Tony Xu of DoorDash: Surviving 1,000 Days of Startup Hell

Well, whenever you can ship something in 43 minutes to test your idea, I think that's pretty good. And certainly this is, you know, 12, 13 years before the rise of LLMs and AI tools to make it so easy to do that. But basically, the four of us wanted to test this idea that if you wanted to offer delivery from places that never offered delivery before, what is the fastest way to see whether or not consumers would care? I mean, at the end of the day, delivery is not a new idea. And so, we thought actually one of the reasons why maybe delivery in 2013 hadn't been around yet was just because nobody wanted it. So, we shipped PaloAltoDelivery.com. That alias was available for $9. And so, that's why we got it. Not a super scalable URL, but we were able to get it. It was a static page where you saw eight PDF menus of restaurants that we frequented in Palo Alto. And the only way you can order is you can read through the menus, you can call a Google Voice number that would ring the cell phones of the four founders and one of us would pick up. We would take your order, place the order on your behalf, go and get the order, deliver it to you. And I used to be at Square and so I had these card readers which was one of their earliest products. These white dongles that you could stick into the audio jacks of iPhones and that's how we would collect payment.

No, actually, yeah. I think one of the biggest misconceptions when we were founded was just how wide open the space was. Where there were about a million restaurants in the states and maybe 20 to 25,000 of them offered deliveries. Most of them were pizza shops, places in New York City, some in Chicago, some in big city centers. But outside of pizza places, maybe a few Chinese restaurants, nobody offered delivery. And so the real grand question or experiment of DoorDash, PaloAltoDelivery.com was, okay, what about everyone else? What if you can enable everyone to actually offer delivery? What would that take? And first of all, would people care? And that's really why we shipped something so quickly, just to see if people would actually come and place orders.

They were mostly, honestly, faxing orders, believe it or not. So they would be a website that would receive orders, if you can believe it. They would fax the orders literally into machines that would sit near the kitchen or the payment systems inside these restaurants. Then the restaurants would actually go out and do the deliveries themselves. So they were lead gen companies at the time.

No, we did. So, when we started, I guess to take a step back before we shipped PaloAltoDelivery.com or even how we got there, you know, my co-founders and I really got connected because of an interest in small businesses. You know, I think my story I've told publicly, which is really, I mean, I grew up coming to the States as an immigrant from China and my mom, you know, put food on the table by working three jobs a day for 12 years. One of those jobs happened to be at a Chinese restaurant where she was a waitress. I got to hang out with her, wash a few dishes when she allowed me to. That's kind of how I grew up while my dad was getting his PhD at the University of Illinois. That was, you know, the first 10 years or so of childhood growing up in the states. And that moment and experience always just gave me a deep appreciation for what small business owners represent. I mean, to them, there's no such thing as work. Work-life, it's all the same thing. There's no concept of a weekend or a Saturday. Saturdays and Tuesdays are exactly the same days. And you just kind of get into this process where that becomes your identity. And it's actually one of the most fascinating things I find about the great experiment that's America where, you know, because it becomes this all-consuming thing. One of the nice positive derivatives is actually they don't just create great experiences like a restaurant or a bar or a furniture store or a t-shirt shop. They actually create the GDP for all the cities that we live in. That GDP is what allows us to have great neighborhoods, schools, all the positive things that happen from a local community. And that was always my fascination with it. We had no idea though when we're looking at starting DoorDash about anything related to what these business owners' problems were. And so my co-founders and I, we spoke with 300 maybe businesses up and down the Bay Area from San Jose to San Francisco, restaurants, retailers, service businesses, and it was actually a baker who showed us a booklet, a 3-inch binder of delivery orders she had turned down. She was a one-person shop who had no ability to fulfill or desire, frankly, to fulfill all those orders. And that was just a very strange moment for us where I said delivery is not a new idea. It's 2013. No one offers delivery. Why? And that's really what prompted us to think about, you know, launching PaloAltoDelivery.com to see if people cared. But to your question on logistics networks, you know, we said, okay, well, if the first place in which we can help local businesses is by building a logistics network, we have to pick a place to start. And this is where I guess the math brain, you know, comes in for me where when we studied every category of local retail of where we would start, whether it was deliveries for restaurants, grocery stores, convenience stores, retail shops...

We looked at all of them. And we had this hypothesis that if you wanted a chance of creating a logistics network that could actually be successful, that can be very fast, that can be very flexible, meaning it can deliver in 30 minutes or it can deliver longer than that, you needed network density. You needed the most number of connections between consumers and stores. We kind of targeted restaurants because there were a million restaurants. You know, if you compare that to the number of grocery stores, there was maybe a couple hundred thousand grocery stores. And you looked at other categories of retail, restaurants had the highest count of stores. And so very quickly, you know, we made the assumption that if there's any vertical to get started in doing deliveries it would be restaurants and prepared meals to give us a chance to build the highest density network so that one day we can deliver everything else.

Yeah. Well, starting in Palo Alto was, I mean, not a conscious choice. It was just where we were students at the time. But one of the earliest experiments we ran at DoorDash was doing deliveries in Palo Alto versus doing deliveries in San Francisco. So a city center if you will. That was close to where we started the company. And one of the fascinating things we found out, and we didn't understand why initially, was we were actually completing deliveries faster inside Palo Alto than we were inside San Francisco. Obviously San Francisco is a more dense place. But one of the things we learned early on though was that obviously, you know, in Palo Alto, you had much easier parking. You had a lot fewer apartment complexes where you had to go up and down the stairs and figure out where the lobby was or the right elevator entrance, things like that. Palo Alto had the following which is if you looked at places like Palo Alto it really represents I think most cities in the US or a lot of the world where you have main streets and then you kind of have, you know, in the spokes outside of this main street hub of commerce you have where the people live. And if you actually thought intelligently about what that really told you, you can actually build a very efficient logistics system if you just understood how to manipulate some of these hubs and spokes. And so this was one of the earliest hypotheses we had that you can actually make a logistics business as efficient in a place like Palo Alto versus San Francisco that was guided by that experiment. But the second thing was actually just in talking to customers. What customers told us was they said, look, in San Francisco, I can just walk down the elevator and head out the lobby and we could probably find a few places to go and eat. In Palo Alto, you'd be walking for miles before you could achieve something like that. You know, the closest set of restaurants near Stanford University where we started this was 2 miles away on University Avenue, as an example. And that's true in a lot of places in America. And so if there was any place we thought where there would be the highest interest from consumers and a possibility where you can actually make the math work, it was places like Palo Alto. And the question to us was just how many of them are there.

Yeah. In the very beginning, it was just the four founders.

Yeah. I mean, we saw, I mean, one of the benefits when you do the deliveries is well one you see how hard it is to actually bring you a burrito on time every time correctly. And the second thing is you get to see who the customer is. And you saw the customer actually almost always was a mom, you know, who had young children, who had not a lot of time, who didn't want to cook every single meal, who wanted just looked for any solution to save her time. And so when we did those deliveries, we just saw, wow, well, there are a lot of young families out there, and let's go find out where they hang out. Let's go find out where they live. And that's why we had that sense that we can build a business with this audience to start.

Yeah. I mean, I think that was probably a derivative of the discovery, but no, I think in the beginning, especially when you're looking for product market fit as an entrepreneur, you're looking for someone who actually just wants your product organically. And we could tell very quickly that someone who has young children who maybe doesn't want to take a stroller, pack it up, pack all the things that come with the stroller, then put that stroller and the children into the vehicle, then get it out and then somehow get inside of a crowded parking lot or a restaurant. Well, there are a lot of those people. And if we can solve it for that group, then we believe we could build a business that can easily grow organically. You're right. I mean, there's a second derivative, which is there more mouths to feed when you have a family than when you have one or two people living inside of a city. But that wasn't the first thought we had.

Yeah, totally. And the presence of single family homes made it a lot easier for sure. That was one of the benefits of delivering to places like Palo Alto. But again I think it just came from this very simple experiment which had an anomalous finding which is why is it faster to deliver in Palo Alto than it is in San Francisco? Why is it faster to deliver in a less dense place in other words?

We chased where the... I think when you're starting out the number one thing every entrepreneur is looking for is do you have something that someone else wants and is it real? Meaning like it's not artificially inflated with discounts and marketing dollars and just other ways to inorganically grow. Will people actually use it? Will they actually tell their friends about it if they actually like the service? And that's what we found early on with places like Palo Alto.

Especially when we were called that, right. Yeah, we had no money. Exactly. We ran this out of my bank account and that's why I knew early on even though look we didn't have any models or unit economic forecast or anything like this but even though I was running out of my bank account where I also had student debt at the time my bank account wasn't going down every single week or every single month so something was telling me that maybe this has a chance of working.

Exactly. You're not paying yourself, free labor, right? Just your time. You built a $9 website. Yeah.

We used Find My Friends. We used Find My Friends to track the drivers, which just happened to be all of you, our co-founders.

No. No. No, we had no money to market or advertise.

It was all kind of like self-funded. This entire activity was self-funded until we had to start recruiting drivers and actually testing this out beyond just the four of us.

Yeah. Yeah. I mean, in that time period.

We may have had one or two.

Yeah. Very quickly we realized, well, we're in class and so, you know, we took turns doing deliveries while we were in class, but at some point it's tough to be a student and do the deliveries.

We had maybe 6 months left before graduation. I mean, we were effectively Stanford's delivery service, you know, for the first half of 2013. We were effectively Stanford's delivery service. Then we get DoorDash the URL and the company name and then we launch out of Y Combinator in the summer, June 2013.

Yeah, with a very small group of users because in the beginning we actually did not have high volume. I mean it was probably 10 orders a day, something like that. And maybe our high day was like 21 orders a day, something like this. Most of them however were done by a small group of users at Stanford. When you see that fact that the same customers are ordering over and again even though it wasn't growing like wildfire but our bank account also wasn't getting depleted, it gave us enough conviction to keep going.

I think we viewed it as a project more than we viewed it as a company. In fact, we were barely incorporated. We were not incorporated when we were running this at Stanford University and then we just got incorporated when we actually got into YC. But at the time it was just like let's just see what the next phase should be. I think sometimes when you start these projects, you absolutely should have a point of view on maybe where this can go in terms of going the distance, but the most important thing is to just get started and then to have a sense of what the next two or three steps are. No one is able to know everything about the future. And for us, the summer was really instructive. I mean, the summer, I think doing the deliveries ourselves for the first six months gave us the clarity that the summer was really about answering three questions: would consumers want to pay us six bucks, which is what we charged. Are there restaurants who would be willing to partner with us for 15%. And, you know, could we afford a wage that we could pay dashers, the drivers, for the service? That was it. That was the entirety of the YC summer. It was not about demo day or raising the most amount of money or becoming the most popular at some event. It was just answering those three questions. And if we had enough conviction answering those questions, then we'd keep going.

Yes. Yes. That was... yeah. Look, I mean, I think we had a lot of classmates at Stanford who looked at us and just thought, boy, like I thought they were smart, but I guess they want to spend their time doing this. And so look, in the beginning of a lot of these entrepreneurial ventures, nothing looks that amazing, right? We were working out of an apartment. We had dashers in that apartment. We had the co-founders live in that apartment. We worked 10:00 a.m. to 2 a.m. every single day, but it wasn't like this glamorous exercise. But nor did we seek that, you know, we were just trying to answer those three questions that summer. We didn't care that much about what our friends were doing. Clearly, we thought it was interesting enough to keep going that if we can actually answer these questions, I think we're actually on to something.

No. No. It's actually why we had to do the deliveries. I mean, the reason why we were so hellbent on doing the deliveries besides the fact that we had no idea whether we had any business recruiting other drivers was how does this work? How should it work? I think DoorDash early on, even to this day but early on, was so hard to explain because it was actually even to build the MVP, yes, to test it was just this website, palmut.com, but we had to build like four things. We had to build this website for consumers. We had to build some app for the restaurants to actually receive the orders. We had to build an app for the drivers, the dashers. And then we had to build a dispatch system that actually could oversee all of the operations. So even in the very beginning, we realized, wow, this is actually pretty interesting. It's just such a fun problem that you in order to actually just bring you a burrito, you have to build these four things. And then to do it really, really well, I mean, that's why we did all the deliveries to figure out how you actually do that.

Absolutely. Because I think most people, and I totally get it. I mean, think of DoorDash as a consumer app. You know, most people think of us as lunch and dinner. And I think what they don't see is everything behind the scenes. I think a lot of times I think you can look at products like ours, especially as a consumer, and you say, wow, this looks like any other product. You know, there are so many of them. But then I would ask the question, well, how come one just gets used more often than the next or the others? And it comes down to everything that you can't see. You know, one of the things we say a lot internally at the company at DoorDash is it's always the data that you can't see that kills you. Because if you can see a truck coming at you, you're just going to dodge and get out the way, but if you can't see it, you're dead. And it's no different with our business. Our business is one where all of the magic or the secret sauce, if you will, are in things that you cannot see. You know, no consumer is sitting there while they're ordering DoorDash thinking about what the Dasher experience should look like or what the operations should be to get the best quality experience at the most affordable price or what are the ways in which you take out every single friction and cost with a restaurant or a retailer and make sure that all the items are actually there even when they're not there. You know, I think all of these things are the things that make DoorDash special and make DoorDash an end-to-end experience that's very difficult to replicate. But yeah, I think early on we knew that because we did all the deliveries. You know who knows it? Your competitors.

Exactly.

Okay.

Yeah.

Correct. Yeah, he runs all of our European business.

This again starts from actually doing the work ourselves and realizing that if you actually want to get something on time, I think it's very easy to think about when you're just intellectualizing it on the outside when we're getting started. Oh, maybe there's a traffic issue or maybe the food is taking longer than it should. Whatever the reasons might be, but you actually have no idea actually what are all the sources of delay in an order until you actually go and do the work. Sure, there might be some of the issues that I think you can think about on the outside, but then very quickly you realize that there's a lot of seconds of delay in every motion. In fact, there's about 20 steps you can decompose a delivery into. And there's delays at each one of those moments. And that's even more complicated if the delivery today happens outside of restaurants, they happen inside shopping contexts like groceries or retail items or if they happen inside malls that are multi-story, sometimes below ground, sometimes above ground. And one of the things you start realizing is, wow, actually there are a lot of causes for delays. And there's no way that you're going to know about all of them until you literally actually encounter it for the first time. A lot of what's difficult about DoorDash is we're trying to build a structured data set in a world that is chaos. That's the physical world. One of the reasons why there's all these sources for mistakes, for delays, for costs that ultimately yield into costs and good or bad experiences for customers is because there is no data that exists. There is no nice data set that a company like a Google or somebody else has organized for you. Because it's all physical information and it's also changing all the time. When you go into a grocery store and somebody moves an apple from aisle 6 to aisle 8, is that always going to get documented? Of course not. Those are the kinds of things we have to work on every single day. And you wouldn't know that. You know what? If I told you the cause for a delay was because actually somebody was homesick that day. How would you know that actually until that event actually transpired? And what would you do to respond to that event if that were to occur, which happens every single day, you know, when we're doing millions of orders every single day. The one in a million event happens a lot and the one in a thousand event happens way more than that. And so building a system that can ideally detect and prevent these issues, but then also a very fast twitch muscle to actually be able to build this, I mean, almost like an emergency response system when something actually goes awry to fix it. That requires doing the work over and again and building the system that can learn over time to get better and better and better. Most of the time we have no idea. We start with these experiments and that's why most experiments fail. But when you get enough goodness out of it, if you can get the 5% out of tens of thousands of experiments to work in one year, that has the benefit on all of your audience for the next year. And then you just keep going and that adds compounding surplus for all of the audiences.

Ideally. Yeah. Yes. I think when we are at our best that's what's happening. But it starts with actually building a system that actually wants to learn. If you think about like why do we have to learn? It's because the physical world is not structured. It's not documented anywhere. You can't scrape it. It's constantly changing. I mean, there's a winter storm right now, for example, in the Northeast. I mean, these are all things that happen differently.

Yeah, I know. We would have no idea here. We're spoiled here in the Bay Area. But like, but in general, all these things are happening every single hour of the day. Okay. There's going to be some missing item today. There's going to be some order that took a lot longer today. There's going to be some incorrect gate we entered at an apartment complex. There's going to be some dasher who's going to get lost coming up the stairs of this office building. There will be guaranteed. And so the question is like well it would be impossible to try to figure out all of that if you can't build a system to learn how to do this. So the most important thing is actually building systems and building a system that at DoorDash really starts with testing things in a very operational, hacky, do things that don't scale kind of way into then taking the things that ultimately work, the ideas, and actually building products around them and then engineering the ones that actually work so that you're actually very efficient with this learning loop so that you can go from learning to shipping something that actually works. Because you know it's a resource constraint with how many engineers we have and how many things that we can actually ship especially when the stakes are high and you want to make that loop as tight and as fast as possible. That's how you build a system in which you can learn thousands of things and you just have to keep doing it over and over. You know our business is one where we believe we have to earn the right to serve you the next day. Even though you ordered with us today, thank you very much for your business. We have to earn it again. The scoreboard goes back down to zero tomorrow and we have to just do that all over again.

Of what?

Very early at DoorDash. We learned how hard it is to keep someone's trust and how easy it is to lose it. And you know, I think I may have said this before, but there was a Stanford football game in which we lost a lot of trust where we were late on every single delivery because we didn't have enough drivers on the road. We had no ability to shut down the website, but we had a lot of those...

This is the third month of our operations, September of 2013, where it was a Saturday. We had no ability to fulfill the orders that came in. We had no ability to even shut down the website. So, we couldn't even stop the floodgates. And usually...

We had floodgates not 3 months in on that day specifically, for whatever reason, because of when the game ended, people wanted to order DoorDash for dinner in Palo Alto. And for whatever reason, that volume spiked pretty hard. We had no ability to turn it off and no ability to fulfill. So, we were late by at least an hour on every single delivery. I think when you go through experiences like that, but not just once, but we've had a lot of those kinds of experiences at DoorDash. I mean, I still do customer support every day. I see them literally every single day. When you see that you can lose someone's trust on one order, you realize that you got to earn it again the next day. And there is no such thing as this, you know, just set it and forget it kind of mentality. Yeah, that came a lot from the early days. But I think this daily reminder when I do customer support is also another great reinforcing function.

We were late on every single delivery. And I think it was probably somewhere around 10:00 p.m. or something where we're tallying up all the refunds that it would cost us if we wanted to make right and kind of give back everybody their money.

No one was asking for anything. We were... the night was over. We finished our last delivery and we said, okay, that was a terrible night. What are we going to do about it? We could complain about the orders or something, but at the end of the day, I think within a very short period of time, 15 seconds, we decided, okay, we got to make right by the customer. So, we got to refund everybody. Now, the complication is we had no money at the time. I was having a hard time raising... I mean, this is a pattern for me. I've had a hard time raising capital for the company in the earliest years. And that started right from the beginning. I mean, we were maybe two or three weeks of cash out and this refund would have cost us about 40% of the bank account. So it would have just made the two or three weeks and just shrunk that into even fewer days. But yeah, you're right. Nobody asked us for the refunds. I'm sure they were pissed, but nobody asked. We did the refund right away and then we stayed up that night actually baking cookies and we delivered those cookies at around 5:00 a.m. before we thought when customers would awake. And the idea was we'd rather die trying to be excellent or at least die trying to do the thing that we want to stand for than to live to be mediocre and not something that we'd be proud of. And that's what we did.

Look, these things kind of happen in steps, right? So it started with the four of us doing the deliveries and okay, well, we can keep doing the deliveries. But at some point we're going to start running into scale issues. I mean, four people can only do so many deliveries. So, of course, we're going to start recruiting dashers. We're going to start recruiting consumers, selling restaurants. And you start noticing as you do the deliveries, well, you have to build products to scale yourself. That's one. Two, you also just start noticing all the problems. And whenever you see a problem recur more than once, you would say to yourself, aha, maybe that's an example of a...

problem that we should actually build something for or actually run an experiment to see if we could actually solve. So I think very early on, the bias for action turned into this experimentation mentality. Now, we didn't have any organizations at the time or anything like that. It was just a few of us in my apartment. It wasn't like, 'Okay, there's this rigorous system that I'm talking about.' That's probably the earliest inklings though of how we thought about, okay, you can go from doing things that don't scale to identifying hypotheses to test to then running experiments and then to shipping products. That was probably the earliest time, the first year of the company.

You fast forward maybe a year as we started launching into multiple cities, all of the general managers of different cities. So you could be running Boston, someone else is running Dallas, someone else is running a different city. They would be reporting into me. And you start seeing that, oh okay, well patterns actually emerge from city A to city B to city C, but they are still quite local. There's slightly, you know, for example in Boston there's not a lot of cars. Car ownership is one of the lowest in Boston in the United States versus other places. There's some strange setups because of the historic nature of the city in terms of that hub and spoke nature I was describing that actually violate that setup. So there are local nuances and you start realizing, well okay, well how do I actually teach this way of doing things that don't scale all the way to shipping some feature that we know is going to work to each one of these people so that we could run more experiments at the same time and then we would just build more products that would actually go across all of these different patterns. So that's kind of how this thing has morphed over the years where you basically start with some basic scientific process if you will. You meet some point in which you have to figure out the next iteration in order to scale that process and then you just keep that going and you're always testing whether or not you're delivering better for customers. That's always going to be the north star metric of whether or not this process is actually making a difference or not. Is it better for customers if it's faster, cheaper, more efficient? Like what are the

Yes.

Customers are always going to want more and more selection. They're going to always want more and more affordability. They're going to want faster deliveries.

Well I think when you just think about what people want, I actually think it's pretty easy because we can play that role ourselves. And I think you just ask, you can ask very basic questions about what's the direction of travel of certain things. Like for example, like do you think people are going to expect more convenience or less convenience, especially in a world where you think that people are earning more, whether today versus the past, tomorrow versus today. What do you think they're going to do with those dollars? Is it going to go more towards consumption? Are they going to expect or demand more convenience or less? I think when you start asking questions just out loud, you get the common sense answers in which you can build a business around.

Yeah. Always. It's not rocket science. I think the rocket science is actually how do you make it happen?

Yeah.

Well, why do you do this? You know, I was saying earlier that for a few reasons. One of the things that we're talking about earlier is that so much of the magic or the difficulty of building a company like DoorDash is in all the things you can't see. And so, the first thing you got to do is you got to build observability everywhere. Of course there's observability with dashboards and systems and increasingly, you know, AI tools, but also I can see the inbound of customers who write us, whether it's a consumer, a merchant, a dasher, an advertiser, and I can choose to ignore them, but those are freebies. I mean like how lucky am I to actually have a product in which people care enough, even I mean usually they're not very positive emails, but they care enough to actually let me know. I think the greatest killer of a business is usually silence and here they actually care enough to actually let me know something went wrong in their experience. I owe them certainly not just a response, but I owe them the responsibility of actually solving that problem ultimately. And so, first, it's an obligation to the customers. Second, it's actually something that I want the rest of the company to do. I think one of the easiest things as companies get a little bit bigger, perhaps earn a little bit more success is there are more obstacles between them and the customers or the jobs to be done. For example, when you become a company, all of a sudden there are the only things that kind of get spotlighted are the financial metrics, your revenue, your profits. None of which are metrics that customers care about.

There are no metrics in what we report to as a public company that customers know about probably or care about frankly and that always is quite bothersome to me because it's because of our ability to serve customers that can hopefully achieve strong financial metrics that investors care about. And so a lot of what I'm trying to do is building as many reinforcing and repetitive mechanisms and motions, including things that I do individually, that will allow this company to always recognize that the number one job and the only religion at this company is to solve problems for customers.

It's a tough one. I think that usually there's always an element of truth in what customers are saying and it usually becomes a trade-off discussion for different teams. The reason why it's a tough decision is because it is so easy to always just veer on the side of the data because almost always when a customer notices something that is wrong or there's an anecdote that may be a quote unquote edge case. It's usually at some tail of a distribution. A distribution of the wait times for customer support, a distribution of how friendly we were when we actually took the call, a distribution of how on time we were or how late we were or how accurate we were, or what are the number of items of the types of SKUs you care about in a particular category of lettuce. Just lettuce, not vegetables, but just lettuce, right? So, it's always some tail example. And so the data is probably always going to win when it comes to a some sort of a prioritization discussion. But when you actually think about how to make a product better, it's going to almost always by definition be in improving the edges, and that's why a lot of times what I like to do personally is I love to spend time with a lot of our power users, whether it's the top dashers or the consumers who order the most often or the merchants who we've been doing business with for a very long period of time and also the new users. They're at the tails of the distribution of almost every outcome. A new user who's never touched DoorDash before and for the 13 years that we've been around will absolutely tell us about how easy or difficult it is to place their first order in a way that someone who's been used to all the things that we've been training together with customers on have figured out. A power user also sees all the issues too because they have the most shots on goal for some chaotic event to happen in the real world that we couldn't capture. And so those edges of the distribution are almost always where the anecdotes are that are the most valuable that you have to pay the most attention to because they almost always will disagree with the data and they are probably worth the most in terms of improving your product.

So the ones I love the most are actually the really long ones, the ones where there's a lot of gold. It's probably like the research you do on founders which is the longer almost the better because you get to study the distributions. When it's a short email about something you already know about there's not as much perhaps interesting material in it. I love the 2,000-word emails especially from dashers who will give many use cases of why the logistics algorithm broke for them and it becomes almost like a debugging exercise of both physical world things that have occurred, things about our systems that probably broke, and things in our products that couldn't interface well enough between the physical world and our systems. And so then I go into our debugging tools and I actually literally track the order. And every single step I'm watching and

Yeah. And you know, once I start figuring out potentially where the sources of error are, I'll either generate the hypothesis and call the dasher or email the dasher, depending on the best way to reach them or the consumer. And then actually find out whether or not there's a nugget of insight there of something we actually could improve. So put a different way, can we put a spotlight on an anecdote that improves the product? That's the opportunity I'm looking for.

Yeah. Well, the eternal mission of DoorDash is to grow and empower local economies. We say this a lot. And the reason why it's eternal is because I think it's a fight worth fighting for or a cause worth fighting for forever, which is the best way to grow the GDP or the happiness or the safety of a city is by making the small, medium, and large businesses in that city successful. They produce the vast majority of jobs and consumption dollars for the economy and the monies for the police department, the fire department, the parks, the schools etc. the hospitals. So the question is like well how do you actually make them successful? One of the most positive tailwinds of why this could be a very fruitful eternal mission is because the physical world is always changing, right? And it's hard to just scrape it. And it's one of the things I love the most about it. It's hard to just scrape all that information, say job is finished, and then put it through some LLM or something. Well, A, that data is always changing. B, it's not organized at all. And C, it's not just an it's not like some relationship between a text editor and a person. I mean, there's a lot of people. There are three people involved on every single order at DoorDash. At least there is a consumer, there's a dasher, there's a merchant, at least three people. Now given that we do more complicated things, there's even more sometimes. And for those people, this could be their identity. Back to what I was saying about small business owners and how they believe that what they do, it's not an office job or something that they just use to earn money so that they could spend consumption dollars or something else. This is like their livelihood. This is like who they are. When I think about those kinds of people, I want those people to win. And so if we have to eternally always look for the edges of the distribution to keep improving the product, of course we will. And if we can do that and we can make them successful, then they're going to make many things about the cities and the neighborhoods that we live in continue to be sustainable and very, very thriving.

Yeah. I don't even want to think about the alternative. You're totally right. I mean the alternative is it's a very robotic world where maybe we buy things in one or two ways or from one or two places. That's not a world in which you're going to grow the GDP of these cities. And actually, that's a world in which you may take away some of the identity, I would argue, of some of the neighborhoods. I think one of the reasons why people love neighborhoods or that there's certain neighborhoods that they maybe preference is because there's a personality to it. So much of the personality is given by who the businesses are and therefore you and your friends want to go frequent and go hang out in those places in addition to your homes and things like that. And that's what makes it tick. That's what makes a place feel awesome. A city feel awesome. And so I think that's an eternal mission worth fighting for.

Well, I mean the first is we have to structure it. So one of the things that I think Google so brilliantly did was they did organize a lot of the information on the internet and they made it searchable to everybody. Right now the first thing we're doing is we're still collecting lots of information and then right now we're trying to do two things with it. The first thing is we're certainly trying to grow a merchant's business by allowing you to search for their stuff through our app and we'll bring them incremental business that way. The other way is we're actually trying to make it useful for them. So, we're giving data back to them telling them when