Three-Sided Marketplace Learnings From Lyft & Shipt With Vishal Kapoor | EM Group Chat #200

Welcome back to Everything Marketplaces, where we talk with founders and leaders from some of today's top marketplaces. So this is episode 200, which is really a great group chat to just have with Vishal Kapoor. Vishal is currently the VP of Product Management at Shipt, which is a marketplace that's enabling personalized shopping and delivery from retailers nationwide. Prior to joining Shipt, Vishal was at Lyft, so some really great marketplace experience. So this is a great chat with Vishal. We're going back to his experience at Lyft. He walked us through some of the biggest challenges that he's working on, shared some of his key learnings from Lyft, We did a deep dive into Ship as a 3-site, more operationally complex marketplace. Vishal shared how they're leveraging AI. We discussed topics like marketplace matching and pricing. We also had a great group Q&A. I did want to mention that we had to edit out a few parts of the conversation that were off the record, but I think you'll find it a great watch to the end. So Vishal, it's been a while since your past group chat, and I'm really excited to have you join us here today for the follow-up and actually a 200th episode. And we have a lot that we're going to dive into with all of your Marketplace experience and also to some of the recent advancements with AI. But before we do, I thought it'd be great if you could start off by briefly sharing your background for those here that have just joined us and might not know.

Thanks, Mike, for having me. You're right, it's been a minute. And, you know, it's an honor, I would say, for me to come back and, you know, come around on the second double century, the, you know, the second century of this. So that feels exciting. I am, my name is Vishal for everybody who's dialed in. I am a VP of product at a company called Shipt. Shipt is a delivery company, grocery delivery, grocery and other deliveries as well. It's a delivery company which is a 3-sided marketplace, right? So you have the customers, the stores or the retailers, and then you have the gig workers that go and pick up whatever needs to be delivered and then actually go into the delivery, either shopping or only picking up and doing the delivery. And I've been with the company for over 4 years, 4 and a half years at this point. I lead, me and my teams, we lead a large portion of the company. Essentially, it's, so we have orders, we have deliveries that are coming in, we have a .com website and we have an app. So, there are orders that are coming in first party that are coming in through that. But we also have business-to-business relationships where orders are coming in through APIs, for example. So, one part of my team leads all of that order ingestion, which is a huge part of the business of all the orders coming in. The second team actually leads once the orders come in, how do we actually clear the orders, what we call as market clearing really? Like how do we If you think of an order lifecycle, how do we manage the lifecycle of an order? The third team actually leads— well, actually before that, another team leads the network design, which is if you open the app, let's say as a customer, what are the stores? How far away stores would you see? What are the stores? What would you see? If you want to buy alcohol, if you want to buy pharmacy, et cetera, what does that configuration, network configuration look like? So that's one. then it's clearing the order. And then the last one is for all the gig workers, we have a separate different app compared to the customers. So there's a, just like, you know, all the other delivery platforms, we have a shopper, we call them shoppers. We have a shopper app or a driver app. And so one of my teams also owns the app and the experience as well. So it's a pretty broad footprint across the company. And, you know, I've been here for a minute and I'm very excited for the chat. Mike, just, you know, reminiscing very quickly, I know that we talked about, marketplace dynamics a lot 3 years ago. And obviously my role has grown and my experience in this company has grown. So I'm looking forward to the conversation again.

Yeah, it's a great kind of quick background and also to overview for sure for us. I appreciate it. So a lot that we're gonna get into here. And as I kind of mentioned, you know, 2-sided marketplaces are challenging. So, you know, you add another, make it a 3-sided, more operationally complex marketplace, there's lots of challenges. So, and I guess to kind of go back to your experience at Elef though, you know, Could you share a little bit more about what you were working on at Lyft at the time and maybe some of your key marketplace learnings there?

So there were a couple things that I was working at Lyft. So before Shipt, I was at Lyft and that's another on-demand marketplace, if you will, ridesharing, which is, as Mike just said, it's a 2-sided marketplace where you have essentially the supply side and the demand side. You have customers and then you have gig workers who are going and driving the customers around essentially. So there my experience was mainly related to 2 areas. One part of my job was to measure and build systems for measuring business metrics internally, right? Like what are the KPIs? What do we look at? Even basic things like what is the definition of a ride or what are rides or what's, for example, what does daily active users mean? Right? Like, is it rides? Are the people who are taking rides, are the people who are opening the app? Like, you know, there's like all these types of things, right? Which, you know, in different contexts, there are different reasons or different audiences, different use cases where the metrics might be different. So one part of my job was that, was measurement and definitions and building products for that. The second more exciting part of my job was working on Lyft's pricing system, dynamic pricing system, which was which was essentially one of the main reasons for gig workers, I would say, one of the main reasons why they, in any gig company, the gig workers come to these roles because they're trying, frankly, trying to make a living, right? So this is a system that actually disperses out all the money, like anything that anybody earns at Lyft. So one part was that, the other part was the other side, which is what do you charge the customers? Because obviously all these platforms like Lyft, like Shipt, like all these delivery platforms, Uber, Instacart, take your pick, they're all essentially aggregators. They're people in the middle and they're trying to provide services to multiple sites, however many sites they may have. So pricing is an important component and one of the systems that I owned there was also managing pricing across the different stakeholders and different users. in the marketplace. So those 2 things, Mike.

That's really great. So we'll probably get into pricing a little bit more detail and, you know, in general for marketplaces here a little bit, you know, later in the conversation. But, you know, kind of transitioning to Atiship, and you kind of gave us a brief overview and then share a little bit more about, you know, what you're working on there. But as far as when it's concerned as a 3-sided operation complex marketplace, could you maybe share a little bit more about some of the specific kind of challenges with it?

3-sided marketplaces are significantly, I would say significantly more complex than 2-sided. I did not realize, you know, when I first came into the job, I was like, yeah, you know, this is more of the same. You know, I come from, come from rideshare, how difficult can delivery be? It's a lot more difficult. And there are a few learnings that I had from— I had slash I have from my time at Shipt. Number one, in general, I would say that any regulated— any kind of regulated marketplaces which are hyper-localized marketplaces like a couple that we just mentioned or anybody, any other marketplaces. It's difficult and it's difficult to innovate or slow to innovate because of the regulations. Because of the regulations, the regulations are not just, it's not just not always national regulations, it's also state regulations, it's also city regulations. So, if you want to run a business in an area, in a neighborhood, in a geography, you have to play by the rules of the playbook in that geography specifically. So, that makes it challenging. slower to innovate, it feels like slower to innovate. Before Lyft, a quick flashback, I actually was at a gaming company called Zynga. And games are the exact opposite of anything you can imagine, like you can do no wrong, right? Like just you concoct anything and you throw it on the wall as an experiment and you see if it sticks and if it sticks, you scale, right? So, it was a very fast learning curve and just really like trying by experiment and seeing what users respond to, what they adopt, etc. When you come to a regulated marketplace, it's not that, you can't do that. You can't just like willy-nilly play with matching models, pricing models, things that actually change the livelihood for some of the people that participate in the marketplace. So, it's significantly more challenging and it feels slower to innovate. But the thing that I learned and I realized after going through it for a while was It's not necessarily slow, it's more the way I— and this is the way I tell myself and my teams, I coach them is it's not slow, it's thoughtful. Like you have to be very careful and thoughtful. And once you do that, once you build a product to respect all the regulations and the laws and all of that, you have a more defensible product. It's more difficult for somebody to come and build a product that does so many things which is necessary to operate in a city or a geography or a county or whatever. It's not that easy because other people will have to spend all the time and the effort to do the same thing that you did. So that's number one. I would also say that the second thing that I learned from— the second important thing, a slight sidebar from what we just discussed that I learned at Lyft, which was very interesting in terms of user behavior or user how users behave was that, and we used to summarize this in a statement that people actually know how to value, people don't know how to value their time versus valuing their money. And I'll give you guys a simple and short story and most of us will be able to probably relate to this. So imagine yourself, right? You book a ride at one of the rideshare companies, take your pick. If the ride And just bear with me for a second and I'm sure at the end of this story you will all agree with me. So you're taking a ride, you open your app, you can get a ride within as fast as 3 minutes or let's say you have to sometimes wait 10 minutes for a ride. Now going from 3 minutes to 10 minutes is more than 3x, right? Like you're waiting more than 3x the time but generally what we saw was at scale we saw that people are willing to wait let's say up to 8, 9 minutes, there was a cliff point. After that, when the pickup ETA exceeds a certain cliff point, people say that this is too long, I'm unwilling to wait, and they book another ride, right? So, or they bounce to your competitor. So that's one example. But take another example. If you open both the apps, right? Like both the rideshare apps in US today, and you look at the price difference, and if let's say if you see the price of the ride on one of the apps is $10 and the other is $11, You will like 9 out of 10 times or 9.5 out of 10 times, you're going to go with the 10. Now from 10 to 11 is only a 10% jump, but from 3 to 9 minutes is a 300x, more than a 300x jump. People don't know how to value that time. That's maybe what we would summarize it. But that's an example that also makes you realize that building these things is— and I'm guilty of that, by the way. So it's not like— This is how all of, all of us human beings are wired. So, you know, just, just another learning that I had from that space as well.

So those are my really great learnings. And also too, I think many of us here can relate with the example that you just shared as well. So, so another topic that I wanted to discuss in detail, and this is something that's been coming up more regularly in the community, is matching. And you, of course, have some really great experience at marketplaces where matching is core. And, you know, not only at Lyft, but probably where it's a little bit more complex and there's more considerations with that with Shipt. So could you maybe give us like a high-level overview and then we can get into a little bit more detail about, you know, the role that matching plays at marketplaces and then maybe help highlight some of the specific considerations from your past experience?

So the role of matching, I think some of it is very obvious. Some of it is very obvious, right? Like let's take a delivery marketplace, right? Let's, you know, think of Shipt or anybody else, right? Amazon, you know, whatever, whoever you want, right? Instacart, Uber, DoorDash, anybody that does deliveries. The first, I think there are some very well and obviously understood rules of matching. For example, if there was a customer that was in San Francisco requesting a delivery, it would make no sense for the matching system to match them with a gig worker in New York, right? I mean, it just makes no sense. So that's very basic. If a customer wants a delivery between 1 to 2 PM in a day, or because that's when they're around, they're waiting for lunch, it does not make sense to give them that delivery at dinner time, right? Those are obvious things that a matching system has to sort of stick with because without that, the system is clearly failing. But then I think that's the obvious, that's the table stakes. But then I think it starts getting more and more interesting. Which is now you have a diverse pool of customers and a very diverse pool of orders that are coming into the marketplace, different types of orders. Some of them are heavy, some of them are light, some of them pay more, some of them don't, right? Depending on what is inside the order. How do you create a system where you are incentivized? And then on the other side, on the gig worker side, it's equally similar, right? Like some of the— one of the examples, one of the simple examples, I'll take a sidebar here, is again, coming back to regulations. In New York, where I live, I believe there is a regulation where, think of a rideshare platform, that drivers are not allowed to work, I think, above a certain number of hours per day. Take that as a simple example, right? I think it's 12 or 16, something like that. I think 12 hours. What happens is after 12 hours, the app is going to basically block you out. The driver app for any rideshare company is going to block you out because that's the rules of the rules of the land. That's the law of the land, right, in New York. So now if you were matching, if there was a matching system, and let's say, you know, an A driver was on their hour 9.5 and they had 30 minutes to go, you cannot match a ride that takes 1 hour with this driver, for example, right? So it gets nuanced and complicated as you start getting into the heterogeneity of both the supply side and the demand side. The use cases are so diverse. And when you're trying to create a complex system, which is essentially most of these are all machine learning driven systems, but when you're trying to match across a diverse pool of 2 sides, for us it's more complex because there's 3 sides, there's a 3rd side involved, but that's where it becomes more and more complicated. Just to give you guys some flavor of what this is.

Yeah, that's helpful. Yeah, we could probably go super deep. We'll refrain from doing that as far as that ship, 'cause then there's routing and batching as well, right?

Yeah.

Related to this, and this is something that we've been discussing, is with the recent advancements with AI and LLMs, the capabilities. And so with marketplaces like early on being able to kind of match better and faster. So is this something you're seeing and like how do you think this will impact a marketplace's ability to match from the earlier stages?

Yeah, so I think let's take a step back and think about AI using the new AI capabilities for early stage marketplaces, right? Like matching is one, let's say matching is one area, but there could be so many. I think in the area of matching, just to put a pin on that one, one of the things that the newer technologies are doing really well, which I'm observing, is they're doing personalization really well, which is, for example, if you were to, as another example, another matching example, there are so many layers to this, but one more layer is I think now one of the rideshare companies, I believe it's Uber, have introduced a feature, or at least I saw or heard about it, I think a month, month and a half ago, where they introduced a feature, maybe they're testing it, where female riders can actually request female drivers. So that's a constraint in matching as well, right? Like for security reasons, if you wanted that. Now, those type of things where personalization comes into picture, if you, for example, preferred a certain type of music, If you preferred a certain type of language and so on and so forth. AI is very, very good at, for example, today's AI is very good at, for example, there you go. Molly is acknowledging that it's there. So AI is very good at, for example, maybe matching a driver to you who probably loves to listen to the music that you probably want. I mean, maybe the driver has not said that, but maybe AI is able to infer it. is what I'm trying to say. So, when it gets to retaining a user and giving them a superlative and a differentiated experience from everybody else, this is where I think the AI and the secret sauce actually comes into picture. Now, that's at scale, we're talking about a company which is at billions of dollars of market cap and you have millions of users and so on. So, that's one example. But for early-stage companies, for founders, What I would say, I love this line which I heard somewhere, the edge case for a big company, which may not be interesting to a big company, could actually be your business model. It could actually literally be your business model. It could be an edge case for somebody else. And the biggest example I saw for this was, I was very recently, sidebar, but I was very recently listening to an interview by Zuckerberg, Mark Zuckerberg, when he formed Facebook. And it was kind of like a duh moment for me, but it was also very inspiring and to understand how a founder of that caliber thinks. So this was his interview. So somebody asked him, when you guys created Facebook back in the day, how did that go? And his answer was that, I can bet you, Mark's answer was, I can bet you this idea was not new. There were so many social networks out there. There was Hi5, there was MySpace, obviously was the biggest one. There was this— Google had something called Orkut, if you guys remember. And there was Friendster and there were like 6 or 7 social networks out there. And the line that he used was so funny. He was like, so what we were trying to do, we were basically a group of children. Like we literally were children and we were trying to build this thing. And he was like, I can guarantee you that there was some VP of— like this is something I tell myself all the time. There was some VP of product somewhere in some company who was like, you know what, social network's not interesting. Here we are after whatever, 15 to 20 years of him trying to build Meta. So that's one example of somebody's edge case could actually be your business model and it could be a super hyper successful business model. So founders, especially founders that are built from a different— cut from a different cloth, you're obviously trying to add value into the world. But I would encourage people to think that way. The new technologies, The other thing I tell myself and my teams is generative AI is so new and it's so, well, it's starting to find its use cases, but still it's relatively so new that everybody who is using it is basically at the starting line. Nobody has like a significant advantage over the other. Therefore, if somebody could come up with a use case where you're able to generate value for your users or your customers, you are probably as good as equally good as, for example, a big player out there which is investing multiple billions of dollars. Google versus Meta is a great example where if children could do it, in his own words, I mean, this is Mark, but if children could do it, we should actually feel so inspired and we should feel like with this new technology, this new wave of technology, we should find ways to catch it is my only 2 cents for founders.

Yeah, no, that's a really good point. Yeah, that's something we're always kind of discussing here is founders finding these kind of new interesting use cases and ultimately kind kind of proving that out, which can then turn into a business there. So I guess if you were to jump over to a much later stage, so say for instance, like Shipt, how are you approaching AI? And maybe are there any kind of examples that you can share as how you've implemented it?

First, I will start with a disclaimer that this is new technology, right? It is emerging. It's really emerging. There was another interview, and I will cite the source for this. There is a blog called Well, Lenny's podcast, right? Lenny Rachetsky. So he blogs with people. So he had a conference very recently. I'll quickly cover the story. He had a conference in San Francisco where he had the Chief Product Officer of OpenAI and Chief Product Officer of Anthropic on stage. So Kevin Weil is CPO of OpenAI and forget the name of the Anthropic, the one at Anthropic. They're both there and somebody in the audience raises their hand and he says, Kevin, can you actually tell us 2 months before you release a foundational model, a foundation model of what you are releasing so we can build APIs, we can look into integrating? And Kevin half-jokingly raises his hand and he's like, can somebody tell me when we are releasing a foundational model? And the way he explained the process at OpenAI is they train a model. And this, I'm sure you guys have been hearing this a lot in the industry. They train a model. So the research team will actually train a model, make updates to core technology, et cetera. And then when the model is trained, let's say GPT-5, right? After 4, when GPT-5 was trained, then they ran a bunch of evals, right? And evals is the new sort of the new thing that everybody's now starting to— and what is an eval, right? Eval is, for example, can it crack a math test, right? Can it actually suggest a cooking recipe? Like how does it feel when you talk to it like it's your friend, right? Like there are all these different evals. There was the OG eval, the original eval was called the Turing test. Like back in the day when Alan Turing was like, can you— the original eval was, can you actually tell if you're talking to a human or a computer? If you're behind a curtain, could you actually do that? And LLMs have far, for a while now have broken the Turing test like on its face, right? Because they are so good at faking and mimicking a human being that you wouldn't know that you're not talking to a person. But anyway, I think the point is that it's so emerging that his whole point was we don't even know. So for example, with GPT-5, they found that it improved significantly on coding, but then it backslid on the connection with humans. And everybody here probably knows that that's why there was a backlash and people said, we want to go back and talk to GPT-4 because this feels like you're talking to a robot. So there are evals, long story short. With that caveat, what I would say is it's evolving. So we are also exploring use cases and trying to figure out just like everybody else, what is it good for? But generally there are a few areas. One of them is we obviously use it for, or we are exploring using it for internal research and data analytics, right? Like you point it to a database and say, okay, what was, can you tell me the value of a metric or can you tell me if something is off or whatever? And generative AI is good at creating stories around very specific questions that you're trying to ask, right? So it can plot and it can generate graphs and things like that. So one of the use cases is that. The other one is obviously you're trying to simplify the life of somebody by using agents, right? Like yesterday we would ask somebody to, for example, and just make this up, right? In a table, like you're setting, I don't know, let's say like you're uploading a bunch of zip codes. I'm just making this up, right? Like, like this zip codes, you know, you know, like you have stores in these zip codes. So now you sign— I'm completely making this up— but you signed a new partnership. There are all these stores from all this new partner in all these. So now you got to go and upload a CSV or got to upload an Excel file somewhere. Well, now you can just go and tell generative AI, like, you know, like, can you just upload all these? Can you just add this store to all these zip codes? You don't have to actually go and upload and finagle with files or anything like that. So that Internal productivity is another use case. Obviously, we are looking at optimizing some of our existing technology, things like customer service, right? Talking generative AI is very good at mimicking humans. And if you talk to a generative AI chatbot versus older chatbots, it's very difficult to distinguish whether you're talking to a human versus a chatbot. And that's where I think the technology shines. So, we are also thinking of and evaluating use cases of— we have a large customer service team. How do we use some of this technology there? And lastly, I would say more and most importantly, and this is where I was saying, where I'll ground us all, what I was saying before, I don't think anybody in the industry has really cracked this problem, is how to actually use it to impact the end users, right? Like, for us, it is the 3 sides, as I was— I'll come back to where we started. It's our customers, it's our partners, retail partners, retailers, as well as it is our gig workers. There are some obvious things. For example, a lot of our gig workers, for example, newer ones who are not experienced might come and ask, how do I start shopping at Shipt? What do I do? I see this icon on your app. I don't know what this means. Can you explain what this means? Do I have to do something differently? So those things, very— so in the older world, in the previous world, you would either Google it, Or we would basically point you to a webpage or something where you would have to kind of find that information yourself. Now with a chatbot, and we are exploring this, it's not like, you know, with a chatbot now, the thing is that instead of parsing through those, that's the original use case of LLMs, isn't it? Instead of parsing through a trove of documents, you ask a specific question and expect a very, very specific response. So, you know, it makes their life easier. But so, and there are, you know, use cases on the customer side again, like generating recipes, building baskets for the recipes, etc. So there are a bunch of these ideas that we are exploring. I don't think anybody in the industry has really cracked that secret sauce yet. So there's a lot of opportunity on the table, I would say, around these use cases.

Yeah, it's definitely early and lots of cool different ways that we're seeing being leveraged for marketplaces, especially operationally complex marketplaces. Yeah, for sure. Is there anything that we haven't covered that might be helpful for our marketplace founders and teams here that you like to discuss?

I would again say what I said before, be open to exploring new technologies. I always tell myself, Shift is, we are as big as we are, but as any company, we always have ambitions to grow and grow our business, make more impact on our customers, and give them a better service. And as I said, for all the other much bigger players out there, for example, their edge case might be your business model. Especially for founders, especially for new ideas, especially for the ones which have never been tried before. I would say don't be afraid to try these ideas, be willing to take risks. And the other thing, me being a product person, a product guy, I would always say always think backwards from the customer. If you are able to— last thing I'll say, right, like on the topic of ChatGPT itself, right? So, you know, it came on the scene about 3 years ago, 4 years ago. And it took the world by storm, right? Like everybody knows that. But if you go and look at Altman's, Sam Altman's, the CEO's stories, he was leading Y Combinator, he was a partner, he advised so many companies. And if you go and watch his X posts, one of his posts was so funny. He was like, everything that I've done at ChatGPT was exactly opposite of what I actually advised anyone. Don't use a weird name, right? Like who knows what is this GPT? Don't use a weird name, right? Like, have a user, like, have a UI, right? Like, have a, you know, do design, right? Like, you know, do like spend time on, you know, like market like this and that. Like everything, it was like literally everything I've told people, like build something with a clear use case in mind. There was no, they didn't know what they would do with the LLMs, but they went ahead and built this. And then, you know, like, you know, so it is the fastest growing company in the history of the world, right? So far. So like these things happen and, you know, as founders, you know, I know that the founders are very— that's where I think they are cut from a different cloth is that they have that keen eye to see things or see possibilities where nobody else sees them. So I would just say that lean on yourself, lean on your instincts, but definitely try and catch the wave. This is a new wave. It's not a flash in the pan. This feels like it is here to stay. There are new distribution channels that a lot of people talk about, how yesterday we had search engine optimization, tomorrow we might have I don't know, like chatbot optimization or something, right? Like, so, you know, keep up with those things and try and figure out where you can scale your startup and get eyeballs and get interest in the problems that you're solving from a large-scale audience in the world. So those are my parting words, if you will.

Those are gonna be really helpful. You got a few head nods with that. And I wanna make sure we can save time for the group Q&A, but I actually had a question that someone sent in advance, and I'll just jump on here and help kick things off by asking, Which is around testing pricing. And I know you actually led a whole workshop on experimentation, right? But when it comes to, you know, initially testing pricing, you know, I guess like when and how would you suggest early stage marketplaces test pricing?

The simplest example I'll use is marketplaces like Amazon, Walmart, and others, right? They have their own subscriptions, right? For example, you know, Prime, Walmart Plus, whatever, but they also have product pricing on top. So I'm assuming that we're talking about product pricing. There is a— and this is another thing. There is a— the general philosophy where people generally start is, you price with what you call as the cost of goods sold model, which is, this is how much it costs me to make something, right? Like whether it is a physical thing, whether it's a digital thing, this is how much it costs me to make a unit of this. And then you add a margin on top and then you basically, you know, like you do that. There is a different model, which is what is people's willingness to pay, which is exactly opposite of— so one thing is what you think the right price is and you're pricing inside out, which is based on what is the cost of your goods sold or cost of your goods manufacturing and you're pricing on top of that. The other one is, let's say even if you did that, nobody's willing to buy because they thought you were too expensive as an example. Or The other side also happens sometimes where you price there and people might feel it's really underpriced, therefore it might be cheap. Like there are case studies where if you price too low, people will probably walk away from the product because they think the product doesn't have good quality, for example, right? There are studies where for the same product, sometimes when the pricing was significantly improved or significantly increased, it actually improved sales. So there are situations like that. So what I would say is, I would say is experiment with it right from the start. What is the right time to experiment pricing? I think the— so here's the thing, here's the thing, right? So, you know, if you— there are 2 schools of thought, right? If you've tried to do it too early, you might think that I don't have enough data, so am I doing it right or wrong? But also because you don't have enough data, you're not going to get— you're not going to probably get a lot of like negative feedback if you change. But pricing is one of those things where you will get a lot of feedback. So if you have a smaller audience, it's not always a disadvantage. Like if you had a large audience and you were trying to change prices on them, that could be a major disadvantage as well. So my point is that think of it as a superpower, in fact, that you have smaller number of users where you can probably test prices on small cohorts, et cetera, and then basically price up from there. But yeah, but then generally speaking, like it's cost of goods sold. Don't be scared to be aggressive. Like, don't be scared to be, you know, to try to infer like what a user's willingness to pay will be for the product that you're making. And I am a proponent of working backwards from a user's willingness to pay model. So I would say, you know, try to get to that as fast as you can versus always pricing based on what you think is the cost of goods sold. So that's like a high-level answer if you will.

Yeah, that's gonna be really helpful. I guess related to that, are there maybe any kind of common mistakes that you see marketplaces make around that or?

So I was actually, I was gonna bring up that story as well. There was a conference called Marketplace Risk, which happens in San Francisco. I keynoted that conference when I was in San Francisco. I keynoted that about 4 or 5 years ago. And after, and I was, and it was a talk about pricing and dynamic pricing, which was my experience at Lyft, you know, the pricing systems I'd worked at. Somebody after, right after the talk came to me and said, This was, by the way, this was in 2021, I think. So right when we were getting out of COVID and somebody from the audience comes and says, I was very inspired by your talk about not being shy. Some of the things I just said, try and figure out what is users' willingness to pay. And their statement was, we had a startup during COVID where the demand was like through the roof, right? Like everything was like upside down essentially. And they're like, you know, we had it in COVID where we had to shut down the startup. And the reason was, you know, now when I'm listening to your list, and this was, I think they were trying to sell, I think some merchandise or something like that, if I remember correctly, right? Something in supply chain, but they had a storefront and they were trying to, this was 5 years ago, but they were trying to sell some merchandise. And their response was, Now that I'm listening to your talk, I'm going back and I'm kicking myself and I'm saying that we should not have been afraid to raise prices. We were so scared. They were a startup, right? This is the point I'm trying to make. You don't have that many users. You can only fail so much as a startup, right? Like, unlike a big company, you could actually drop billions of dollars in value if you get it wrong. But as a startup, you can only fail so much, right? And your customers know you and they are with you because they love you and they find value and so on and so forth. But anyway, this person's response was that we kept being defensive and defensive and defensive, and there was just a hard rule in the company. And this person was the founder, a co-founder. And it's like, we made a hard rule in the company that we would just not raise prices above a certain margin. And then finally, now listening to your talk and you talking about ridesharing companies and things like that where these type of models are used, now we're starting to realize that— now I'm starting to realize that we should have been a little bit more, we should have had some more appetite for risk. We were so risk-averse that eventually we basically burned our company to the ground and we had to shut it down. So that is my story for you that don't be that, right? Like it can only go so wrong at the end of the day. So don't be afraid is what I would say.

Cool. So I actually had a question that Noah sent and he said that he's building a 3-sided marketplace. So when it comes to kind of going from 0 to 1 and getting started with a 3-sided marketplace, Do you have any maybe specific kind of tips or suggestions?

What kind of a marketplace are you building? Can you, like, what is it exactly?

So we have a marketplace between hotels and creators. So a lot of hotels, they want to do VR events and Pilates events. We've done a bit of market research on companies that want to sponsor these. So they want to sponsor these events. Our thing is that we've done, we've built a marketplace between hotels and creators and that works. Should we manually try it first? What are things that we should be aware of before we actually build a line of code?

So, I would say that this is a very ironic statement and I make this a lot actually. Coming from technology and being a technologist, I always ask exactly the question you asked. Do we need to build something to actually try an idea? It's very expensive to build something. It's very, very expensive to build product, to build software, to motivate your team to do it, especially because if it fails and over and over and over again as a startup. It's just like you only have so much emotional battery before which you run out and you start to feel like— but that's the life of a startup. You will have many, many, many, many failures. That's just the life of a startup and you just have to keep trying and trying and trying, pulling yourself back up and going. So, I would say number one, the answer is even forget a startup, but even in my teams, when we try a 0 to 1 idea, I'm just like, do we have to like go to engineering? Do we need engineering? Like why? Like can we actually validate that somebody cares? At the end of the day, you're trying to find out if users care, right? Like whether they will come, right? So that's one thing. So yes, absolutely. I would say try and figure out a way around your system to validate the idea there's a there there versus upgrading or changing your system to validate it, right? So that's one thing I would say. Because upgrading or changing is expensive. Even if it's not that many changes, at the very least, you have to now undo, probably undo, because it's dead code, it's not being used, it's like all of that. So there is some probably effort to take 2 steps back when you took those 2 steps forward in the past, right? So there's all of that. So I would say try and do it separately. One of the other things that happens as a result of this is I am now starting to see different companies. We are not doing it, but different companies, they actually have like test apps. So they have like their main app, which is like their main thing. But then they also have this like side app, right? Where they're just like, you know, like this is my like labs, you know, where I'm going to like go and see like anything is allowed and what can I do? And many companies are doing it, by the way. Like it's not very obvious because they don't want to like associate the main brand with the labs thing. But they are, they have a labs thing because they are trying to experiment and figure out. It's very, Easy to do. Like, you know, there's no brand damage or anything like that. You walk away from it, right? You just like pull it and walk away from it. So that's number one. Number 2, I would say for attracting and making this a part of your business, and I don't know the specifics.

No.

If you want to talk separate, you know, we can have a separate conversation. But if you want to make it a part of your business, you have to tell the events. I'm assuming you're going to events companies. You have to tell the events companies what is in it for them.

Right?

Like, for example, you can bring an audience to them which they would not have had and maybe find a way to cross-promote people. Like, there is demand here. You should come to us because we have demand and we have venues for which, for example, for you, like, it took you like so many hours. Yes, you are an event company. You know how to like program, you know how to get the programming ready, you know the media, you understand AV, you understand all of that. You understand narratives and content and whatever else, right? But you need a venue, you need, and these are 2 most expensive parts of your, for example, of your operation. And we can do bulk deals with you. Like we can actually give you bulk discounts if you do it at a chain of restaurants or we can bring an audience to you. So always like try and tell them what is in it for them. This is just standard building a business 101, right? So those would be my 2 pieces of feedback if you will.

Awesome, thanks. Hey, Molly, saw you raise your hand. Do you wanna jump on?

Yeah, so I run Recess, marketplace for afterschool activities, classes, and camps. And we are hiring a product manager. So things are just very wild in the product and eng space. And I just had a question if you have your favorite places to source candidates. This would be more of like a, I would say junior, kind of like early to mid-career role. And it seems like LinkedIn is not necessarily the best spot.

It's not the worst either. Because it is the largest. I mean, I don't know, OpenAI is saying that they're building a recruiting thing. I don't know, there's talk somewhere. They talk about everything nowadays. They're also building a consulting thing where they're trying to go after the McKinseys and the Bains, but they also are saying that they're building a recruiting product to compete with LinkedIn. And so I don't know about that, but it's not the worst. One of the ways for me, what I will say is because it is, I would say, the most authoritative for me, I would say, or the largest, maybe not the most authoritative, but the largest repository of professionals is LinkedIn, right? It is a good place for me. I would say that for whoever we have hired, most of them have approached me on LinkedIn, my teams. Now my teams are hiring. So they have approached them on LinkedIn. It's not the worst source of resumes. What I will say, Molly, hiring is one of the most difficult things in a company. It's one of the most, it's the most tedious. It doesn't get easier if you are higher up. It doesn't get, if you're a founder, not a founder. And unfortunately in this day and age when there's so much of a talent pool and slash, people don't even know if they want to hire people at a junior level because AI is basically just making those jobs redundant. So there's more and more and more people, more and more graduates without a job. It's even more like it's a deluge of resumes. There is no easy way to do this. You have to just like, I mean, you are a founder, so I'm preaching to the choir here, but just like roll up your sleeves and just white-knuckle this thing. That is the only way to do it. We did it, like I did it when I built my team at Shipt 4 years ago. As an example, and I say this sometimes, I took 9 months to hire my first PM. And this was at Shift. And I was supposed to hire 3 and run 3 PMs for 3 teams. And I actually took 9 months and I was running all the teams myself. And I took 9 months to have my first PM because I just didn't find the right person. And there were so many interviews I had to go through, but it's one of the most difficult things that you'll actually have to do. So if you need any other tips, I'm happy to offline on that, but there's no easy answers unfortunately so far. That's what I'd say.

Awesome. And then another related topic since to AI, of course, which has come up in the community, I wanted to ask for someone, which is around balancing like speed when launching new features and the product roadmap. So since it's, you know, become much easier and faster to kind of, you know, build and launch new features and a lot of times, you know, at the early stages, like going from 0 to 1 and kind of building liquidity, I guess, like, do you have any tips maybe for founders on like, you know, how to think about like speed of launching new features and balancing the product roadmap?

So I would say that at scale, right, at a company like Shivk, for example, I generally follow a 70-20-10 rule. What that means is 70% of our projects are focused on improving things, making things that work basically, and you're trying to make them better. 20% of the time is spent on things where we have very, very high confidence but we haven't built them yet. And 10% is we just don't know what we're doing, like let's just go and try, for example. Like completely new ideas where we don't know if there's a there there, could be completely zero to one, could be somebody just came up and some leader said let's go and explore this or somebody came and said let's try and explore this. So that's at scale. For startups I would say like you can probably like adjust this equation accordingly, which is your 70% could be ideas where you do have high confidence, but then 30% could be things that you're spending, assuming that you don't have a high product market, you don't have like a business case where you're generating a lot of revenue right now and you're still trying to find it. It depends on what stage you are, obviously. But if you're still trying to find it, I would say then 70-30 sounds like a good split where 70% of the time you are you have ideas, that is why you founded the company and you're trying to take those on the road and see what works. But 30% of the time you're purely exploring, for example, using things like generative AI and all that. And 70-30, I also, when I use that framework, I use that and I mean to use that framework in terms of the time you spend as a company, you yourself as a founder, in terms of the resources you allocate, right? 70% of your entire company, eng teams, design, whatever is focused on things that are very high confidence, 30% you don't know and you're exploring. So, take your pick. But I would say that at a point where you know that there is a revenue-generating idea or there is a revenue-generating stream, as a startup, the first thing you want to do is sustain yourself, right? Like through funding and— sorry, through your own revenue because your funding is going to at some point run out, right? So, for that, I would say as soon as you find that, definitely make sure that you are tripling down on that. And the simplest example I will use is a large portion of revenue for even the biggest of the companies, right? Your Googles, your Metas, and all of that, it comes from digital advertising even today. Like 90-95% of the revenue comes from it and they spend an inordinate amount of time and money on it as they should. Because it's generating disproportional return on investment. And then, you know, same idea, right? Like they are now using the returns on that investment to invest everywhere else, etc. So it's not very different, but I would just say use the 70-30-10 framework and adapt it to whatever you want to do.

Awesome.

That's gonna be a helpful framework. And I know one of the challenges when discussing marketplaces is so specific and kind of nuanced once you really get into it as well.

Yeah, that's true. That's true.

Hey, Lucius, you wanna come on?

Hey, Vishal. My name is Lucius, founder of Bytesight. We're a video-first food delivery app. Thanks so much for making the time. I really appreciated it. One of the things that I'm kind of curious about from your story, your time at Shift, is like how much of the value that you guys drive for your customers are you guys thinking about from like the perspective of like differentiation and like answering the question of like, how are we different from Instacart or from any of the other places where you can get the same products delivered in about the same timeframe? versus just delivering a great service? Because I know as someone who's fundraising, like, I think I get a lot of those questions, which is like, how are you different and what are these things that you're doing different? But to me, I also think there's a lot of value in like, are we just doing something great and are there a lot of people who value it? And like for you guys, it seems like a big part of the business is like we're doing something great and a lot of people are willing to pay for it.

So. It's a great question actually. And I would say that, I would say that, you know, I'd use another story and this is slightly sidebar, but I'll use it anyway. Bezos used this example and he has used it and I saw him somewhere on YouTube, I guess, but I saw him say this, right? So he's like, people ask me what will change in the next 5 years, right? Like what should you be, what should Amazon be going after? What will change? What they don't ask me is what will never change. We like to go after things that will never change. For example, we want to go after being like a low-cost retailer. They were a low-margin leader for a long time, loss leader or low-margin leader. We want to be one of the lowest margin retailers that you can find. We want to have the biggest catalog, the biggest selection that you can find. We want to deliver things as fast as possible that nobody else can do, faster and faster and faster and faster. Because we know that people want those 3 fundamental things no matter what. I mean, the thing is, here's the thing, Lucius, they built an entire fleet of Prime airplanes, scooters, like you name it, right? From scooters all the way to airplanes. There are trucks, obviously, like everything in the middle, right? Like from last mile to first mile to whatever miles you want to put on it. They basically built the entire fulfillment infrastructure. Why? Because they're trying to deliver packages to the customers faster. And then there's warehouses and this and that, like the whole thing, right? But ultimately they're trying to deliver things to customers faster. They're trying to do it cheaper, et cetera, et cetera. So I would come back to, I would answer your question in a very similar way. This question comes up a lot to us. Are you trying to differentiate with something or what? And my answer is very simple. You're just trying to serve the customer. You're just trying to serve the customer. If you do an outstanding job, the customers will tell you, by the way, right? They will tell you if they expect, for example, if they, you know, some way, maybe they will, maybe they won't, maybe you have to read between the lines of what they really want. But, you know, like if people are churning away from your product, you know, you will take a look, for example, at, you know, what's out there and try to assess like, what do you need to do in order to be able to, you know, quote unquote differentiate? But all of that, whatever you do at the end of the day, you know, ultimately is in the service of the customer and doing the job that needs to be done for them and basically making their lives easier. If you are the number one metric that I tell my PMs to measure anytime they launch anything or think about anything is adoption. Do people care? If people don't care, it doesn't matter. Do people care? Are people willing to use it? Well, there is a small asterisk there, which is If you're giving something for free and people care, like when you actually put a reasonable price tag and people stop caring. So, remove that, which is, you're basically making them care, artificially making them care. But under normal circumstances, if you release a product where you're not giving iPads for free, right? If you're giving iPads for free, people care, right? But take that edge case away. But generally speaking, when you release a feature or product, do people care? The first thing is if people care, And this is another thing I also like to say, where there is demand, supply will follow. It will follow, right? So focus on demand, focus on the fact that you are the one, number one, you build product that people care about. Like that's your differentiator is that you build something that people care the most about and therefore they remember you as a company that can do something that you can do, which nobody else does like the way you can do it, right? So make people care about you. Like that's the only thing I would say. Everything else is just like, it's just an outcome from that. So yeah.

This is probably a good way to kind of wrap things up here because I know we're almost out of time. But do you have maybe a few kind of learnings on the competitive nature of marketplaces that you could share with us?

Yeah, I mean, I think we covered some of this, right? Like it's more difficult to do it at the beginning, at the top of the conversation. It's definitely more, if I compare Lyft with Shift, my 2 recent employers, it is definitely more difficult to build it run a 3-sided marketplace. I think one of the things that makes it challenging is— so very quickly, I'll give you why. So at Lyft, what happens on a rideshare marketplace generally, what happens is you have a unit of demand, you have like a request coming online. It's an on-demand marketplace, but you either in that moment, you can either serve it or you don't, right? So either you can actually match it with somebody, like we talked about matching systems, If you can't find a match, you're like out of luck, right? So you can't do it. In the delivery world, it's a little bit different because the demand actually comes, typically comes a little bit ahead of time. So it's more like an e-commerce/retail marketplace where you're placing a demand, but when you're serving the demand, that serving needs to happen in real time or near real time. Meaning like as a customer, if you, in the morning, like let's say at 8:00 AM, You're trying to plan for your dinner. You place an order at 8:00 AM and say, I want my dinner, I want my delivery to happen between 4:00 and 5:00 PM so I can start prepping my dinner at 5:00 to 6:00. And then at 6:00 PM, my dinner is ready. So you want your deliveries between 4:00 and 5:00. Now at 8:00 to 9:00 PM in the morning, we have to actually forecast whatever you're ordering for the 4:00 to 5:00 PM. Will everything, is everything that you're ordering be available on shelf in that store wherever you're ordering from. And this is, we are talking about thousands and tens of thousands of stores forecasting across every store, across every product across, right? And getting it right because one of the things that we also hear a lot and experience a lot is I was trying to make chicken pasta or chicken Parmesan or whatever. And I ordered chicken and the cheese and a couple of other things. And you didn't deliver my chicken. I'm like, now what do I do? But you have to, at 8 to 9 in the morning, you have to actually be able to forecast very accurately that you will be able to serve it. Otherwise, you upset the customer. And if the customer has to go to the store, you've kind of lost that. That was a loss. So it becomes more challenging when it comes to these because the demand comes ahead of time and the serving of the demand needs to happen like in the future somewhere. So you're sort of relying on a forecasting system to be able to do something like this.

So yeah. That's a great way to wrap things up here. And it also helps kind of, you know, demonstrate some of the additional challenges, you know, with 3-sided marketplaces here. And at the same time kind of ties it in what we were mentioning earlier, you know, solving a consumer problem. It might seem so easy, you know, from the consumer side, but, you know, everything that's involved kind of behind the scenes with all the kind of complex coordinations and systems. So, well, you know, really appreciate you taking the time to join us today for the fall follow-up group chat. And now we kind of bounce around a little bit, but I covered lots of ground here. So, you know, we appreciate, you know, joining us. And thanks for also too helping answer some of the questions that we had here.

I hope that the 200th episode was worth it. You know, yeah, I hope that, you know, people were able to get something out of this talk. And if anybody needs to, you know, double-click on anything, I would love to help out. Okay, thank you so much.

Yeah, it was definitely worth it. So, you know, awesome 200th episode. And I'll also be sure to include a link to your previous workshop on experimentation. I know everyone will Have fun kind of going through that as well. So thanks again for joining us here, and you know all you do to pay it forward. And thanks everyone for the great questions.