Marketplaces Deep Dive with Ramesh Johari
01:12:26 Share

Good afternoon LPs. I will say it feels good to get me out of the interview seat and have another episode with a real and proper guest. You couldn't handle the pressure. Yeah, very much so.

Very much so. So David, who do we have with us today? We are super lucky to be joined by our advisor at Wave, Ramesh Johari, who, in addition to being a very, very awesome and critical advisor to us at Wave, is a professor at Stanford of management science and engineering. And he teaches courses on data science and platform and marketplace design, all of which is super core to what we do. And then additionally, outside

outside of Stanford and Wave. Previously, he was head of data products at Upwork. He took a leave from Stanford to do that. And he's now also a senior advisor to Airbnb, Uber, and Stitch Fix. Really, really critical to the whole Silicon Valley marketplace ecosystem from Stanford to great companies to

Wave. We're really lucky to have him here. So thank you for joining us. Yeah, thanks a lot for having me here. You know, it's exciting to talk about marketplaces. I could talk about them all day. So it's literally your job to talk about literally my job. It's my job, too. So great. So that's what we're going to talk about today. Everything all about marketplaces, how you start them, how you build them, how you operate them, and then how you optimize them at scale. And Ramesh has seen all of it. Let's start with start

So Ramesh, one of the first things you kind of told us here at Wave as we were getting going was about your framework for kind of the three fundamental, I think you call them friction points a marketplace can address, but really like jobs that a marketplace platform does in a market. Can you talk a little bit about them? Yeah. And, you know, this whole thing really started when a while back I was talking to some students about

about like, you know, what it is we're selling. You know, when you start a business, you think you're selling something. So what is it you're selling? And the initial thought was that like, if you look at, you know, say Airbnb, you know, that's a marketplace that's selling lodging. You look at Uber and Lyft, that's like a marketplace that's selling transportation. You look at Amazon, you know, that's a marketplace that's selling goods. And

What I thought about that, something made me really uncomfortable about it, which is, you know, when you go to these marketplaces, it's true that you're buying lodging, you're buying transportation, you're buying goods, but you're not buying that from the platform. You're really buying that from sellers on the platform. And so, you know, once you kind of dig into that, you start unpacking it, what you realize is that what the marketplace is selling is actually...

the reduction of transaction costs between buyers and sellers on the platform. Or to use kind of less academic language, what they're doing is they're just making it easier to trade with each other.

Easier even possible. Even possible. Yeah, exactly. That's right. I mean, sellers and buyers are both, in some sense, users of a platform. They're both coming to the platform to get something. And the platform is really... If it's going to succeed, if it's going to make money, it's going to make money because...

the friction of being able to trade, and that's where that word, you know, friction comes in that you mentioned earlier, the friction of being able to connect with the other side of the market is either, you know, reduced or completely eliminated in some cases. So, you know, you asked kind of what are those three friction points that I brought up? And that really comes out of asking yourself, well, what are the things that have to happen for me to be able to trade with someone on the other side, to connect with someone on the other side?

And it's not, let me just preface this by saying, this is not my framework. This is just kind of a common economic way of thinking about things that have to happen for a buyer and seller to be able to actually connect with each other and one of them exchange money for goods or services. So the first one of those is kind of in line with what you said, David, you have to know the other side exists.

So that's what we call search or search and information costs is basically that one role of a platform is showing you, if you're a buyer, that there are sellers out there and showing if you're a seller that there are buyers out there. And then the information part of that is just telling you enough about them so you can figure out whether that's someone you want to buy from. Again, like concretely, if you go to a site like eBay, you want a used iPad, right?

well, number one, are there used iPads for sale somewhere in the world? And number two, like, what do those used iPads look like? Are they scratched up and crappy or are they new and in box? And so, you know, eBay is providing both an inventory of used iPads and enough information about them for you to know if that's the thing you actually wanted to buy. Yeah. Or like, you know, Airbnb. I'm trying to remember.

I think in the very, very early days of Airbnb, there may have been images, but maybe not. Maybe you could have had less things without them. And like, if you're going to stay in a place, you really want pictures. Like, A, can I even, is there a place for me to stay in? But B, like, I want to see it. Yeah, I think that's right. And, you know, it's interesting talking about Airbnb, right? That like a big part of Airbnb as a platform now is thinking about

kind of the quality of images that are on the platform. If you're a host and you want to manage your listing, it's pretty important to have professional pictures in place. And that's exactly for this reason, because

that's a major source of reducing that uncertainty to the potential guest. And Ramesh, how do you think about this first third of the framework, search and information, as it applies to marketplaces like Uber, where the supply is theoretically undifferentiated, where you are actually in a way buying from the platform instead of from the supplier?

I'm actually going to, let's say, take the opposite side here and say that even though it looks like you're buying from the platform, you're still really buying from the supplier. And to make that precise, we have to talk about what's being sold. So it's true in one sense, it's undifferentiated. It's just a car. But the thing is, it's not really a car. When you request a ride on Uber, you're requesting transportation from point A to point B at a given time. That's the thing. That's the contract. That's the thing you're buying. And so you're buying from the

So when Uber is trying to figure out what's out there for you, basically, the question you're asking is, hey, is there a driver somewhere in the world who's willing to take me from point A to point B at this time? That time is usually right now. But you know, sometimes you're willing to wait a little bit. Uber's job is to inventory whether or not reliable drivers exist.

that can take you from point A to point B. The reliable part of it is them, you know, Uber, Lyft, trying to make sure that the drivers who are on their platform have sufficiently high quality or trustworthy, reliable enough. And, you know, do they exist part of it as well? Do we have supply around? Sometimes you open the app and you get 15 minute ETAs. And that happens basically because they're telling you, look, there really isn't any driver around you that's going to get to you in time to take you where you want to go.

So in that sense, you know, I think even with something as highly structured and, you know, kind of close to commoditized as rides, there's still this question of as a user, as a rider, are there cars out there that are willing to take me, you know, from...

where I'm going now to where I want to go. And if you think back to it, like it's, you know, it's funny, they're not that old, right? These businesses are not that old, but you go back to like pre-Uber Lyft days and think about the friction of figuring that out. I mean, you had to call a taxi company, you're waiting for dispatch. We talked about all about this on our Uber and Lyft IPO episode. Yeah. So let me not rehash that. None of us want to relive that nightmare, right? The dark days. Basically the way you think about it then is,

there's an escalating amount of information that you're getting as you get closer to the transaction. So you sort of, all you know at first when you say, I want a UberX and it says four minutes is the sort of the type of car you're going to get and about how far away it is. And then even though you click book ride, you still haven't really made the match to purchase from the person that you're ultimately going to purchase from until it tells you, here's the exact person, here's their license plate number, here's their name, here's their car, all that. Yeah.

It actually goes even further than what you're saying because sometimes, you know, like you request a ride and, you know, for whatever reason, it turns out that the driver is further away than you thought that, you know, you or the platform thought they'd be. Like,

the platform thought they're close by, but they're actually on the expressway. And so then they're like suddenly zooming down. Oh man, I had this happen the other day, literally. Yeah, right. Okay. So, you know, there's even that window after the match where you might decide, actually, this is not quite what I was looking for. Let me try again. It's like they presume the close of the sale, but they give you that opportunity to back out at the last, you know, even after you go. And I mean, it's partly because of that, right? It's like...

It's like that. You need that information to know if the match is the right one. Just to quickly kind of run through the other two. It's actually good that we talked about search information at that length because a lot of what we're going to continue talking about, I think, is going to focus on search and information just now.

to put the hook in, that's going to be kind of the most important distinction between an unscaled and a scaled marketplace is the extent to which it can deal with search and information frictions. But the other two big ones are, okay, let's suppose you found someone to trade with and you know that they have what you want.

So an example might be a guest and a host on Airbnb. The host has a listing available on the dates that the guest wants, and the guest likes the look of this listing. What are the remaining issues? The next one is, can we agree on a price? And on terms of trade is what an economist would say. So it's really like, what's the contract between us? What am I getting and what am I paying you? On Airbnb, that's basically like, what's the nightly rate? You know, what's the cleaning fee? What are all those things?

Let's say we agree to that. We have the contract. It's in place on Upwork, labor market. You know, this might be like, what is the wage I'm going to pay you? How many hours do we agree to? Fixed price versus hourly contract, all that stuff. So that gets agreed to. And the last piece of the puzzle-

And the platform can help a lot in the middle too. Oh, for sure. Yeah. Even from, certainly in like Uber and Lyft, they're setting the price, right? Right. A theme that's probably going to come up is the extent to which a platform is intermediating in helping with that kind of contract setting. And, you know, it's funny, but like one of the earliest examples of a marketplace is really Craigslist, right?

And Craigslist is like the wild, wild west of bargaining and negotiation. It's like that's a place where when it comes to terms of trade... It's only doing certain information. Exactly right. It's out here in cyberspace. It might exist in the real world. It might have used to existed. And this price might be what I'll take. But here's a way to get in contact. We need to do a Craigslist episode at some point. For sure. It's like such a...

like the ideals of the beginning of the internet is like, oh yeah, people are good. They'll like do stuff like, oh man. And especially, I think what I love about Craigslist is that like, at least when I started using Craigslist to try to buy or sell stuff, I think I just didn't appreciate the extent to which like you had to be a good negotiator.

And that was the thing that made me appreciate marketplaces where like that negotiation on the contract was a thing they were helping me with. Oh my God. Imagine if you had to negotiate the price on Airbnb with every time you stay. Like nobody would use it. Yeah. I mean, only like people who really like negotiating would use it. Yeah. It's an incredible friction. And, you know, of course, like listing places to stay, that was a thing that many sites would in principle do. But if the ability to resolve the payments friction is a big one.

And then the last one is, the term of art for it would be policing and enforcement, which is basically like, you know, I can sell you whatever I want and tell you what I'm giving you, but if I don't actually deliver, then it's not really worth much of anything.

So on eBay, this would be, you know, someone's got to follow through and make sure that you actually get the goods you paid for. On Airbnb, both sides actually would want guarantees. So for example, you know, if you're a host, you want to know that the guest is not going to destroy your place. Airbnb has a host guarantee that helps with that. If you're a guest, you want to know that the host can't just cancel at the last minute and screw you over. So they try to do things to mitigate that risk.

On labor markets, it's typically the case that the workers want some guarantee they're going to get paid for work they do. Amazon Mechanical Turk has this issue where if you're a worker, you can often find yourself in a situation where you're doing work and you actually don't really get compensated the way that you would have expected to because the guarantees aren't really in place for that. It's important for platforms to help that entire chain. It's not just finding someone. It's not just knowing enough about them. That's search and information. It's not just arriving at a contract, knowing who's paying what.

for what in return. It's also knowing that that's executed and forced and delivered. So that's the whole chain. And, you know, back to your comment, David, marketplaces that work really help with all of these things. And those are, those are what I call it. Like a lot in the entrepreneurs, especially first time marketplace entrepreneurs, they started and they're like, Oh great, I'm just going to solve the search and information issue. And then while I'll have a, you know, billion dollar company and like a lot,

A whole lot of the early kind of finding product market fit phase for marketplaces is around like, okay, we have a market that we think is interesting. You can solve search and information, but then you got to create the incentives and dynamics such that like the sides are actually going to interact with each other organically. And like that comes down to all this stuff in the middle often. I already hinted a few minutes ago that it's important to distinguish between search and information and these other frictions. Yeah.

And the reason is pretty simple. If you're starting a marketplace, then you don't have any liquidity. So talking about search and information is a bit silly. I mean, it'd be great if I have the most amazing search tool in the world for freelancers. But if I have no freelancers, that's not really doing anyone any good. I guess what I'm fond of telling people when I teach this kind of stuff is that, you know, when you're starting a marketplace, don't start a marketplace and figure out why

Why are people going to show up in the absence of you solving the search and information problem? And it's interesting because like what you said is true. Most people associate marketplaces with solving that search and information piece. But to get started, often you need hooks that are not about search and information, that are about something else. So, you know, some really fun examples I like here, you know, one platform that we used when my kids were a little bit younger in the city is UrbanSitter, which is a platform to find babysitters.

anybody who's hired a babysitter knows this, you know, you arrive at the end of the night and you don't have any cash in your wallet. And that's a total pain. You know, how are you going to pay the babysitter? And so like, there's a very simple friction there, which is just, if I could just pay with a credit card, that would solve that problem.

And that's basically it. You know, that's a product that... What friction is that addressing? It's addressing this like bargaining and negotiating. Basically, the payment's friction. Yeah. Square for babysitters. Yeah. And you get that addressed. You get people signed up with their Facebook networks. And suddenly what's happening is you're building out the network on both sides and...

organically because people are having this payments friction addressed. And now you're building up liquidity. You build up liquidity. Suddenly you can actually do something about search and information. And, you know, Urban Sitter in their trajectory was able to change their monetization approach from getting you to

pay so that you were basically being able to use your credit card to something where they could charge you for contacts with babysitters and so on once they had a little more scale. Yeah. This was a big kind of eye-opener for us at Wave when we were starting. And so much so that we actually had Ramesh come and talk to our investors this year at our annual meeting, mostly around this topic of like, hey, not all marketplaces...

can or should actually start as marketplaces. Sometimes you need to do things to kickstart the market to be able to actually then build into a marketplace. I'm not sure about this term, but I think calling it hacking the cold start is not unreasonable because it's probably going to be the case that there's something bespoke about it in each setting. Now, that said, I think there's a few things that one can think about that maybe at least help give you patterns of

So, okay, the first of those we kind of already alluded to, which is, you know, look at the target

that your target market area that you're thinking about and ask yourself of those other two frictions, is there something I can address? Like I said, often that's payments. Sometimes it's, you know, it's guarantees. It's providing confidence and quality or making sure that people know what they're getting. This is often the case in labor markets in particular, where, you know, you're not in the same place. How do you know that what this person is telling me they're doing is actually getting done? Yeah. Even our portfolio company, QuotaPro, which is a tech-enabled scrap metal recycling brokerage,

We were just talking about this today. If you just threw up an open marketplace and said, hey, I've got scrap metal to trade, it wouldn't work at all. You would have so much fraud and so many claims and all this. One of the big things that Quotapro did in the beginning was do policing and enforcing and say, hey, we're actually documenting and we know what the materials are that are going into each of these containers. Yeah.

I think escrow in general is a really good way to accomplish this. If you're trying to figure out, you know, what is going to be the tool that gets me to have supply and demand sort of converging into my software, even though it's not a marketplace, like just offering good escrow service that seems like it has the right feature set for the people in that discipline seems like a really good way to just create some value out of the gate without already having massive supply and demand on the platform.

It's interesting you mentioned escrow, you know, with Odesk, which is the predecessor to Upwork, where, you know, you mentioned earlier, I was on leave there running data products. You know, one of the interesting things about Odesk was that they have both employer and freelancer guarantees.

And the guarantees basically tell the freelancer, look, if you work an hour, you're going to get paid for that hour. And they tell the employer that if we told you the freelancer worked an hour, then they worked an hour. And, you know, that was really important because what they did is they instituted this kind of time tracking and like keyboard and like mouse event tracking software. Yeah, I remember.

You could think of that as kind of oppressive. But the idea from the freelancer side is like, look, this monitoring is allowing Odes to certify to the employer that you were doing what you told them you were doing for that hour.

And you're going to get paid. Because we can do that, we can hold funds in escrow, which is exactly what you brought up, and then you're going to get paid. Yeah. It's not like if you were a salaried employee and this were happening, that would be like 1984. Right. But here you're working for lots of people. Like you want to ensure that you can get your money. Exactly. I think that's a great point. And that idea of setting up trust in an environment where trust is naturally hard to come by, you know, it helped them kind of get to the point where there was a little bit of scale. Yeah. Yeah.

Yeah. These are all these things that like, you don't have to think like, oh, I want to start again. You want to start a marketplace. So many people's minds go to the search and information problem right away. And it's like, solve some of these other problems first. Right. And then let's talk about search and information. Yeah. And I wanted to mention just a couple others on this cold start issue. One of them is kind of obvious, but it needs to be thrown out there, which is just pure subsidy.

And you look at some of the markets that kind of started, I mean, capital is obviously cheaper now than it usually has been. And certainly one way you could solve this problem is basically by subsidizing one side of the market to bring it on. Subsidies can amount to like giving away something for free. It can also literally be paying. So let's say that you're in a labor market, you need supply in the form of workers or freelancers to be around. You can offer literally like payments to make that happen.

So that's one. And then the last one I wanted to mention is actually a kind of an interesting growth hack, which is to think about marketplaces where buyers can become suppliers. So think about eBay. If you use eBay and you had a good experience as a buyer, you're much more likely to be someone who's going to go sell something on eBay in the future.

Airbnb too, yeah. Yeah, Reid Hoffman is fond of mentioning this as an example of kind of how to get a marketplace going. If you can see it in a situation in which maybe there's a kind of a side of a buyer that could be a seller, then if you can tap into that, that's like a built-in way to get the engine humming.

We want to thank our longtime friend of the show, Vanta, the leading trust management platform. Vanta, of course, automates your security reviews and compliance efforts. So frameworks like SOC 2, ISO 27001, GDPR, and HIPAA compliance and monitoring, Vanta takes care of these otherwise incredibly time and resource draining efforts for your organization and makes them fast and simple.

Yep, Vanta is the perfect example of the quote that we talk about all the time here on Acquired. Jeff Bezos, his idea that a company should only focus on what actually makes your beer taste better, i.e. spend your time and resources only on what's actually going to move the needle for your product and your customers and outsource everything else that doesn't. Every company needs compliance and trust with their vendors and customers.

It plays a major role in enabling revenue because customers and partners demand it, but yet it adds zero flavor to your actual product. Vanta takes care of all of it for you. No more spreadsheets, no fragmented tools, no manual reviews to cobble together your security and compliance requirements. It is one single software pane of glass that connects to all of your services via APIs and eliminates countless hours of work.

for your organization. There are now AI capabilities to make this even more powerful, and they even integrate with over 300 external tools. Plus, they let customers build private integrations with their internal systems. And perhaps most importantly, your security reviews are now real-time instead of static, so you can monitor and share with your customers and partners to give them added confidence. So whether you're a startup or a large enterprise, and your company is ready to automate compliance and streamline security reviews like

like Vanta's 7,000 customers around the globe and go back to making your beer taste better, head on over to vanta.com slash acquired and just tell them that Ben and David sent you. And thanks to friend of the show, Christina, Vanta's CEO, all acquired listeners get $1,000 of free credit. Vanta.com slash acquired.

You've talked to us a little bit about this continuum of you've got a pure marketplace on one end. In the middle, you might have something like a platform. We can maybe come back to that, maybe not. But in the beginning, you could have something like the simplest instantiation of entering a market would be just as a service. Would you consider that almost like a subsidy? Like say, given market, I know there's demand for something, might be hard to attract supply. What if I just go employ a bunch of supply and then sell that as a service to demand?

Is that a good route to potentially becoming a marketplace? Yeah. I mean, I think when we had that conversation about service, then platform, then marketplace, I think part of what we were getting at there is that the service side of it is really kind of addressing one of these frictions that we talked about early on. I mean, one thing to keep in mind is there isn't really anything else on the table. You don't want to ultimately be in the business of selling

something, you want to be in the business of monetizing these frictions. And I already laid out what they are and there really isn't anything else. So you kind of know that if you're selling that service, that the service is really trying to address one of these. Like payments would be a service, right? Like if you're a payments processor, like that's kind of the service you're offering.

When you turn into a platform, I think my view of that was that that really happened when the data that you were collecting through the service was making you much smarter about both sides of this market. And you really become a marketplace when you're able to then deliver on search information and matching. A different way to think about that kind of service initiation piece of this really is that like that's a different way to talk about what it means to start by minimizing one of the other two frictions and then, you know, getting yourself bootstrapped. Yeah.

Cool. Well, our last sort of topic for starting a marketplace was addressing the cold start problem, but I feel like we've addressed it. Yeah, that's right. Great. We have solved the world's problems already. We can all go home now. Turns out after you start a marketplace, as Airbnb and Uber and Stitch Fix and Upwork and others can tell you, you still have a long way to go. Let's talk about the building phase. Like, okay, you've started something, you know, whether you're fully on the continuum of

to a full marketplace and you're doing information and searching information yet and discovery or not, you have something going on. What are the key things during this phase? Like, what do you want to be like capturing and building, you know, about attributes, about supply and demand, various data signals, propensity to transact? Yeah, I actually think that's a really, really good and kind of very timely question as well, because I've been thinking a lot about what scale means. So as an example,

Let's say I give you two different marketplaces. One of them, I tell you, has like 10 million buyers and sellers on each side of the platform. And the other one I tell you has only like 10,000 buyers and sellers on each side of the platform. We can think of examples of each kind. I mean, one market where I talked to someone recently is RigUp, which is a labor market for contractors that work on energy platforms. Yeah, really interesting company. Yeah. And, you know, that's clearly quite a bit smaller than, say, something like LinkedIn, right? So definitely there's markets at all scales.

I mean, I think we're all conditioned to think that like the 10 million, 10 million market is inherently just a bigger market, a more scaled market, probably a more sustainable market than the 10,000, 10,000 market. And I think one thing I'm really starting to grapple with is that that may not be true. And the reason that might not be true is because even though you've got 10 million and 10 million,

If you haven't really found a way to make matches efficiently between them, if you haven't delivered a lot of value to either side, it kind of doesn't matter how many people are signed up. You know, when we count profiles, we're really just counting the number of signups we got. That may not mean anything when it comes down to the actual liquid match you make up. Right, because LinkedIn, you know,

is not like you could call it a marketplace like but like there's not a lot of transacting that's happening on the platform. There's definitely a marketplace certain of their businesses are marketplaces like you think of LinkedIn recruiter. Now I'm not going to make the case that this is actually true but you could one could postulate that AngelList

their recruiter tools are actually better than LinkedIn's recruiter tools because everybody on it is either looking to join a startup or a startup. Whereas LinkedIn, it's like this whole morass of everyone ever and all of the jobs ever. And like, maybe it's harder to make those matches and bring the right high quality candidates onto the search results. And the first 10 results on the page, when you compare it against, hey, everybody on this other place is either a startup or looking to join a startup. Yeah.

Yeah. And actually, honestly, I probably don't even need to be here. You really hit the nail on the head for kind of what I was about to bring up, which is the reason I bring up this like 10,000 versus 10 million example is that

$10,000 might be the better play when you're able to deliver a much more kind of known sense of quality on each side of the market, right? So, you know, why is rig up kind of a compelling idea? It's because there's enough structure in what it means to be a contractor on an oil rig that they can actually deliver a matching process that's really tailored to that environment.

why is it hard when you try to make matches on LinkedIn? It's basically exactly what you just said. It's that there's so much heterogeneity that you've got this platform now that has the benefits of liquidity, but the curse of liquidity, which is that you better be able to explain every single variety of things that can show up on your platform. It's not that it's unsolvable. I mean, you look at Amazon, obviously, it's like a massive marketplace and they've gotten pretty good at it. But

It does mean that sometimes there will be value there on the table by delivering more structured matching and marketplace design around something which is maybe carved off. Yeah. And so for a company and for entrepreneurs working on something like this, it sounds like what you would want to do is really understand why

what the nature of a transaction is and what are the propensity... What are the attributes of each side that supply and demand are looking at when they're evaluating the other side and do everything you can to build product around that? Is that a good...

Yeah, I think that's right. And what's interesting about what you just said is, you know, I previously emphasized a lot that bootstrapping might amount to reducing bargaining, negotiation, policing, enforcement costs. But actually, there's a way in which bootstrapping might involve reducing information costs, right? It may very well be that even in the absence of liquidity, if you just made it easier for users to be able to evaluate information,

Maybe you do that by giving them benchmarks that make it easy to compare against, right? Like as soon as you've done that, you've really made it, you've already kind of potentially gotten that engine humming a little bit. And what I find interesting is that sometimes scaling back your focus is

can actually really help you in designing that product better. And so I know that in the lead up to our interview here, you had pointed me to a podcast of Bill Gurley's that you were listening to. And I think it was interesting. One of the points he makes there is that... Bill was on the Invest Like the Best podcast where he makes this point. We'll link to it in the show notes. Yeah. And so what was cool is that this idea that maybe you started something in a city and you're scaling up really well in that city.

our instinct is like, oh yeah, let's go find 15 other cities I could do this in. Maybe the right answer is you should nail it in that city because maybe there's something unique about the city that you're in that's giving you a chance to really scale up. And I think something you should not forget when you're running a marketplace is, the phrase I like to use is durable market advantage. And what you really want is you want to be in a place where you've

attained the barrier to entry that's implicit in what a marketplace is. Because if all the buyers are in one place, that's where the sellers want to be. If all the sellers are in one place, that's where the buyers want to be. If you nail that in that one city, you've bought yourself time to figure out what needs to happen elsewhere. You know, compare that with like the equivalent of spreading out to a bunch of cities that are all different is exactly what we were saying about LinkedIn with having potentially a lot of different markets that you need to treat simultaneously. It's almost like if you're going to

expand the scope of your marketplace to be very broad, you better be very, very good at basically data science, at figuring out how to classify the type of supply that each of those individual demand points are looking for and how to be an effective matchmaker when you do have this vast sea of things you can't assume that you could have assumed if it was a much more narrow pool of supply. Right.

which maybe I'm jumping ahead a little bit, but really brings up sort of the value of data science in a modern marketplace.

Well, let's go there. We have one of the best people in the world to talk with us about this. We've talked about this a little bit on the Rover and Dog Vacay episode, on our episode with Phil from Rover. Based on what I lived through and saw at Rover, I've kind of been of the view for a long time that really a huge, huge portion of the product of a marketplace is data and data science and getting really good around this match and everything that we're talking about. Yeah.

I assume you're going to agree with me. Yeah, actually, I do. And I think that one of the things I really want to call out there is that marketplaces, I think, I've given you a lot of ways of distinguishing what they're doing. But another maybe important dichotomy is between marketplaces that are what I would call closed cycle versus open cycle. And by that, I mean that there's some marketplaces where the buyers and sellers go there and the match happens, and maybe they don't collect a ton of information post-match.

So think about like advertising, a lot of advertising markets, you know, it's not really clear if the platform eventually finds out, you know, how well that ad worked out for you, right? Because somebody goes to your site. The famous closing the loop problem. Yeah, that's right. But there's a lot of marketplaces that are really closed. So like Airbnb, you know, the payments get processed through Airbnb. You know, there's a lot of incentives provided for both guests and hosts to leave ratings for each other.

Upwork, it's the same thing. Employers and freelancers rate each other. Upwork really manages the whole thing. The thing that actually drew me to Odesk in the first place was this idea that labor markets are as old as human history, basically. And here's a closed labor market.

potentially for like the first time, you know, this like at scale digitized closed labor market. It was as a scientist, it's like an amazing experience to see that. Man, Odesk makes you, once you sort of glimpse the very few types of jobs and work that is actually done in Odesk, it makes you just like,

yearn for the future where maybe all labor is done that way. I don't know if maybe it's overly idealistic, but it does sort of paint a dotted line toward could we live in a world where sort of all work is hyperdynamic and very modular in the way that Odesk labor is in this very sort of narrow segment.

I'm going to be careful not to go there because I feel like you guys could do an entire episode on the future of work and what that's going to look like. And I have a bunch of colleagues, you know, that spend most of their time thinking about this. Why are we not there yet? You know? But at least like thinking about it kind of more, a little bit more narrowly through this data science question that you asked about. I think the reason this kind of closed cycle marketplace is really a fascinating study in how data science can transform business is that

If you think about what we need to be able to make better matches, the first thing you would ask for if you were a data scientist is information on past matches. Right?

right i mean anyone who knows anything about machine learning what do you need to do good machine learning you need kind of labeled examples right you what worked what didn't work well if you run a closed marketplace you're just generating these examples every day because you're a closed marketplace so what's great about kind of a closed cycle marketplace is that you really get constant training data to be able to make matter matches from your marketplace itself and it

You know, that kind of cycle back from having algorithms that provide search, provide recommendations, provide matching technology, provide pricing, seeing what the outcomes were, then feeding that right back in to make it better the next time around. That's actually probably, I think, one of the most exciting aspects of data science in a marketplace. I mean, it presents a lot of interesting challenges, too. So you have to be careful not to just kind of convince yourself that what you're doing is great because...

because what you're doing looks a lot like what you did before. You really need to make sure that you're constantly experimenting and trying new things. Kind of an interesting aspect of data science inside these marketplaces is the role of experimentation and making sure that you're able to confidently validate new strategies. - Yeah, I wanna talk about experimentation in one sec, but just to put even a finer point on everything we were just talking about in the role of data and product at a marketplace,

It always blew my mind, you know, how much I learned from the Rover experience and then here learning from Riley about at Airbnb. You know, when you run a search on both of those marketplaces, like you talk to people and they'd say, well, I don't understand exactly how data is helping me and how machine learning is making this better. And you don't realize that when you run a search on Airbnb, the results that you see

are very different than the results that somebody else would see who's running a search concurrently in the same place. And like the, the function of the search bar and the results that are given, like that is like has such an impact on conversion on the business, on your propensity to book or not book, even when the user on the demand side is actually choosing supply versus the marketplace making the match for them.

Yeah, and I mean, I think what you said is true even in marketplaces where maybe it's not so personalized, either because of sort of fairness or ethics considerations or just because it's not feasible to do so. In the end...

One of the most important things that search engines did for us in principle was take a litany of options and make it into some structure that we could interact with. The cognitive friction of sorting through options will basically never go away. Arguably, it's gotten much worse in the last decade. That's like a very fundamental role of data science. And I should add that that's a place where data science meets product design pretty immediately. At

at least speaking from the data science side, there is a hubris that can set in where you kind of say like, I'm going to let the data speak and I kind of know what's best because I'm close to the data. One of the things that's important is that, you know, when you're building search and recommendation tools, these are things where they're really right in front of the user. And, you know,

There's often small changes to the way those things are designed from a product perspective that have a first order impact along the lines of what you're saying. So I guess one plea there would be, you know, to ask data scientists to work productively with their PMs. Yeah, exactly. Yeah. Well, like a good point, you know, I think maybe one example of that is like,

you can optimize your way into a local Maxima, but not a global max. But we saw this at Rover with, um, in New York city in Rover, when people would run searches for sitters, for, for their dogs, you know, we returned results on the map, same as we did in many other cities.

But we didn't realize that like transportation in New York works really different than in other cities. And actually what users cared a lot about was where sitters are on subway lines that they were close to. Not as much on like geographically how far away is this sitter from me. And so we totally revamped. But if we had just like run our machine learning algorithms on it that we were doing across the rest of the country, like we wouldn't have necessarily seen that. Right.

Another kind of similar example like that is that a lot of search engine, I shouldn't call it search engines really, but just like when you start and you display search results, the simplest thing to do is display them in time order. So like eBay, right? What's the simplest way to display listings? Just like which ones show up most recently. That's really sort of an early stage solution. And you could see lots of reasons why just listing things in time order is probably not the best thing. I think Craigslist might still be in time order.

Yeah, yeah, yeah, I think you're right. Simply times. But, you know, I think it also creates like weird marketplace dynamics because it means that you're constantly delivering the most visibility to things that are most recent. And that makes it a lot harder, you know, think about freelancers or employers and labor markets. It makes it a lot harder to really kind of build up a profile because you're not getting necessarily rewarded necessarily.

for good outcomes, right? So thinking about the ways in which visibility becomes almost like a currency, like something which you want to manage and you want to leverage so that you're directing match attention in the right way, you know, that's an important part of our business. Well, and then as you build up the corpus of supply and demand,

if you're growing, you want new supply and demand coming on the platform. Where do you see them in these, you know, in the rankings and in the list results? Like, it's super important. So many fun stories about that. So, I mean, I think one thing I've kind of always found interesting, this is a pattern across many, you know, different examples is, okay, like, you know, you're

you're watching your numbers and then you notice one month that your new buyers aren't doing quite as well. Like they're not converting as well or whatever. So you're like, all right, you know, I want to treat them better. How do I treat them better? I'm going to connect them to my most experienced sellers.

But then you find out that, wait a second, your experienced sellers aren't as happy as they used to be. Yeah, they like transacting with experienced buyers. Yeah, exactly. So you think about changing that over and you're just kind of playing whack-a-mole with the batches. And this really brings up something important, which is many of these markets are what are called matching markets, which basically says, if I'm choosing to match this buyer to this seller, it also means I'm implicitly choosing not to match them to other sellers. Right.

And that externality that it's, it's very different from Netflix is how I like to put it. Like on Netflix, the fact that you watched a movie doesn't mean I can't watch that movie. But on Airbnb, if you choose to stay at a given place on a given night, I can't stay at that place on that night. Yeah. I'm laughing. I've been thinking about this a lot recently because Jenny and I host on Airbnb. Whenever we're out of town, we list our house. And recently we've been getting a ton of requests from,

from people that are new to the platform and have no reviews yet, which means obviously Airbnb is sending them our way. You got to auto-disable that. That's a recipe. We have it so that you can instant book if you don't have any reviews, but you have to contact us. And it's like, you know, it's really frustrating. Sorry, no social proof, no validation that you're going to keep my house safe.

I guess the way I'd like to talk about is what's the level of certainty of the thing that you're getting from the market? The simplest example is impressions versus clicks versus conversions in advertising markets, right? It's interesting to look at that funnel. If I'm buying impressions, obviously I should be paying less for each of them because they're kind of worth less to me and the idea that they're going to convert or not. There's some more uncertainty there. Clicks are one step further down the funnel. Conversions are kind of the best thing.

What happens is though that you're basically transferring risk over from the buyer to the platform as you go down that funnel. If I'm selling you conversions, I'd better be making sure they're conversions. That transfer of risk, that is an important choice that's made in some of these advertising markets. And I do think like, again, you could do a whole episode on the distinction between like markets for impressions versus markets for clicks and advertising and how did they, you know, how did each of those emerge the way that they did? I think

this instant book example on Airbnb is another one of those. Like thinking through like, what's the level of certainty of the contract that I want to offer? The lower the certainty, kind of the more liquidity, but the lower the prices and then kind of the less sure people are about what they're actually doing. I mean, actually, I think that from a product perspective, Airbnb has done a really nice job here, which is like, if guests who have reviews...

allow a really simple setting to allow them to instant book. But if you don't have reviews yet, you have to communicate a bit first because then that does leave it like hosts. You can decide, maybe sounds like Ben, like you do is like, I just ignore those requests, you know, or I decline them or you could decide to accept them. The leverage shifts all around. If it's two days out,

You know, I might accept that request if it's two days out and it's for an eight day stay and then they're really great communicators. I'm probably going to accept that request, but it shifts the sort of burden of responsibility of that trust away from a feature of the platform to letting the two parties sort of decide that for themselves.

Yeah, I think that's right. I will say just as a kind of a side thread inside all this, I do think that one of the most important metrics the marketplace should be looking at is the long-term value of each side of the market. And in other words, you know, to use the language you're using earlier, I think it's really important to be asking yourself, you know, how happy are my experienced users? And the reason that's so important is because you're hoping the new users become experienced users. And what they're hoping is that, you know, it's worth it to be

to become an experienced user. So I do feel like it's best to work your way backward from that. If that metric isn't looking good, or if there's, you know, you may have high liquidity, but high churn, that's not a good place to be. Some marketplaces would often be better served to really make sure that you're delivering long run value for the users you have there. And then worry about that growth aspect. We're, we're so focused, I think on growth in a, in a kind of cheap capital environment that you can sometimes lose sight of the fact that, uh,

people need something to grow into. If the value's not there, they're not going to stick around. We have this conversation with almost all of our portfolio companies very regularly. I mean, as entrepreneurs, they're struggling with how to balance growth versus quality and looking to us often to give them signals about that, about like, oh, what is our next round of investors? What are they going to expect? And

almost always we're like, look, you have to grow. If you don't, you know, startup equals growth, right? In the words of Paul Graham, if you don't grow, you know, you're either you're dying or nobody wants what you're selling. But we really encourage them to do it in a like, in a way that is not compromising long-term quality. If your churn metrics are high,

you basically need growth to make up for the churn. So a simple way to address growth is just don't churn as often. Either you could dump a lot more capital into the business or you could just fix the holes in your bucket.

I know you want to talk about experimentation. This is like a cousin of data science. I'm usually part of data science, but... I think, you know, one of the things I just wanted to mention is that I really feel like the kind of marketplace businesses that, you know, we all have heard about have really, they've been on the vanguard of A-B testing and experimentation as a way to improve the platform. And I think A-B testing and experimentation, they really have such an important role to play in these markets because...

there's so many externalities of changes that you make. And many of the choices that you have to make do involve trade-offs.

some of the things that you're going to do, they're going to divide the pie differently as a way to think about it, right? Like we talked about experienced versus inexperienced users on either side of the platform. But there's other examples of that too. I mean, there could be things which are dividing the pie differently between buyers and sellers as well. And there could be things that are kind of making it easier for certain types of sellers to be able to connect with matches on the other side. All these different kinds of things. I think it's important to understand that

so much of the choice that you have to make is like, where do you want to operate? What trade-offs are you interested in? And evaluating those, those trade-offs, I think to be quantitative about it, uh, having an effective experimentation methodology is a pretty big deal. Now, one thing I will say is that I know, you know, most of the listeners of the podcast are probably fairly early stage, uh, either about to start or have just started. And,

Maybe. We have plenty of folks that are listening at big established marketplaces too. Yeah, great. I think it's good to think through when do you do this, right? Because it's a fairly kind of cost-centered task to build up that infrastructure. But I do think that as you start to scale, it's something to make sure that you kind of give A-B testing like a serious... Well, Riley often talks about as they were building up this discipline at Airbnb...

he and they implemented this concept of backstop metrics of like run lots of experimentation to optimize, you know, whatever your goal is on a particular metric, but do that,

without harming other metrics that are up funnel or down funnel from where you're focused by a certain amount. And that was what their backstop metrics were. So there were some guardrails of like, because you could go wild and do something that would like, oh, my objective is to increase conversion rate. Like, well, great, I'm going to lower my prices. The notion of guardrails, one really great thing about that exercise of setting up guardrails is it does force you to ask questions.

like what matters to you and your market. So, you know, an example of this is that almost every marketplace is going to care about conversion rate from, you know, essentially like if you look at, you know, buyers that come on the platform, what fraction of requests or, you know, whatever it might be like jobs convert to actually being filled.

that's probably a good guardrail metric. Because you really don't want to be tanking that with something new that you do without some pretty good justification. And so I think thinking through these types of guardrail metrics down the funnel on either side is super useful.

The flip side of it is that sometimes you really have to kind of think outside of the box. And I think for many marketplaces, that moment is going to happen when they switch from that early cold start phase where you were like bootstrapping up to where you really want to monetize search and information. Because I think one of the things that changes there is like, look, maybe at that point it's

going to become okay if some of your customers are not actually paying for the thing they were paying for before. Because now what you're really trying to get them to do is switch over to this model where they're willing to pay to be able to connect to the other side, right? That can have big influence on a lot of these core marketplace metrics, but it makes sense because you're changing your business model, basically. So, you know, it's one of these things where you want to be aware of what stage of the journey are you on and tailor the metrics and your expectations of guardrails to where you're at.

Ramesh, I'm going to go off script here for a minute. I think we would be remiss to do an entire episode on marketplaces and not at least touch on take rates a little bit. And obviously we could do a whole episode on this, but in your mind, at a high level, what goes into figuring out how much of the value the marketplace itself can capture?

Oh, man. Yeah. I'm glad you said this could probably be a whole episode that lets me off the hook about having to expound forever. But yeah, that's a very, very big question. Something I just want to highlight there is that you asked me how much of the value can the marketplace capture? I think at least as important a question is from who?

One thing I found really interesting about this is that from an econ perspective, it really shouldn't matter from who. So what we mean by that is like... Right, because there's only, you know, there's money in a transaction. Yeah, right. I mean, let's say you charge the seller, like they could always pass it through. So this is, again, it's one of these places where like...

the kind of technical data science side of the world has to hit reality at some point. It does matter. It matters if you're the buyer or the seller paying. And there's plenty of evidence that the exact same fee charged to the either side, you know, it doesn't yield the same outcome. So that's rule number one.

And so I think it does matter. And that has something to do also, by the way, with where is the friction getting resolved, right? So like, if you're asking, I mean, the most extreme example is if you charge someone a subscription fee to be able to get into your platform, they don't know what they're paying for when they subscribe. And they don't know what value in return they're going to get. So there's this high degree of uncertainty or charging me up front for it, right? Contrast that with, like, say, again, like paying for conversions and advertising, like

high degree of certainty for what i'm buying you know it's a different take rate potentially that that could work but okay so in terms of like figuring out how much of the value you can capture

Again, there's so many different aspects of this, but I think one thing I'll emphasize is, you know, how much do you expect them to return to the platform? Because really, when you say value, right, the value isn't just the value of this match. It's really kind of... The long-term value. The long-term value of that match to each side. So you've got this participant that comes in, you know, buyer and seller, they match. They go away, they're still a buyer, they're still a seller. And what's going to happen after that? Are they going to match with each other again? Yeah.

Are they going to match with other buyers and sellers again? And, you know, so when you're asking yourself, like, how much of the value can the marketplace capture? I think you have to think through some of this downstream value. So there's a lot of marketplaces that are just basically lead gen where you connect and then you go your own way, right? So now the goal of that marketplace is really to capture as much of the value as they can up front. And that would be common in scenarios where you only are a buyer or a seller of that thing

one time. You're not going to be repeatedly doing that over and over and over again. Or perhaps you're monogamous in your relationship with that buyer or that seller. You're going to match once and then you're... Yeah. I mean, I like to joke with my wife about the fact that dating markets kind of shouldn't get too good at their job because you don't necessarily want people to never come back, right? So it's important that not all the matches... I think I've seen...

billboards like for hinge maybe when I was in New York that they have a campaign going right now like something like we love it when you leave the platform

In that case, really what's happening is that you're not even worried about disintermediation or any of those things, like the usual kind of things that marketplaces are worried about much more repeated interaction worry about. By contrast, I guess I have to say that in most of the instances where I've been involved with marketplaces, kind of a metric of interest, and this goes in line with thinking about long run value of your experienced users, a metric of interest is what's the frequency at which people return?

So I think one thing to think about when you're starting is like, there's usually should be at least one side of the marketplace that has reason to come back often.

So if you think about Airbnb, right, like the, you know, the vacation traveler on a couple vacations a year, three, four vacations a year, maybe 10 on the upside, right? Your business traveler, maybe. Right. And so that's not like super frequent. Yep. But the host, they've got a lot of nights and they're basically managing those all the time. So at least one side of the market has, you know, that high frequent interaction with the marketplace. Same thing on labor markets, you know, freelancers at Upwork, Odesk, you know, they are kind of there, you know,

all the time. And so that is one of those metrics, I think, to think about. When you find a marketplace where really both sides are fairly infrequent, that's much more challenging because now what's happening is that you're not really benefiting from that

you know, repeated kind of closed cycle marketplace that I was talking about. It's much more one-off matching as you talked about, Ben. And the data science, you know, benefits, the network effect benefits of that are not quite the same. I remember having discussions, you know, again, keep talking about Robo on this episode, but it's my most hands-on, you know, marketplace experience. When WAG first launched,

and they were doing just dog walking and we were getting into dog walking too at Rover. And I think we just merged with dog vacay. Wag had a 40% take rate. And we were like, this makes no sense. Like a dog walker is like a very high repeat transaction, you know, model, but on both sides of the market, why would you have such a high take rate? Interestingly, I think it survived for a long time. I don't know if they've lowered it since, but, uh,

We're just like, God, this should be much lower. Well, you know, as with many things, I think the initial take rates, like who the hell knows, right? Yeah, you just pick a number. One thing that's changed, I think, is that there is a little more, I don't know what to call it, kind of social norming around take rates. Like,

You know, there's like the norms of around 15, 20% in the ride sharing industry, you know, in labor markets, it's kind of 10%, something like that. In lodging, it's sort of 10, 15% range. So I think it's rare now that you'll see like the 40% take rate. But maybe one other thing I'll just add to this is that it is useful.

for entrepreneurs to retain some flexibility in exactly what that take rate is going to look like. And you can do that a number of different ways. You can do that basically by explicitly saying, you know, the final take rate is going to be something between X and X. And, you know, they'll be determined as a function of kind of the transaction, the timing, all that stuff.

Or, you know, you could say like our take rate's actually something high and then you're discounting it down for now. You know, there's a number of different ways to handle it. But one of the reasons that's so important is that the marketplace changes over time. It goes back to this theme that a marketplace is not always a marketplace when it starts. And that means that what you're monetizing and therefore the rate you're monetizing it at could look quite different now and in the future.

You don't want to hamstring yourself into saying that, you know, every transaction is going to be exactly the same. There's going to come a time where, I mean, if you think about search and information, if that's really what you're monetizing, you know, that maybe does mean you want to shift to a model where you monetize a lot of the match up front. If what you're monetizing is, say, payments or something like that, and payments is an ongoing aspect of the relationship, then that might be something where it's like a percentage over time.

Those are really different. And if you commit that, hey, I'll just never change my take rate, you know, you can potentially lock yourself into something you don't want. So Ramesh, you might be able to help me dispel something that I've sort of thought for a long time. You can tell me sort of myth or fact. But the reason why Uber or Lyft can justify almost twice as high of a take rate as Airbnb is

is because the supply is a lot more commoditized. And any platform where you have a relatively commoditized supply, the quote-unquote work or more of the work or more of the value is being delivered by the platform itself rather than the supply itself. Whereas if you have something extremely differentiated, like unique vacation homes everywhere, then they can actually justify taking more value in that transaction. Is that...

true or am I overfitting some model to very few examples here? So let me preface this by saying, and I probably should have said this sooner, that definitely anything I say on the podcast, it's my own view and should not be attributed to any of the companies. You got it. I mean, you work with every great marketplace company close to it. So it's not like, yeah, we can't stay away from everything.

I think one of the reasons it's hard to answer that question observationally by what we see in the market is that we're not in steady state yet. You know, and I think you could see that in the dynamics of the ride sharing industry and most other marketplace industries. So, you know, when we think through, like, what is the sustainable take rate, kind of

Kind of one of the challenges is that we don't know what the eventual market equilibrium is going to shake out at, right? And taking the way you framed it and putting out the pros and cons of that point of view. So on one hand, it's true that when things get commoditized, then potentially the value of the platform is relatively speaking, at least relative to the supplier, is higher. Any informational advantage the supplier might have had sort of starts to evaporate if there's a commodity there.

The flip side of that, though, is that when things get commoditized, competitive pressures grow. Competitive pressures from other platforms. And kind of that's what you see in, you know, many of the delivery industries, right? And so that's a little hard to, you know, work out kind of what's the eventual equilibrium because you've got one of the advantages when you're operating in a more sort of informationally heterogeneous market environment, or at least one where, you

and the buyers have some information that prevents it from being a commodity is that it also means it's harder for a new entrant to build up an advantage against an incumbent. And this, by the way, has like- In a situation like that,

It really is a lot of value of being where all the buyers and all the sellers are. Yeah, and I think this highlights something about data science that I didn't get to say earlier, by the way, which is that there's actually two types of incumbency advantages that scaled marketplaces have. One is kind of the obvious one that like buyers are, you know, you want to be where all the buyers are, you want to be where all the sellers are. But the other big one is that, of course, now they have a library of data built up on what works. And, you know, so to your point about commoditization versus not, the thing is that data incumbency advantage of a scaled marketplace is much higher when you do have that heterogeneity.

Because now you know a bunch of stuff that other marketplaces could try to enter don't yet know about what kinds of matches work, what kinds of matches don't work. So anyway, this is all a long-winded way of saying the usual academics answer, which is that it depends. I'm just going to cop out. You did segue nicely into, I think, David's next question here, which I'm going to steal from you because I'm already talking, is, you know,

how do you think about competition, especially in a world where we are now, where there's sort of two things. One is with the advent of the internet and smartphones and everybody being online all the time, we theoretically are in one globally connected market for everything, which is what everyone in ride sharing thought, but it's not quite what materialized. So at least we're in a, uh,

winner take most or some sort of world where there is a race to go and get a lot of supply and a lot of demand quickly. And that, you know, allows you to have staying power. The second thing about the world that we're in right now is one where, as you brought up earlier,

Capital is wildly, wildly available and accessible. And if it is clear that something is working and investors and founders and everyone believes it to be a winner-take-all, there's this massive acceleration into trying to take all. So how on earth should entrepreneurs think about that with sort of this access to global markets quickly through the internet and the capital ecosystem we're in?

When capital is so abundant, I think that's an interesting environment to be in because you can't defend against that, right? So I don't think that should ever be the goal. I mean, you basically have to accept that that's the race you're playing and you hope that you win in that race. I don't think one walks into that saying, I'm going to defend against abundant capital. Because in the end, remember that one of the three approaches to scaling a market is subsidy, right?

And if capital is abundant, you're willing to subsidize. It doesn't matter how much you've scaled your network effect. I just pay people and bring them over. We've seen this so much in a bunch of markets over the last five years of just like somebody might have a better market or might have a better product or might have a better lead or might have a data science advantage. And somebody else comes in and dumps half a billion dollars in subsidies. Yeah. And I think part of that is there's like, again, another episode you guys could do on sort of efficient social capital allocation and

Like, what should society be doing? And I think, you know, a lot of that is tied to the way in which we reward capital. And I do think that as an academic, at least, that's something that does bother me because I think there are a lot of social benefits if marketplaces can scale. And, you know, the good side of marketplaces is really that there's a lot of frictions that they're taking away that, you know, none of us like. The bad side of marketplaces is that with concentration, you know, comes market power and many of the

adverse effects of market power. I think as a society, we're probably not grappling with that on the right terms. It's, you know, I think, so there's a lot, there's a lot that you guys could go down that road, but coming back to this question, we'll have you back for our antitrust episode. Yeah, exactly. But coming back to the competition for, you know, I want to take more like the founder entrepreneurs perspective, like how should they think about competition? And the phrase that, you know, I was telling David earlier that I like to use this durable market advantage,

So when you think about scaling, one part of scaling is that you've brought a lot of buyers and sellers onto the platform. I guess what I want marketplace entrepreneurs to be careful about is the fact that buyers and sellers have shown up, it does not mean they'll stay.

And so the question you want to ask yourself, and this I think is a question you asked from the beginning, why do they stay? So why would you want to still be here even after you've been able to transact? The first piece of that question is you transacted once. Why do you come back again? That comes back to the frequency issue we brought up a second ago. The next piece of that question is you've transacted again and again. Why do you transact with more than just the same counterparty, right? What value are you deriving from my presence here?

One thing I find so interesting about the pace at which the Valley and a lot of these businesses move is that because things are changing so fast, everybody's very short-termist.

But that's, from a strategic standpoint, can be a real problem because you're not thinking through this durable market advantage question. All your lampposts are focused on short-run metrics. You forget to ask yourself, if I scaled, would I have a thing that I could defend against competitively, right? And I do think that's important to think through. You know, I recognize that there's a lot of short-term versus long-term pressure in delivering business value. You know, we can see in many different industries with marketplaces or these competitive pressures, if

If you get scale, but you can't defend the scale, then it's not really kind of, you know, it's not durable market advantage. Yeah.

Yeah. Oh, man. Well, when you come up with the right algorithm for developing durable market advantage, can you please tell us first and then give us a couple years head start? Yeah. And then we can tell everybody else. There's a reason I'm an academic. If I was that good at it, I probably wouldn't be at Stanford. Wait, I'm not going to just let that lie. Come on. I'm sure you're great at it. Yeah.

Well, I think that's a great segue into maybe our last question or our last kind of couple questions here is like, given all of that, if you were to be a marketplace entrepreneur or the marketplace entrepreneurs you talk to or us at Wave, what are some areas that, you know, would be interesting? Either particular markets that would be interesting to attack from or start to build a marketplace in or just dynamics that you would look for?

I know in our early days, one of the interesting conversations was about in talking to potential investors in the fund, what's the scope of the opportunity? Yeah. Oh, man. So many LPs asked us, like, haven't all the great marketplaces already been built? Like, what else is to be built in the future? And I guess what I found out through that sequence of conversations is that really, like, the common mentality is that marketplaces are vertical, right?

There's manufacturing, there's energy, there's like automobiles, and then there's marketplaces. And that was really interesting for me because I don't think of them as a vertical. I think of them as a business type.

And once you think of it as a business type, it's really kind of, you know, running across potentially every vertical. So, you know, when you say what markets or areas do I think are ripe? I mean, commerce is part of every vertical. So in principle, every vertical is ripe for disruption. And, you know, like you brought up earlier the example of scrap metal and I brought the example of, you know, contractors on oil rigs. I mean,

Like, would you have thought of those as sort of pre-existing? I wish we had those at the tip of our tongue when we were fundraising for our first fund. Right. And we could have said to LPs, like, yeah, here are some examples. Here are some examples. Right. And so, yeah, that's my first answer is that I just, I sort of think in principle that the benefits of digitally intermediated marketplaces are basically, you know, the scope for that is unlimited. Right.

There's a different question, which has to do with where you can create durable market advantage, right? And we can have a lot of business value. And that may not be in every one of these, you know, potential verticals, but...

I guess two that I'll highlight that I like, and then there's kind of one other broad comment I wanted to make about this. So one of them is labor. I continue to be fascinated by labor. And I've always had a soft spot for the kind of work that Upwork does. It was great for me to see the value that that generated for the participants in the marketplace. And so I do think that in terms of opening opportunity, there's a lot of room in how we think about labor market intermediation. And one quick aside there, I feel like

labor marketplaces for a long time, people thought like, oh yeah, like LinkedIn or hired.com or like recruiting. Right. And like, I think that was a little bit of a red herring and like, yeah, there was value, but I mean, obviously LinkedIn had a lot of value, but like that's

That's not what we're talking about now. Yeah, and I think that partly has to do with this question of repeat interaction. Yeah. Right? It's, again, closed cycle versus open cycle marketplace. And I think you're exactly right. I mean, I imagine that's kind of what you're alluding to is that a lot of the kinds of labor markets we're talking about now are really where there's more of a closed cycle. Yeah.

And you really do get those benefits of... It's not recruiting of helping somebody hire an employee that's going to be a W-2 employee for them for a long time. Right, right. Exactly. I mean, LinkedIn does have components that look more like this closed cycle marketplace now, as Ben brought up earlier. But yeah, definitely kind of conceptually, I think that's a major distinction. To dive in on labor a little bit...

you know, it seems like the first couple decades of creating labor marketplaces, you know, starting with Mechanical Turk and Odesk Elance, and we started Spare5, which recently sold to Uber, which ended up narrowing tremendously all about data labeling for autonomous vehicles. To dive in on my comment earlier, work that is being modularized and being able to be done on sort of a one-off basis started at the most

you know, sort of basic atomic types of work. And it would feel ludicrous today to say, you know, you're going to go and on a labor marketplace, find the labor that the Fortune 500 CEO does. Like that seems like it's ridiculous that we would ever get there. But it does feel like we are slowly sort of taking on more and more either, I guess, complex types of labor and finding ways to bring them into an efficient marketplace model rather than, David, as you suggest,

do your best to expensively go and generate leads to hire them and then do work within your company.

A lot of that has to do with the fact that we're just getting generally better at structuring information. You know, I think things that we can do to extract information from free text now were not possible, you know, 10 years ago. And that's really changed the notion of what information, say, a job post or an employment profile contains, right? So that's as kind of methods in AI, machine learning, natural language processing, image recognition, advanced, that all of that has changed.

kind of spillover effects on the types of markets you can intervene in.

The other actually meta statement I was going to make about markets is I actually think there's opportunity in sort of platformizing the creation of markets. There are pieces of building these marketplaces that are repeated and that I think we can, you know... It's funny, somebody in the acquired Slack brought this up as an idea recently and I was maybe a little pessimistic on it, but I want to hear your, I want to hear your bulletin. Yeah, I think I understand your pessimism. I think it's a challenging thing to do.

The thing that stands out to me here is that the idiosyncratic part of a marketplace is often going to be defining the characteristics that matter, as we said earlier, on the buyer side and the seller side. Which is exactly why I was pessimistic on the idea. Yeah, but the flip side of that is that the algorithmics of matching and of recommendation and of search is...

there ought to be a way to package those so that at least for an erstwhile marketplace to get started, it doesn't have to do all this from scratch. And, you know, already like there are, you know, platforms for search, right? So are there platforms for matching? I mean, not really. And the question is like, is there an analog of that? Are there platforms for two-sided pricing? You know, is there an analog of that? So that's maybe a little bit of the academic in me speaking, but I do think this is a thing where I think it's kind of like an elastic search, but for marketplace

place matching. For matching and for pricing. Exactly. Exactly. Yeah. I mean, there are enough companies. I mean, to your point, there's now a very successful thriving labor marketplace for oil rig workers. Like, there are going to be a lot more. Yeah, that's right. Exactly. Yeah. It goes back to this thing that, you know, I got to that point because I realized that really you could intermediate in anything. And if you lower the friction of getting there, then of course it makes, it opens up the number of

potential opportunities. That can be built. So the last comment I just wanted to make on this, I said I had kind of one broader point and that's

Partly a personal point, but something I've been reflecting on is just how we can bring what we've learned through this, I don't know, let's call it a decade of marketplace innovation. How can we bring that to sectors of our economy that otherwise probably just aren't going to have access to it? So here I'm thinking about kind of public policy. You know, I live in the Bay Area, things like urban transit. What are the things we've learned here? You know, how can we bring those insights out?

And so I do think like something that's interesting for entrepreneurs in this space to think about are, you know, how that process of matching and, you know, reducing frictions, how does that play out in actually helping solve policy problems? I'm particularly interested in urban policy problems. So simple one here is like housing, right? I mean, matching for housing is a huge problem. Part of that is an entrepreneurship opportunity. But part of it that I would like, you know, us as a sort of technology community to think about

is how we engage more productively with what the urban public policy problems are. That goes for transportation as well. It goes for healthcare. It goes for labor. And every single one of these, there are kind of first order public policy questions that are associated with the markets that they're intermediating into. And, you know, historically, I think that engagement has always

been kind of lagging, right? It always happens. Lagging or adversarial. Or even adversarial, yeah. And my view on these things is that the reduction of frictions should be a net social benefit. So then the question is, you know, how do you end up at a more productive interaction with the public, you know, policy domains that you're influencing that way? Again, partly my academic hat speaking there, but I think in the end, that's kind of a win-win for everyone. Because if you are a scaled marketplace and you're successful with that barrier to entry, you're

you know, you are going to bring on this kind of antitrust magnifying glass and it's better to be proactive and engage with that than it is to kind of be caught flat footed. Amen. Amen. Well,

Well, Ramesh, this has been awesome. Thank you so much for coming and sharing your vast marketplace knowledge, not only with us at Wave, but here on Acquired too. Where can listeners find you? Obviously, your students are very lucky to take your courses at Stanford, but where on the internet can folks find some of your work? You know, I've got my Stanford webpage. It's not the sexiest thing in the

world, but it's, uh, you know, but it's, um, we might need a, we might need a vertical labor marketplace for you. Yeah, that's right. Um, but, uh, but yeah, it's, you know, it's got a bunch of stuff that I've, that I've worked on there. And, um, there are a few places where talks I've given have been recorded. And so, um, in particular, there's, uh, one, uh, it's called the engineer as economist. Um, it's a play on an old paper by Al Roth called the economist as engineer. Um,

And the idea of that talk is really that engineers are building marketplaces. I mean, you think about most of the people employed at these marketplaces in the early days, it's often a lot of engineering talent that's getting it off the ground. Yeah.

they don't really realize that when they make that choice about how bidding is going to work, how pricing is going to work, how matching is going to work, they're doing the economics. They're being economists. They're being economists, exactly. And so I really wanted to get people thinking about what it means to engineer markets and kind of the operational challenges of that. So there's a few talks like that that I think would really help your listeners. We'll link to them in the show notes. Ramesh, thank you so much. Yeah, thank you for having me. We really appreciate it. Yeah, this has been awesome.

We'll see you soon. Listeners, thanks so much. And we'll see you next time.