Fireside Chat: Using AI to Launch New Products in eCommerce

Alex Wang: Hi everyone, I'm Alex. I'm the CEO and founder of Scale AI. I'm sitting down here with Toby Espinosa, VP at DoorDash, I'm so excited to be chatting with you today!

Toby Espinosa: Excited to chat with you, Alex. It's always a pleasure. Nice to meet everybody and welcome to this event. It should be fun.

Alex Wang: Toby, you've worn lots of different hats at DoorDash. I've just been so impressed by you and your operational abilities ever since I met you. Tell us a little about your journey, pre-DoorDash and at DoorDash. We'd love to learn about it?

Toby Espinosa: Absolutely. Well, I appreciate that Alex and I thank you for hyping me on this, the virtual fireside chat. My name's Toby, I've been at DoorDash for around six years. I like to say when I joined DoorDash we weren't even the largest food delivery company in our hometown of Palo Alto, which also happens to be my hometown. There was a competitor who was actually beating us in market share. Since I joined, my first job was launching new markets. I would live on the road, live in the office, I'd turn on new markets for DoorDash in Phoenix, Atlanta, and Vancouver.

Toby Espinosa: Then very early on, we saw something very interesting going on. Which was, whenever we would turn on a new market, large enterprise restaurant chain. Think McDonald's, Panda Express, Jack in the Box, Chipotle. Whenever we would turn one of those restaurants on we would see a ton of consumer activity. A lot of consumers would come to the site, merchants would perform really well and our volume would spike. Very early on, we were always looking for ways to scale our business. And one of the early ways we saw was this little business which we eventually became the enterprise restaurant business. Which was going out and getting the largest restaurants on our platform.

Toby Espinosa: So I built that with a small, amazing team from just three of us to run 120 folks over the last four years. We also, with that business, then went into grocery. So, partner with large grocers. We then partnered with large retailers, a lot of things to come there, and then I took it International, so we went to Australia and we went to Canada with a lot of the same first principles that we had when we were founded; being merchant first, thinking about all sides of the marketplace, dashers and consumers.

Toby Espinosa: Lately, I've been working on for the last 12 months, a new project. That is a new business, we're super excited to launch at some point in time. That's been my journey. I've gotten to hang out with you every once in a while, which is fantastic.

Alex Wang: So sweet. But it's incredible. To your point, I think what DoorDash has accomplished, just in terms of the ability to expand, and grow, and go into new verticals, new categories, new businesses, is just really incredible. Actually, I want to start there and spend a bit on that. DoorDash started by facilitating door-to-door restaurant delivery. As DoorDash continues to grow, how do you all think about developing new products and new services? Because you're launching a lot and it's been super exciting to see and it sounds like you guys have more stuff to launch. It's been really incredible. How do you think about that?

Toby Espinosa: Yeah. It's really interesting. I thought a lot about this as we have this core platform and then you're trying to expand new businesses on top of this core platform. It's frankly a lot of, even the way we think about launching new business. Also, using AI and technology to scale new businesses, are actually similar pieces, similar philosophies. Which is, solve a problem by figuring out if you can create it into a process and then scale it over time.

Toby Espinosa: For us, our business was actually founded to empower local economies, period. When Tony, Stanley, and Andy founded DoorDash a long time ago in Palo Alto, they were actually going door to door trying to figure out what was the largest problem they could solve for local businesses. It actually just happened that it was delivery. First use case of delivery for local restaurants, who were turning away revenue because they didn't have the technical capability or infrastructure to fulfill those deliveries. Seemed like the largest problem at the time, it was a large problem at the time. We founded that core business, that core DoorDash marketplace that you alluded to.

Toby Espinosa: Then as we've continued to expand, I think the way that we've done this is very simple. Very similar to the way you work with I think a lot of your customers, is customers come to us with a problem. It could be our core merchant customers, so a local business has a problem. A great example of this is DoorDash Drive, which is our white label capability. Merchants wanted the ability to fulfill their own deliveries or they wanted to own the consumer channel on their website. Think papajohns.com or walmart.com, they wanted the ability to deliver their own products but didn't have a fleet to do it or didn't have the logistics capability to do it. They asked us, could we step in and build that for them. And we did. That's one example.

Toby Espinosa: Another example could be in the core consumer experience, delivery for the end consumer can sometimes feel, if you're ordering a $15 burrito and you have to pay for $3, $4, to get it to you. It can feel like some hurdle that you have to go over. We built and did a ton of user research on, well, what if we did a subscription service? Where for $9.99 you would get unlimited delivery for every transaction, could that work? Could that work with all sides of our ecosystem? And so we went out and we built that business. We do a very simple process of, what is the problem for one of our core customers? How big is this? Is it a large enough business opportunity for us to dive head in first? And then do we have a competitive advantage? And if we do, if all those things match, it's like you're in Vegas and you see the slot machines and they go ding, ding, ding, they all match. And then we move forward. We've been very blessed to have fantastic people at DoorDash, great operators, great product teams, great engineers, that have then gone out and fulfilled on the dreams of our customers.

Alex Wang: It reminds me a lot of, at least what's been written and said about how Amazon has approached this problem. You and I, we both talk about this, how we admire Amazon and they have a similar thing where they have this PRFAQ process, they try to identify customer problems that are big and meaty as markets to solve, and then they go after them. One aspect I'm curious about is, how do you weigh and take into consideration all three of DoorDashes key stakeholders? I think one of the unique instances about DoorDashes business is just that how... it is this three-sided marketplace with merchants, customers, and dashers. How do you take them all into account?

Toby Espinosa: It has to be a balance, right? I think often our perception of business generally or the way the world works is you have some sort of problem, you just go head first and you forget about everything else on the sides and in reality, you and I both know that's not actually how it happens. You have stakeholders everywhere, you need to work really hard with your customers, you need to work cross-functionally internally and in our case, we just have an incremental customer and we have a three-sided marketplace instead of a two-sided marketplace that we have to think about. It can be challenging. I think the beauty of our organization and working with a lot of people. We've been privileged, as I said, to work with some of the same people for the last six years of my life. Is that as long as you trust the people that you work with and also trust the other sides of the marketplace and respect them, then you can do great things.

Toby Espinosa: But often, solving a problem might start with one side of the marketplace. In the Dash pass example that I alluded to, we started with a customer first. Thinking, how do I drive incremental economic value for our customer? For pickup, that was the same way that our pickup business was built, which is now the largest pickup marketplace in the country. Then at the same token, we have merchant first businesses, the drive business that we started. I think it wholly depends on who we're thinking about first. But then as we start to scale that business, we then start trying to make sure there has to be a check and balance to make sure that all sides of the marketplace work with us. That's a bit about the way that we do it.

Toby Espinosa: I'm curious, from my vantage point, I know you have been a student of DoorDash. But I've also been a student of Scale and you for quite some time. And I know you started on data labeling. The speed of execution of launching new products is also pretty incredible. How are you thinking about taking that first core business and then launching new products and suits at the same time you're also launching new categories?

Alex Wang: Yeah, 100%. A lot of parallels between what you had said. But I think that our mission is to accelerate the development of AI applications. We really view AI as the greatest secular shift in technology that will happen over the next few decades. Our goal is to help that happen as quickly as possible, help AI be applied to as many use cases as possible over the next many decades. The story of how I came to start Scale was that I was at MIT, I was studying AI and machine learning in a very theoretical fashion.

Alex Wang: It was super, super exciting and you could be a big dreamer and imagine all the amazing use cases. Whether that's in agriculture, or healthcare, or transportation, the list goes on and on. But I just realized very practically that there were hurdles along the way. In particular, there's a huge hurdle around data. In particular, just getting great datasets for

Toby Espinosa: Interesting.

Alex Wang: To rebuild great machine learning on top of. Similarly to your story about how DoorDash started, I think that Scale was similar in that we knew what we wanted to do was enable AI applications to happen more quickly. We like to say this thing at Scale which is that, if software is eating the world, AI is eating software. We really think about it is like, it's this really important trend. Then we locked and attack on like, "Okay, what's the biggest media serious problem that we can solve at the start?" That was really around data because it was organizationally the biggest bottleneck. That was very exciting, that served us really well. We got started in the autonomous vehicle industry, which was probably the most heavy-duty, most large amount of data, most safety centric, the highest use case of AI. Then we've expanded from there. Really, for us, the way we view it is all about being customer-centric as you mentioned. It's all about, how do we solve and break down more problems for our customers so that it is just easier for us to accomplish our mission through our customers, right?

Toby Espinosa: Yeah.

Alex Wang: The way we look at it is that, our job, our mission is to continually take the hardest immediate problems that our customers have and then build beautiful, elegant solutions and products to solve this problem. Then keep doing that over and over again. We think if we just do that and we just keep doing that over and over again, that's a virtuous cycle. For us, we view this transition of going from a data annotation partner to our customers to a full AI readiness and AI infrastructure partner for our customers.

Toby Espinosa: Interesting.

Alex Wang: You mentioned one part that I really agree with is, you guys were thinking about, you've built this amazing platform, how do you keep expanding out from that and keep helping your businesses? I think we think about it really similarly. If we identify a customer problem that we know we can leverage either a technology, or expertise, or operational capability, we're game and we're going to build products. And so that's what we've done when we launched Scale Nucleus to solve data management as a problem for all of our customers. We launched Scale Document to solve document extraction and document processing problems that we noticed across tons of our customers and ones that were holding back for businesses. We've built new products that serve eCommerce companies and marketplace companies. We built products to serve our original customers that have autonomy with better mapping capabilities and so in a very similar way that I think you mentioned early on. It's like, "Let's identify customer problems, let's build beautiful solutions to them, and let's keep going."

Toby Espinosa: Let's keep going. Yeah, I love that. I also love the story, which I hadn't thought about before, you focused very early on, on the self-driving car use case, which was the first use case. Which probably at the time was one of the meatiest and hardest problems to solve. If you can solve that problem much like delivering a pizza versus delivering a physical good. Our analogy's a little different. One is a self-driving car where there's actual large problems being involved and the other's a pizza but in our mind, a pizza is the hardest thing to deliver because it has to get to you on time, it has to be warm. If you get a bottle of shaving cream delivered to you, it's okay if it shows up cold because you don't care. I love that you started with that first meaty problem and then were able to launch from there and work with your customers on finding those new businesses. It's really interesting.

Toby Espinosa: Just as an aside, what new business would you not say yes to then? If you think of, I'm your core customer and you have these capabilities, you have some of the smartest minds. What is not a new business that you would go into?

Alex Wang: Yeah. Well, it's always tough to be prescriptive about these things. Because, I'm sure Amazon early on thought there were things that it wouldn't do.

Toby Espinosa: Totally.

Alex Wang: Pharmacy, it's got a voice assistant, they have grocery delivery. Now they have everything under the sun.

Toby Espinosa: Totally.

Alex Wang: You never want to box things in. But for us, again, it's about servicing the mission and it's probably same for you guys. Hey, if a product doesn't empower local economy for DoorDash that makes it unattractive and it makes it something that it's not furthering the mission, is not furthering the crusade, and for us it's similar, like if there's a use case where it's not helping AI achieve greater impact in the real world and greater impact with our customers. Then we'd rather focus on problems where that is the case.

Toby Espinosa: Totally. Makes total sense. And that was a hard question, I understand. If you ask me the same question I would have probably answered in the same way. We can't be prescriptive, we can't say no to new businesses.

Alex Wang: No. I loved the segue to, we talked about AI and we talked about DoorDash. But we haven't talked about AI at DoorDash. First, tell us a bit about how DoorDash uses AI to solve problems and scale new businesses?

Toby Espinosa: Yeah. It's very interesting and it's akin to the problem that you solved very early on. Which was, if we don't have a feed of data then what how can we use AI to scale our business? If you think about our core business, delivery for those folks on the other side of this chat listening, delivery's a very easy concept to understand. We've all had it for decades and decades, it's me ordering a pizza on Friday night, clicking through, it goes to the restaurant and someone delivers. But if you were to actually look at a map of all of the different things that need to go right in a delivery, it's pretty insane. And we actually have this visual at our headquarters.

Toby Espinosa: Think of the consumer ordering, taking time off, maybe they're doing something on the side, it goes directly into the point of sale or maybe a tablet in the restaurant. Restaurant gets it, they start making the food, at the same time we have to dispatch and make sure that a Dasher shows up at the exact right time to pick up the food. Every one of these pieces of the story, there is something that could go wrong or there's an opportunity where things can go right. So that by the time that pizza gets to us, it's warm, hot, ready.

Toby Espinosa: Then think about that across all different types of products in the last mile, grocery, alcohol et cetera. When you think of our business that way there's, I don't want to say infinite, but there are a very, very, very high number of opportunities for a company like us to leverage technology and AI to scale our business in a very fast way. The way that we problem solve a DoorDash, akin to your early founding story, is we'll put a SWAT team on a problem. We'll talk, maybe it's our white label business or maybe it's our marketplace business expanding our marketplace business. And we'll put a team on it.

Toby Espinosa: It could also be, maybe there's an issue with menu accuracy or maybe there's a long Dasher wait at the Cheesecake Factory in downtown San Francisco. What we'll do is we'll put a small team on it; an operator, a product lead, engineer. They'll come together, come up with a small MVP, see if we can solve this problem with a bunch of duct tape is maybe the best way to think about it. And then if we can get the duct tape right, then from there we'll say to ourselves, "Okay, can we scale this into the behemoth, can we scale this into the platform?"

Toby Espinosa: The second that we scale it into the platform, then you see massive amounts of output. In our business, because of the complexity of it and because it's dollars and cents of rapid, rapid scale. The small changes in our system, powered by technology with a massive amount of data, means a lot to us. That is really the core of what we're doing, which is finding cents and minutes, shaving these off over time, optimizing this massive logistic system first, massive marketplace system second. Eventually, now a massive software business as well. To help us continue to build businesses. That's a bit about the philosophy and how we've done historically,

Alex Wang: I love it. As I've talked to enterprise leaders who are thinking about applying AI to their businesses, I think they face super similar challenges. Which is that they have these huge, huge businesses that were 1% on point, 1% improvements, of huge... and the ability to identify problems, test them at small scale and scale them up, it's so meaningful to the business. I love what you mentioned beginning, it really depends on, you have all these businesses running and you need data feeds and you need data to be able to feed the beast of AI to make these things work in the first place. I thought that your framework was great and I think it's a good framework for other businesses to keep in mind. Is like the places where you're really going to see gains and an ROI from AI investments is when you apply it to your behemoth of a core business.

Toby Espinosa: Yes, 100%. Your core business and then new use cases on top of that core business too. I think is a great opportunity. And I think, much like you said, if software is eating the world, AI on top is eating software. I think we will all see more and more massive gains for these new businesses sitting on top of platforms because of that exact dynamic. If we can scale to some small tweaks in our logistics algorithm, means we can now do grocery delivery and alcohol delivery and a bunch of these other things and then we have this logistics engine that can pull in data to say, X, Y, Z Dashers are alcohol enabled and so they can do this delivery. You can then take that platform and turn it for any new use case. It's a pretty cool opportunity to think about for any business leader.

Alex Wang: Yeah, totally. Then how do you think about like, you all are launching a lot of new products as well, right? We talk about launching these new businesses which may or may not initially have the scale of the core business and you may not be able to get that kind of leverage. How do you think about applying AI towards new businesses or new products and do you think about it more as enabling new kinds of customer experiences or do you think about just like getting the business on the rails where as it gets scale, you're going to get those gains from AI?

Toby Espinosa: Yeah. It's a great question. There are some truly bad decisions in business. But most decisions live in the gray and most of those gray things have to do with time. When I think of applying AI to a problem, it's about time. At some point, every problem, most likely you're the expert, will be solved by some sort of AI technology, I don't doubt that. But a great example of this is DoorDash, and this is a funny example. We got a call from a large customer. DoorDash has a white label delivery business which I alluded to before. It's Dasher for order. Anyone can use our logistics system in order to fulfill their own needs. Chipotle uses this. If you go to Chipotle you use their app and you order, it's a Dasher. It's our logistics engine in the background. A bunch of large grocery companies use this, Walmart uses it. Albertsons uses it to do grocery delivery. Even some marketplaces use it, Drizly uses it and others, to do their own deliveries.

Toby Espinosa: This kind of drive capability, the early innings of this drive capability actually started because a customer, Wingstop wanted us to build the small order fulfillment capability for them. And we really wanted their business on our marketplace and so went out and built them this business. The question for us was, "Can it scale?" And it absolutely has scaled. We then got a call, subsequently from a very large grocer, and me being the excited partnerships person that I was, was looking to sign a deal. I said, "Absolutely, we can apply our white-label technology to the grocery use case."

Toby Espinosa: Well, as you can imagine, it's very different, as like any of us have. Picking up Chipotle or picking up from a very large grocery store. They're two completely different situations. One, there's a massive parking lot, where you don't really know what door to go into. There's a much longer wait time and the basket sizes are much larger. So think small Chipotle had a burrito versus massive grocery bill that's $150.

Toby Espinosa: So, we put a SWAT team on it and launched our MVP and the first car that came to pick up a grocery delivery order was a smart car. For those of you who don't know what a smart car is, a smart car is basically like a small, tight little car where it fits to people. And guess what? Groceries didn't fit in the smart car. Now, at the beginning-

Alex Wang: I know. I thought you're going to say it was Tonys' car, by the way.

Toby Espinosa: It wasn't Tonys' car. It was a smart car. The thing about this is, and this is along on the way of getting the point, which was we didn't have any labeling in our early drive logistics capability. Because we had one use case, right? And so needing to teach our systems that we wanted to assign a different type of Dasher didn't really make any sense to us, that use case didn't exist. But the second that it hit us like a freight train, because now all of a sudden, we needed a different permutation, basket size, or time of day or pickup location, all of a sudden, we had to start adding these labels. And the more labels we add to our system, now all of a sudden, this engine can take control and continue to scale, continue to scale and continue to scale.

Toby Espinosa: And so it's an example where I think in many cases, AI will be necessary for almost every business application. I think the question is, at what time and frankly for us, we both move very fast in business. So it's not like the time is decades, the time is months or years. When you start using the technology to scale business.

Alex Wang: Yeah. I love that example. By the way, I think that's like this... I also loved the way that you describe how the white label business became used for grocery, I think that's so elegant, right? Going back to the Amazon mindset, that's why being customer-centric is going to lead you to so many new businesses, so many new problems. And that just really exemplifies it. When you look back, because I'm sure this was at the time, just a total whirlwind. But what are some of those key lessons or takeaways from implementing AI systems in the white label business, do you think other people can just learn from?

Toby Espinosa: I think one should always ask the question. The beautiful thing about software is that software helps you scale. The beautiful thing about AI is AI helps you scale on steroids. And so I think for every business application that we go into every problem that we're solving, it doesn't actually have to be a new revenue-generating opportunity. There's many examples of a small feature for us, for a while we had Dasher wait time is a metric that we track Doordash, Dasher wait time is exactly what it is. It is when Dasher shows up at a restaurant, the food's not ready. So they're waiting. Dasher wait time is hard for the entire ecosystem because it means our dashers are off the road. So their time is not optimized. Consumers waiting for the food, and there's a problem with the merchant. And so we actually particularly for QSR clients, where the food is actually made in minutes.

Toby Espinosa: In order to minimize Dasher wait time, we actually launched a small feature where when we send an order to a restaurant, that restaurant is tagged as a QSR which means they actually don't fire the order until the Dasher breaks a geofence, which minimizes our time. And that's an application that has nothing to do with revenue. That where we use a tagging system, you build it in the existing massive ecosystem in order for it to scale. And so my only advice or looking back, the only thing that I think about is just like Scale was started, you started with one use case, one problem, it was a very meaty problem with a large market. And then you were able to apply that over time and scale it using fantastic technology. That's the same thing that I would do for pretty much every problem that one has to solve in business. It's the end mentality of doing the micro operator trying to solve a problem in the back room with a SWAT team, and then trying to figure out how to scale it. And the lucky thing for us is now we have technology that helps us scale it. So it's not just human capital that you have to put to put to work.

Alex Wang: Awesome. Well, I want to transition to talk about how DoorDash uses AI to just scale the core marketplace into new categories. Maybe first off just for us to understand, DoorDash has a lot of different ways of expanding its business. But when scaling the core marketplace, are you focused more on expanding into new geographies? or new categories? Or has that changed over time?

Toby Espinosa: Yeah. So I think from our vantage point, as I just referenced before, it's the end, we always think in the end philosophy, and I think it's healthy in some regards and sometimes it's not healthy. Butt here in the world of business, it is healthy. And so when we think about the and we do want to go to New geos, we just publicly announced two days ago, I believe, Japan. So we're now going to Japan, we are in Australia, we're in Canada, we're in the US, and we have ambitions to go elsewhere. But we also have ambitions to go deeper in our merchant cohort. So launching new businesses, for restaurants that consumers may or may not see. So whether that be a white label business like DoorDash storefront, where we power the software for any restaurant to run their own digital business to our drive business. That's another example. And then we also have the ambition to launch new businesses on the consumer side. And so that's going deeper and deeper into new use cases.

Toby Espinosa: I think, yes, we started with restaurants and food, we just launched 12 months ago, we launched our convenience business, which is 711, and those snacking is in that category. And we are now the number one player and convenience. And then we publicly also announced that we're going into grocery. And in our mind, there are a certain number of eatable moments that every household has. And we want to just address more and more and more of those eatable moments. And then once we address eatable moments, then we'll go outside and think more about different areas of commerce. So whether that's pets, or anything else we will we will definitely try to do it.

Alex Wang: The pets eatable moments.

Toby Espinosa: The pets eatable moments. Exactly, whoever's around you eatable moments.

Alex Wang: I love that. I love eatable moments. That's great, very memorable.

Toby Espinosa: I think, again, if we go back, we just like scale, we think about our core customer. And we think like, what does our core customer need? And I don't know how many times... I think even we've been together on a Saturday and the question is, does DoorDash have crest toothpaste? Right? And normally, historically, you have to think we didn't we just had restaurants. And then we added Walgreens and CVS and yes, now we do have crest toothpaste. And so the permutations in the consumer's mind can continue to expand. But going back to how we use AI to do this, I talked a little bit about how we built the drive platform to start getting labeling on the Dasher side to understand is the fleet ready to do a grocery delivery? Think about all the different permutations there for alcohol is the Dasher 21 plus? We have many, many dashers who are under the age of 21.

Toby Espinosa: And so as we build those capabilities on the logistic side, we then turn back around on the consumer side and said, "Well, if our consumers also want these things, all we need to do is then list them." But again, herein lies the data problem, which is I know Alexander Wang's affinity groups for restaurants, I know Indian, I know a bunch of things. I do not know what you would like on the grocery side, because I don't have any of that data. And so this is where the Ops and product thing have to come together to try to spur our consumers to come across and start ordering in new categories. We start learning much more about that consumer and then we can build more models to make every single time you open our app, you have relevant content bespoke to the user who's ordering.

Alex Wang: Yeah. This is one of the big use cases for eCommerce companies and marketplaces which is like, as you guys keep growing into new categories and keep growing your business with more merchants, the skews and options available on the platform are growing just exponentially and you want to be able to still offer incredible customer experiences, incredible recommendations incredible, like use of user affinity using AI. And this is actually the core problem that a lot of folks pull us into. It's like we have this exploding set of skews, how are we going to tackle them, manage them, make sure we properly understand what the user preferences are?

Toby Espinosa: That's right. That's absolutely right. And then how do I help? On top of that, if you want consumers to get better targeting, like better-targeted things on top, that's another massive opportunity. But going back to the forum point, if you don't know, if you're a brand new eCommerce company, and you don't know what your users want yet, then you actually can't apply it. And so again, I think every application will have an AI use case, it's just a matter of time. And again, that time will be weeks and months.

Alex Wang: Yeah. Totally. Well, what were some of the challenges that your team experience when scaling the AI systems for the marketplace? What were some of the hurdles?

Toby Espinosa: Yeah. So first off, I'm glossing over the time aspect. But I think it's really important because I think there have been examples recently, when we're building new products, where we will say, it looks like consumers don't want x, right? And we'll say they really just did for whatever reason. But the problem is, our consumers actually just haven't had the ability to tell us do they like, X or Y, right? And so you can't actually build a model to say what this consumer would like, if you have no data to tell you. And so I think timing is one of those very critical factors. I think also, over time, your model might change, like, preferences might change. And I think you need to be able to adapt. And then as all things are, early when you're using technology to scale things, you might find edge cases. And I think this is a very interesting aspect.

Toby Espinosa: As we all use technology to scale our businesses more and more and more and more, we're all taught to look at averages. So we're all taught in school to look at averages and think about it that way. But the reality is, is technology scales, you should always look at the extremes. Because in those extremes, you'll probably actually find the best little nuggets to update your technology and to scale it further. And I think a lot of people focused on the sector and eCommerce in any sort of use case, don't think in that way. And so, training the next generation of operators who know how to use AI enabled technology and product managers and engineers, you have to start training them to look at that. Don't tell me the average on a weekly basis, tell me the extremes? And what is going on at the edges?

Alex Wang: Yeah. I totally agree. And I just focus on me about this as this is like a school of thought in an operational mindset that we learned working with our self-driving customers and our autonomous-

Toby Espinosa: Well, interesting.

Alex Wang: ... because, for them, they're trying to achieve these perfect, perfect, right? Because they have to be incredibly safe, oh, as one might expect. And so their lifecycle and their operational lifecycle and what we call their ML Ops lifecycle was all about, we're going to drive our cars, we're going to identify, what are the edge cases? What are the rare cases, but those cases are tripping up our vehicles, and then let's tie that back, and let's figure out how we're going to rebuild our system to address those edge cases, and then launch that as quickly as possible like how can we do this in days rather than weeks or months, and it normally takes to update technology, and it keeps going through this loop. There's actually this presentation from Waymo, where they call this their ML factory approach to doing this and even Tesla talk about that was data engine but this operational mindset, which we did learn from this heavy-duty autonomy use case exactly match what you said, which is, identify the edge cases, identify the extremes, address them as quickly as possible, and build your systems so that you can just keep cycling and keep going.

Toby Espinosa: Totally, and what's interesting actually about that is, I love that example. Because in that instance, you're forced to. Self-driving cars, pharma defense, you're forced to look at the extremes. Ecommerce. We're not trained that way. Right? We all fill our support tickets and we think about it, but we're not trained, we think what is the core customer? Who that core customer? What core problem you're solving. We're not trained to look at things on the sides. But we can take learnings, that's an interesting thing where even for your customers in eCommerce can learn from your customers and self-driving.

Alex Wang: Yeah. No, that's super cool. And I think that, frankly, every business can learn from this approach, because it's such a unique way of looking at it. But it makes sense in AI contexts or in AI-enabled framework, because all of a sudden, you have technology that's going to allow you to deal with the common cases super well. How are you going to keep dealing with the extremes and keep tackling the corner cases?

Toby Espinosa: Yes. Absolutely. I love that. You should do a blog on corner cases. I love that.

Alex Wang: You'll co-author with us, won't you?

Toby Espinosa: Yeah. We'll do it, we'll co-author it together.

Alex Wang: On that thread, what other advice would you give to other marketplaces, or eCommerce companies as they're looking to start or Scale AI initiatives? Not everyone sees the scale that DoorDash has seen, but because of that entry of great learnings for the rest of the community?

Toby Espinosa: Yeah. I'll recap the two that we discussed and then maybe I'll go for a third. I think the first one is, everyone will use AI in their business period. It's like saying, I'm never going to use software. It's just that everyone will use AI period. It's a question of timing and hills. So when you use it is the question. And so I think, if everyone out there, every operator changes their mental model to I will use this, it's a matter of when, that'll drastically change the way you run your business. Two, as we just discussed, I love this edge case thing. I think we hire a bunch of incredibly smart people from all top firms. And the first thing we have to reteach is, don't show me averages, show me the extremes, because in the extremes lie the truth. And what's funny about that is, as you said, like, that will only become more pronounced as AI does the middle.

Toby Espinosa: So I think that's a very, very important thing for every operator, every product manager, every person in tech, and any industry to think about. And then the third is how to find people that can do those top two things really well. And embedded in those top two things is the word and, and is not either-or, but it's and which is one of our values at DoorDash, I think is going to be more and more important for every single person that works at our companies, regardless of what role you do or what role you have. And if you can find people that do both the diving really deep into the thing that doesn't scale, understanding the problem from first principles that doesn't scale. And then also can think about how to use technology and a team and human capital to scale something. Those are the types of folks that you want around the table. And so those three things together. Again, it's not do I need AI, it is when do I need it? Two is... What is Two?

Alex Wang: Two is extremes.

Toby Espinosa: Exactly. Two is extremes. And three, that you need to find people that can do the end. I think those are the three things I'd leave everybody with, and hopefully DoorDash listens to those as well. Sometimes we go off-kilter. But if we listen to those things we will continue to do quite well.

Alex Wang: It's so incredible, and amazing insights. I think truly every business whether or not you're a marketplace, or eCommerce company, or not, like can learn from those insights and I love the framing of and, and finding people who have and in them, it's such a powerful concept, but I think that all of us know... like I certainly know who I think of when I think of you.

Toby Espinosa: Yes. Oh, there it is. And I think of you.

Alex Wang: Thank you so much for taking time today, Toby and frankly what you and the team at DoorDash have accomplished is just absolutely incredible. It's really quite staggering and going from what you mentioned, which is the number two restaurant delivery platform in Palo Alto to obviously the global behemoth that DoorDash is now is just absolutely incredible. And I can't wait to use DoorDash in Japan.

Toby Espinosa: I can't wait for you to use it too. We would love the order. And thanks so much Alex for having me and I'm so impressed with all the work you and the team are doing at Scale and excited to continue to watch you change the world.

Alex Wang: Same to you. Let's get at it.

Toby Espinosa: Let's get at it.