Pie is getting bigger to the point I was making earlier. Search has not really declined yet. It may decline but the most of the most I’ve seen is a 3% decline in some categories of sites.

So continue doing SEO. Continue doing SEM. And do index yourself in the LLMs. I Max. Don’t worry. The full transcript of this will be live next week at my blog, including the PowerPoint, by the way, or the presentation that I’m going through. And the next phase, of course, the more important thing if you’re a, if you’re a marketplace founder, is actually what’s coming next.

And what’s coming next is, okay. So if you’re not worried about disruption in in the marketplaces. By the LLMs because they’re not capturing the top funnel. And even that they do capture the top funnel. They don’t capture that much of value because ultimately you have unique supply. You’re doing a lot of work they’re not willing to do, et cetera.

What else is it that you should be doing today that will completely inflect your business? And so there’s six things funnily, that I think you as a marketplace founder should be doing today that will completely transform your business for the better.

So one is cross-border B commerce. Two is simplified listings. Three is enhanced listing quality. Four is just improve the productivity, improve in your company and improve customer care programming, et cetera. Five, improve revenues and six, like maybe have traceability across the circular economy. So let me explain what I mean by all six. So number one, cross-border commerce.

If you were in Europe back in the day, like when, in the days I was running OLX we had a Polish site and we had a Romanian site and we had a Ukrainian site. In fact, these sites were still the leading players in their countries. But Europe was not Europe. Europe was a coalition of different countries and they were all independents.

So you had a French side and a German side and a UK site. But these days with AI you could do a few really cool things. You can. Automatically translate the listings such that you might be in France, where the listing could be coming from Lithuania or Poland or Romania. And you can translate the conversations between the users.

So you can have buyers and sellers from different countries who are talking in their native language in a completely seamless way. So AI allows you for the first time, and again it implies that you have integrated shipping and integrated payments. So not all companies have it but companies like  Wallapop or Vintage or Ovoko in the car parts place do.

If you look at Vinted is about 10 billion in 10 billion GMV and I think like a billion in net revenues and growing like crazy and very profitable. The. Their strength has been using liquidity from a country of their dominant, like France to then go into countries where they’re entering and having supply and products to sell from the get go.

Now, it only works because, as I said they’ve integrated very effectively. Payment and shipping, but it helps even if you don’t aspire. So obviously Vinted aspires to be a ginormous cross border, cross category, $50 billion company, perhaps even a hundred billion dollars.

But even if you’re the only, the dominant player in your core country, like Wallapop is in Spain, Portugal, or Subito in Italy. By. But they have supply from Spain that’s unique and interesting to people in those countries. So they actually launched in Italy, in Spain, in in Portugal, in France. And that is a source of incremental revenues and incremental.

Ovoko has done the same in car parts. So they’re sourcing in like Poland and Lithuanian. They’re selling into France and, but cross border and even in B2B is happening where investors in a company called CarOnSale, which is a big B2B marketplace for cars, used cars between dealers and already 30% of the volume is cross border.

Second big trends and things people should be doing is simplifying the listing, right? So the old way of listing on, let’s say in eBay is I take 20 photos of my phone. I write a title, I write a description. I select a category I select a price. It’s a lot of work and you may not know actually exactly with what’s the best category.

You may not know the best way to sell to describe it. You may not know what the correct pricing is for that item, but these days, especially in certain verticals, you can just take a photo and boom, automatically. The listing is created for you. A few examples were investors in a company called Rebag and Rebag which is a luxury used handbags marketplace.

They have this AI called Clear. You take a photo and it tells you, make model price, you can get CollX is a tool that scans all of your trading cards, tells you which one’s the value, and allows you to list them instantaneously. One second. “but Vinted comes to the US asking folks if they want to open a “parcel” shop and most American’s “say what?”. So there are cultural difference. Don’t you need a native holding company?”

Yeah. So Vinted coming to the US, Connie may or may not work. The difference between the US and Europe is. First of all, in Europe you can ship from like France to whatever, Lithuania for two euros. Their integrated shipping in the US shipping costs are actually very high. The average distance an item is shipped on eBay, I think is like 2000 miles, and the average shipping cost is $7 or $8, while the average price on, on a Vinted is like 30, 40 euros so like whatever, $30-$50. So it doesn’t support a $7 shipping cost.

And so some of the way and also because of tariffs and shipment costs from Europe to the us the liquidity that Vinted has in Europe with so you can’t use French listings to launch in the us It’s too expensive to ship, plus there’s inconvenience and tariffs, et cetera, so that that core advantage doesn’t exist.

And so I think they’re looking at oh, maybe we have local destinations where you can pick up and you can drop off or pick up for cheaper than if you’re using, than if you are shipping through UPS or FedEx. So they’re testing the model. In the meantime, there is a player that eBay just bought.

It’s called Depop, that is doing reasonably well or very well in the US for now. Now we’ll see if eBay is a good steward of Depop on a go forward basis. I would not bet against Vinted. Because they don’t get it right necessarily the first time. Like the first few times or five times or 10 times they went in the UK, they failed.

But eventually they figured it out and they took over the market and they crushed the incumbents. And they have a lot of money. They’re very smart, they’re low cost provider. And it’s run by my former white hand man who is the, my Fixer. And he helped build and turn around like Wallapop with me as well.

His name is Thomas. He’s amazing. I would not, obviously we’re bias A. I love Thomas and B. It’s one of the portfolio winners for F four FJ Labs, and I think Vinted and so could return a huge percent of portfolio. I’m super bullish. Don’t know they’re going to win the US but I wouldn’t, I would not bet against them.

Okay. Simplified listings, as I said. Right now, you can with the new, especially in verticals, you can take photo and poof. You get the listing, you should totally do it because on marketplaces, 99% of visitors are buyers. It’s only a small percentage of their sellers, so it’s usually like 99 of one, more or less.

And if you make it a lot easier to sell because it’s just one photo, all of a sudden you can improve the percentage of visitors that are sellers. And so if you can actually increase the volume of supply, that’s fantastic. How does Vinted make money? Okay, so Vin, the way Vinted makes money is they’re free.

You could be a hundred percent free on Vinted, where they make no money. Now their business model is multifold. You, the buyer have. So the difference is most marketplaces in the past have said, oh, I’ll take a 15% commission, 20% commission from the seller. What they’ve realized is, especially in Europe where there’s a elasticity of supply, so if you take a big commission, the volume of your supply goes down.

They weren’t free. We’re taking nothing from the seller. Instead, we’re going to monetize the people that get value. So if you’re a buyer and you say, oh, I want, I want escrow because I’m not sure I want to be able to return it, and I want to have, I don’t want, I want it to be shipped to me and I want to pay and I want to pay my credit card versus meeting in the street and paying in cash.

You will pay on top of that. And typically they charge 5% plus a flat fee, plus the shipment cost. So effectively they take 9% from the buyer. And most buyers in the countries where they’re well penetrated have chosen to do that. Then in addition to that, sellers can choose to be visible to pay for the first few slots, and so when you blend all that, they can have an effective take rate of around 10%.

As I said, 10 billion GMV and billion net revenues, and yes, many people sell or list on Vinted because they’d rather not throw it away and find a wonderful home. So they sell things for $3, but most people are not selling for $3. As I said, the average order value of ING is around 40.

And it’s both a source of entertainment and a source of income and allowing people to circulate or change the cost by a pretty effective basis. But they’ve built the lowest cost infrastructure, both in the software development side, customer care side, shipping side and payment side, which allows them to make money even at low average order value.

Okay, so number two, use AI to simplify the listings. Number three, you can enhance your listing. So it’s not just you take a photo and then you get the listing, but actually AI can think through. Okay? This is jewelry. You don’t just take a photo like on the table. It’ll figure out which side you’re selling on and change the background.

So sometimes it just creates a white background, but sometimes it’ll put it in nature or put it in somewhat or whatever to increase the conversion rate. And so there are companies like PhotoRoom which both do it for you or can sell to marketplaces such so they can improve the quality of the images to improve the conversion rates.

Again, all these things completely like your business, like just going cross border can increase your business 30% today. Improving listing quality, simplifying the listings can increase, can double or triple the number of listings you have today, right? So this is not like something that’ll happen a year or two or three years.

It completely funnily changes your business and changing your sell, your rate, your visit to purchase rate from. 2% or 3% is ginormous. So all these things are the, top priority. They, what makes the most sense, which of what needs to be done. Number four, now, everyone is doing this but key use AI to improve customer care.

And so you see it like we’ve invested in a company, actually it’s the next slide called Ace Waves. They’re Sierra for marketplaces. They’ve integrated in multiple marketplaces of ours and they’ve lowered customer care costs by 50, 60% in six months. So you can actually simultaneously lower your customer care costs while improving your NPS now at the same time.

You know what’s interesting is like we’re not actually seeing a decrease for demand for programmers. We’re actually seeing just an increase in productivity of programmers. So the existing programmers do more because they’re using Cursor and get up Copilot and so they’re coding a lot faster than before.

So improve your programmer productivity lower your customer care costs, improve your NPS, use the tools to improve your productivity. And again, on the first three things we discussed, we were increasing volume, revenues, et cetera. Here we’re lowering costs while improving NPS.

Fifth thing is another big trend in marketplaces is now in addition to whatever commission you might take on buying a selling, people are selling advertising.

And advertising is like a 95% gross margin product. So if you look at Instacart, the biggest chunk of revenue is actually from brands buying advertising to have the broad show first, and it’s true on Amazon, it it’s no longer the, it’s not the primary revenue generation, but it’s a multi-billion dollar revenue generation where the existing sellers on Amazon or buying sponsored slots to have buyers see their items first and they’re willing to pay per effectively a certain percentage of GMV.

Now they, it could be on a CPM basis. On CPC basis, it doesn’t matter, but like it’s a GMV equivalent. So on Instacart, I think it’s about 5% of their GMV. It’s coming from ads, but it’s the vast majority of profits because that’s the 95% margin product versus.

If you’re doing a commission on transactions, that’s a maybe 50, 60% margin product because you have credit card processing costs, you have returns, et cetera. So we’re investors, a company called Topsort, and we’re investors with all the companies that I’m highlighting here. And Topsort basically has done all the work to actually fi to optimize your revenues by selling ads and it’s actually more complicated a matter of work that you that you think, because if you’re selling a CPC basis. What you’re optimizing for is not neither the CBC nor the clickthrough rate. It’s actually the CBC times the clickthrough rate. And so you have to figure out which ads are going to click well and automatically put the right ads in first and Topsort is doing an amazing job at helping marketplaces add an advertising revenue layer to their business.

And by the way the future of marketplaces is having more and more different types of revenue streams from financing to commission, to perhaps listing fees, to advertising, et cetera to get your effective take rate to be not too high, but high enough and profitable enough that you can scale the business.

And the last thing, again, not guaranteed, but there are interesting use cases where you can actually trace an item. So you own when you buy it, you have proof of ownership, and then you can one click sell it in another marketplace, transfer the proof of ownership with someone else.

And so we’re seeing that start to happen. It means they’re talking about mandating this in Europe. But we’re talking about, we’re starting seeing it happen. Different companies like Tings, which I think is in the Nordics. On a regular basis, which makes the circular economy safer, more trustworthy, because now you know, oh, this is actually the owner of this item. And then, and when you, then you get ownership and it gives you a certain level of trust.

So all that to say, as I said, if I was a marketplace founder today, I would do, instead of worrying, I would index myself in the LLMs. I would keep focusing and building different supply, adding value added services, and doing work that the LLMs are not willing to do.

And then I would immediately do things like cross order core commerce, simplified listings, and as listing quality. Optimize customer care, optimize programming add revenue streams with advertising and build internal recommendation engines and get a certain amount of AO now.

AEO there are lot of vendors are not clearly good. The people I like best for Graphite HQ, I think they’re be reading themselves as we speak. And so to the extent you’re looking for a good AEO talk to Graphite HQ and yeah they can help. But regardless, index yourself. But don’t let the LLMs use you for training data.

That brings an end to marketplaces in the age of AI. But what I want to talk about next, which is somewhat of a non-sequitur, is like what are the types of things we’ve been investing in this crazy world of AI that are not ai and some are not marketplaces that I still think are really interesting.

While the entire oxygen in the air has been taken by ai. So a few interesting examples Palmstreet. Palmstreet is a live commerce live video streaming mostly for selling rare plants, and they’ve grown very rapidly from zero to like over 10 million a month. And there’s a logic.

So live commerce is something that only worked in China for the longest time, and like things like Taobao, 25% of the transactions was Congress and people said or dismissed as, oh, this is a Chinese unique behavior. But there are actually categories where it makes sense, right? If you’re selling a rare plant, which is a high average order value, you can imagine that getting the story of where it’s coming from, how you take care of it, et cetera, makes a lot of sense.

And they’ve catered to an audience of affluent wealthy women in their twenties, thirties and forties, who actually spend a fair amount every six months. They have these pro shops that are sell, doing two streams. Per week, and we’re selling thousands and thousands of dollars per month.

So business is doing very well, completely under the radar, and growing very nicely. But again, it’s not growing from, 10 million to a hundred million to a billion, a la Cursor, Lovable, et cetera. But these are the things that are like, little by little building, interesting ways of attacking the incumbents.

À la Etsy with different approaches. Another big company or another company interesting company we’ve invested. So continuing the live streaming, live commerce platform or category, Whatnot, is the dominant player here. They’re the number one, collectibles live streaming company with billions of GMV.

And we invested in Troffee, which is Whatnot for the Middle East. Again, early days, but interesting.

It, and Connie I saw your comment. Let me switch back so people can see the question. Yes, Whatnot live is exactly in this category. Fanatics is also launching the category they bought with fanatics live, much smaller.

They bought a company to, to do that. So Whatnot is definitely the dominant player in the category. By now we’re starting to see in both in verticals and other geographies.

Next interesting company. We’ve invested in this company called Garage. So Garage is a fire truck and fire equipment marketplace.

And it used to be that, so in the US the fire departments are funded locally and you have neighborhoods that are rich with lots of donations where they’re buying the latest equipment and then you have reasonably poor neighborhoods where they have like really bad equipment. And so it used to be that people would only sell on like Facebook marketplace.

Their firetruck and firetruck are like cost like 30 K on average. And what this founder is amazing figured out is okay. In order to unlock the marketplace, I actually need to deliver I need to find, to integrate shipping, delivery guarantee, et cetera. So they get the firetrucks on a flatbed which is obviously a special one, and they deliver it to the buyers.

And so they’ve created a B2B marketplaces. For firetrucks with the average order value of 30 K, and on average, each firetruck is shipped by almost 2000 miles. And so it’s by integrating the service layer that they’ve been able to unlock the marketplace. Other cool things Pickle. Pickle is a peer-to-peer reasonably high-end lend renting and lending marketplace for dresses.

And it used to be you had Rent the Runway, but Rent the Runway is its own inventory. And people have tried this idea many times before and it never worked. And it never worked because there was no reverse commerce logistics infrastructure that was low, inexpensive enough, and people didn’t trust each other.

But like enough behavior exchange in terms of people willing to be driven by other people like Uber or go to people’s houses like Airbnb that now the time has come and reverse logistics works well enough that they’ve created a peer-to-peer rental marketplace that works very well, especially in cities where you wanted to look great, et cetera.

LA, New York, obviously Miami or the places where Pickle works well and is growing very rapidly.

Clutch is Carvana for Canada. So Carvana is interesting because they’ve gone from, darling to, they went from zero to hero, back to zero again, back to hero, back to zero, and now it’s worth like a hundred billion.

We both seeded and A brand and then recap later the Carvana Canada, and they’re crushing it like, or like a billion in sales. They just read round over a billion valuation. And I like Canadian market actually more than I like the American market because there’s no CarMax. There’s no competition.

So they are the Carvana of Canada with like sub 1% market share with no CarMax. And so I think that company is absolutely going to crush it.

Next school company Manual. Manual is like Hims or you, so. Combination of like hair loss, TRT, and erectile dysfunction drugs. They started in the UK and actually the biggest market is Brazil.

So think of Ro or Hims for Brazil in the UK. Again, the founders are amazing. They’re launching across different categories. They’re going to go with GLP one inhibitors in the future. À la Ozempic amazing company. Also doing extremely well. So all these companies, they’re not AI in any way, shape, or form, except they all use ai.

And that’s the point I was trying to make, use AI to be more efficient, to increase your revenues, to go across border, et cetera. Minus the company I helped build we’re tokenizing financial assets and many of them US financial assets. So bringing savings US type savings to the rest of the world, right?

If you’re in Argentina or Venezuela or Africa, you don’t really have access to US financial products. You can’t open it in account on Charles Schwab and h often you have like arbitrary risk, confiscation of your assets. You have high inflation, and so we’re giving you a US dollar base for now, eventually other currency, investment products. So we’re tokenizing investment products to give you yield and savings in investment opportunities around the world. The main use case right now actually is more, not quite Degen, but like shorter tail sophisticated investors who want reasonably high yield at low risk and that they can loop.

But long term the idea is like democratize access to investments and savings on a global level. Other cool company is we invest in Boom Supersonic. So we had seen Boom at YC in the early days. And then there are many reasons it didn’t make sense, right? Like you needed a license, to you, you’re not allowed to go have a supersonic boom over the US, you need legislation it changed, et cetera. So we invested a few maybe when there was an inflection point and it was a dual inflection point. The US was changing the law to allow supersonic flight over the US and at the same time they realized that they could use their special engine to power data centers and they started getting huge contracts for AI data center powering.

And so we invested then, just when the business was inflecting. Base Power, which was crushing it as a neo utility with like in a home battery backup like probably the hottest company in the energy space. We’re invested in Neobank in Mexico, run by former founders of an amazing FinTech in Russia, who obviously left Russia to do this, completely crushing it.

And these, what’s interesting with these types of companies is they end up being a lot bigger than you can imagine. Like they end up winning in so many different finance verticals. So you look at Newbank in Brazil or Revolut in Europe, and these are like 50 plus building dollar companies. And Plata has the chance of doing that for Mexico.

Numerai, which is actually AI it’s like a hedge fund built by the masses where people upload their different models and then they pay up to the people that are uploading the models based on the returns created doing really well as well. And we’re in Somos, which is a low cost acid light infrastructure fiber provider in.

In Columbia, they started in midi, in Medellín and a country where a lot of things are broken and they have the lowest costs of fiber installation in the world, doing extraordinarily well and growing like crazy. So I sell lots of like super fun stuff happening where investor in pair, which is basically helping people figure out like what they should be doing with AI in different types of companies.

And then Fleequid, which is a B2B used bus marketplace in Europe. So, lots of interesting things happening outside of the world of, fundamental foundational models and LLMs and in fact, all of these guys use AI to do things better and more interesting. Gives you a sense of both what Marketization do in the age of AI, but also what are the interesting things that are happening outside of the core LLM models, which everyone is talking about.

I’ll pause here for a second. See if anyone has any last questions and if not I will bring this stream to an end. The next stream, which I’m thinking of doing next week by the way, will be an Ask Me Anything session. So probably next Thursday at noon where we’ll cover. Yeah, pretty much whatever you guys have asked questions as to what’s been going on in the world writ large.

So with this, I’ll bring this stream to an end. Thank you for joining this week, and I’ll see you next week.