How Zipline Scaled to 2 Million Drone Deliveries | Keller Cliffton

The places that have the most access to technology are the places that appreciate it the least. 99% of the world isn't that. They're dying for innovation, dying for a tech startup to show up and help them solve problems in a really innovative way. That's kind of been the story of Zipline. It's like we went to parts of the world that people ignore. Normal delivery platforms actually have net promoter scores of like -10, -20. The service right now has a net promoter score of 95. We would deliver to apartments. We would deliver to town homes to hotels. We even deliver to public parks. We are hitting new record numbers of deliveries practically every day. This is something that is going to grow a lot.

Sure. The simplest way to think about Zipline is that starting about 13 years ago, it seemed to us pretty obvious that someone was going to build an automated logistic system for Earth. Especially when you look at the way that instant delivery works in the world, whether for an emergency medical situation or for something that you want delivered quickly to your home, we're generally using technology that's like 100 years old, which is a human driving a 3 to 4,000lb gas combustion vehicle to deliver something that weighs on average about 5 lbs. And so we always thought this was bizarre. If you think about it from first principles, that is not the best way to solve that problem. You don't have to be a physicist to realize that using a 4,000lb vehicle driven by a human to deliver something that weighs 5 lbs is not very efficient. So we always felt like it should be possible to design a robotic system that would do that 10 times as fast, dramatically less expensive, zero emission, just an overall better, safer, more reliable product experience. I always envisioned it a little like the way the internet moves information around. We used to pay people to carry letters on horseback, now we send an email, and I think a similar thing is going to happen with packages and I think you also see it happening with people right now. That's the promise of autonomy. It's a global network that gives everybody a totally different level of access to things like healthcare but also to economic opportunity, packages, products.

Well, before we went to Rwanda, we started in a trailer park, which is what we have behind us. On a hill. We call this place Nezero. We were five people. We had this idea about building this new kind of logistics system, but we needed a lot of space to test it. No one thought that this was going to work. People thought that it was kind of preposterous. So we needed a place where we could fly and test and learn. And so we ended up on this farm. This is an active cattle ranch. The owners here were incredibly gracious. 12 years ago.

We didn't know them. They've owned this land, I think, for like a hundred years or something like that. The family has and they've been ranching cattle on it for a hundred years. We met them through a mutual acquaintance and they put up with us. We ended up signing this little lease with them. I think the original lease was like $3,000 a month, which was a lot to us at the time. We actually moved ourselves out here. Some of us were sleeping in these little construction trailers. I would often sleep on the couch. The advantage of being out here is that we had hundreds of acres to design, build and fly. So we would design all the different fundamental components. We had our little manufacturing area here and then we would just be flying every single day. When you walked outside, there was always going to be an aircraft flying around. And if something went wrong, if an aircraft had a problem and basically crashed itself on the bluff out here, we would go out and pick up the pieces and people would be crying while we were picking up the pieces, because it was not clear that the company was going to survive.

The cows honestly don't really mind us. They mainly break in and eat our stuff. They eat our rope and our hardware. So the fences are mainly to keep the cows out. But being out here wound up being this interesting forcing function for even the kinds of people that Zipline could hire. In the early days, the idea of hiring an engineer from Google was like the best thing you could imagine, which is amazing how far we've come. We'd have a lot of people out here but you could tell normal people were so put off by the idea of working out of these construction trailers and driving all the way out to the middle of nowhere every day. Everything we were doing was very scrappy and unfancy, just make it work in the real world as fast as possible, and the company was running out of money every month. So I think that attracted a special kind of person. We attracted people who wanted an epic adventure, wanted something different, were willing to take a lot of risk, knew how to get their hands dirty, and operate in actual real weather and real world conditions. Those traits are kind of rare in software companies, but they are required for the survival of a hardware company. So that's the background on Nest Zero and how we came to be here.

There are a million. Early on when we were designing the first generation of our system, we wanted to test how we would recover these aircraft. It's a fixed wing aircraft that doesn't have landing gear and we didn't have runways where we were operating. So the idea of how to get the aircraft out of the air is not an easy thing. We eventually decided we were going to prototype a solution that kind of looked at how aircraft carriers arrest planes out of the air. My co-founder went to Walmart and bought these two deep sea fishing poles, then basically built this metal hinge and bolted that into the ground, strung a line between these deep sea fishing poles, set the tension, and then we flew the aircraft at those fishing poles and it worked. People were pretty shocked because that was a very ghetto way of rapidly prototyping and showing if this would work or not. We found out that it did work, and then we ended up building a much more advanced version of that which we call Snapcatch. Snapcatch is still in use today in distribution centers all over the world. It's recovered probably 1.8 million aircraft reliably. There were a million things that we had to do out here that looked totally crazy. When we started building this, everybody thought it was so obviously a bad idea, so obviously not going to work.

It's not ideal. You could see the writing on the wall. And I was like, well, it isn't over. We're going and doing this, we're doing that. We're launching in Rwanda. We think that they're going to let us deliver these blood transfusions. We had this grand ambition. The idea was preposterous. Designing a fully autonomous system that would operate in the real world had never been done before. Even to this day, you only see the first system starting to scale now, 10 years later. The idea of starting in Africa was kind of preposterous. Everybody thinks that the best technology is going to start in the US and then trickle its way out to developing countries, rather than the other way around. The technology is extremely complicated and I don't think people thought that a team of 15 or 25 people could field early versions of this technology in a way that would be useful for people. To a certain degree, I think that they were kind of right.

Exactly. This is the definition of Zipline.

Yeah. A certain amount of naivety is 100% required. If you knew how hard it was going to be when you are starting the thing, you probably would not try at all. There are a million things, whether it's the sewage tanks literally never worked here. So we have a truck that comes and pumps these tanks. We had 50 people working out here. We definitely reached the breaking point for Nezero at some point where we had too many people here every day and we would have to pump the tanks every single day and everyone would be sitting trying to write code and smelling this incredibly nasty smell. It was exactly nobody who was coming from the Google Plex, living in the lap of luxury.

Yeah. And you'll smell while you're writing code every day. Suffice it to say, it was definitely a rag tag team of adventurers, and that really set the culture of the company for a long time. People who wanted to put up with that and who wanted to operate in these kinds of conditions. That was a small subset.

I think it's way easier if you don't have a lot of other awesome options to worry about. Previously to starting the company, I was rock climbing full-time. I was a professional rock climber for a year, which is way less fancy than it sounds. I was basically living out of my Honda Fit, driving all over the US and traveling the world. I made about $3,000 that year, and I had a blast. I got to see the world and go to all these inspiring places. I think that was empowering in a way because I realized I don't need a lot to be happy. I didn't have a lot else going on in my life, no real obligations. It seemed to us like it should be possible to do something like this. The idea inspired us as we started spending more and more time looking at problems that we felt would be inspiring to solve, especially in terms of healthcare logistics. We always felt, okay, no one is going to give us regulatory permission. No country is going to give us the ability to fly at scale unless the use case is so incredibly compelling and clear.

Yeah. Lives being saved that we could kind of force it through. So we had this thought that we needed to focus on a small country. We wanted a public health care system. So obviously that kind of eliminated the US. That first country was Rwanda. We were talking to a couple different countries at the time. But I remember this conversation with the Minister of Health of Rwanda where I was going in, totally clueless. I show up in a hoodie as an engineer from Silicon Valley saying, 'Hey, we'll build these autonomous aircraft and we can deliver every medical product in the public healthcare supply chain, deliver it 10 times as fast in a way that's going to save all these lives.' And she kind of looked at me and was like, 'Keller, shut up. Just do blood.' As she was explaining it to me, she was like, '50% of transfusions are going toward moms with postpartum hemorrhaging, 30% are going toward kids under the age of five. It's this incredibly important use case.' And it was very narrow. I actually didn't find out until 10 years later, or eight or nine years later. I was on a panel just a year ago with this minister. Her name is Agnes Binagwaho. I was on a panel with her at UN General Assembly in New York and she told the other half of this story. She had heard me say this and I didn't know this, but basically that morning before I met her, a tragedy had happened where a mom was giving birth in a hospital in Rwanda and she had postpartum hemorrhage. There was this cascading chain of failures where they were supposed to have blood in the hospital, but they didn't. They placed an order for blood, but it couldn't be delivered. Someone from the hospital got into a car and drove to get the blood, but there was traffic and they couldn't get there fast enough. They got to the center for blood transfusion and it took an hour once they got there. By the time they got back to the hospital, this mom had lost her life. There was this big email chain inside the Ministry of Health on how the heck did this happen. She was reading that email chain and then I happened to walk into the office 20 minutes later. So you realize getting any progress in these kinds of startups definitely relies on an element of luck. There was probably more luck involved there than even I appreciate because we walked in with this idea that seemed totally preposterous, totally stupid. There were a hundred reasons for her to tell us no.

But I think she felt frustrated, like how could this have happened? And sort of beside herself enough to be like, 'Yeah, we'll try something crazy.'

I think there was less than zero data. We had not fielded a system commercially in the real world. We were a small team of 15 people who knew nothing about healthcare or aviation. So there was every reason to believe that there was no chance we could do this. We did not yet have regulatory permission, which seemed like a pretty big consideration, and we didn't know how to integrate with a national health care system or with a national airspace regulator. So there were a lot of reasons to say no. And yet this minister basically said shut up, just do blood, and she gave us 21 hospitals and said if you could deliver blood transfusions to these 21 hospitals in a way that is 10 or 100 times as fast, that would save a lot of lives. The president of the country was also deeply involved in the contract. He had a lot of vision. He is a kind of extraordinary technocratic leader. I think he also knew that Rwanda doesn't have a lot of natural resources. So if they want to win, their natural resources are their people and technology and innovation. They've kind of modeled the scrappiness of a startup.

Yeah. The reality was that that country felt like it was a startup in its own way. The modern version of that country is 30 years old and they are moving incredibly fast and ultra scrappy. I could always at 8, 9, 10 p.m. on a Saturday get on the phone with any of the ministers or with the regulator, and we were able to move lightning fast in a way that would not have happened in a bigger, more complicated market like the United States.

Yeah. Genius and talent and entrepreneurship is evenly distributed in the world. Opportunity is not. So I do think people probably underestimate the degree to which a lot of this cutting edge AI and robotics technology could transform major parts of the world. We can build new kinds of infrastructure that can save money and save lives. We're still largely technology is still serving more tactical or rich white people problems in the US. We're focused on these better understood markets, even if in reality they're not necessarily the largest.

I don't think there was any... I mean, we lived there. We lived there for months and years on end. We would travel back and forth between headquarters here in San Francisco and Rwanda.

I would read textbooks. I love reading textbooks. I would bring an 800-page textbook and my goal would be to read the entire textbook on the journey. I loved it. You didn't have an internet connection, so you were locked in. We made that flight hundreds of times. We built what we call the Zip House, which was this little house we rented right next to the distribution center, and we all lived together. I bunked in bunk beds with Abdul, the first team member that we hired in Rwanda. Abdul's kind of become a legend. His story is legendary at Zipline. Abdul and I would wake up early in the morning, take ice cold showers, do push-ups, and then jog to the distribution center together. That was how we figured things out and built the company for the first couple years.

When you're trying to do something really hard, maybe people have this sense that some companies will execute super well and it goes right on the first try and then some companies don't execute that well and it's kind of a disaster. The reality from my perspective is very different. It will 100% always be a show. And then the companies that execute really well can rapidly iterate themselves out of that pain. At least that's what we tell ourselves because the first year was really difficult.

That was our experience. We were doing everything we possibly could to get the system to work at all. We had been given this contract to deliver to 21 hospitals. We only launched one hospital to start. So it's not like 20 hospitals were going without. They were just relying on the old system. We had switched one hospital over to Zipline. The system was extremely unreliable. We could not get it to operate. It was so painful. We could get it to do a couple deliveries a day and then it would break and we'd have to figure out what was wrong with it and fix it, then a couple deliveries a day and then there'd be something wrong operationally. So it was day after day, night after night, long weekends working between the Rwanda and US teams trying to marry this technology with this complicated operation with this health system. We also had to get regulatory approval for what we were doing and integrate with the civil aviation authority, the Rwanda Civil Aviation Authority. We had no clue what we were doing. We built the first version. We were holding all these medical products all of a sudden. Blood is not simple. You have packed red blood cells, platelets, cryoprecipitate, plasma. So four different types of blood and then you have AB and O positive and negative RH factors of each. Really hard to get the right thing to the right place at the right time. We had to be managing this inventory because we were now holding this precious product. We wrote a Google Sheet that we were using to do inventory management. Everything was the bare minimum, scrappy, the minimum viable product, the scrappiest version of what we thought would work. For 9 months we struggled to serve that one hospital. After 9 months of all-nighters and nights and weekends, we were able to get the system working pretty reliably for that one hospital, we expanded to the other 20. And then from there, once it was working for all 21 hospitals, we expanded into a wide range of other products, vaccines, infusions, transfusions, cancer products, basically everything in the public healthcare supply chain. Then we expanded from 21 to 100 to 500 to 1,000 to 2,000 to now 5,000 hospitals and health facilities across eight different countries. It's crazy how many years of your life will be taken off by the stress of the first one. Now we add hundreds at a time and I don't even find out about it unless I look on Slack.

Again, everything that could go wrong did go wrong. I was in the US at the time but we were probably 3 or 4 months in and we were supposed to be able to deliver very precisely. We were supposed to be able to deliver to three or four parking spaces, that was the idea. That's what we called the mailbox. This zip was supposed to be accurate enough that we could put it in their mailbox every time. So they would always know where to collect the package and go back inside. I got this call at 2 a.m. We had a pager duty system. I got woken up at 2 a.m. and one of the members of our Rwanda team was like, 'Hey, Keller, we have a problem.' I said, 'Okay, lay it on me.' They said, 'The zip control algorithm wasn't working correctly. There's a mom who was having an emergency C-section, needed an emergency delivery of blood and we delivered onto the roof of the hospital.' I remember thinking, wow. We missed big. To end up on the roof, we missed by like 50 or 100 feet. It was a big problem. I immediately started asking all these clarifying questions. Then I called Ryan, my co-founder, and a bunch of other people on the team, and we rallied from 2 to 5 a.m. trying to figure out what had happened, why was the GPS so off. We made some updates to the system and immediately pushed out a software update a couple hours later. I called the Rwanda team member back and said, 'Hey, we think we have a solution. It's being rolled out now. But by the way, what happened to the blood?' She was like, 'What do you mean?' I said, 'Is it still sitting up there on the roof, this blood bag sitting out in the sun?' She said, 'Oh, no, no. When we delivered onto the roof, some nurse climbed up and got it and went back inside and they transfused the patient 5 minutes later.' I realized in hearing that, there's this very interesting experience for me. If a startup has a good enough mission, if the need is high enough, your customer will meet you halfway. In this case, we had failed to provide the product experience that we had promised. The product was not working correctly, but we had gotten the blood close to them, even if it was 100 feet away and on a roof. This guy was willing to risk life and limb climbing up the side of the building and onto the roof to get it and go back in and save this person's life. It was very powerful for me. It inspired me. It was like, you know what, we don't have to be perfect. The mission was so important that we were willing to launch things that were not yet perfect, that were actually in some cases bad, and it was still good enough to save lives. We were able to rapidly iterate from there to something that was extremely reliable and dependable.

For a hardware company that's operating at Zipline scale today, Zipline now has 135 million commercial autonomous miles and zero safety incidents. So we take safety incredibly seriously. That's why we have these large test sites. We're at our first test site. And yet somewhere in the world with a fleet this large doing that many commercial autonomous miles, there's always something going really wrong, especially at the test sites. I think you realize that every morning you're going to wake up and whatever is the most screwed up thing, that's what's going to be at the top of your inbox.

So I think you have to realize it's important not to have an emotional reaction because it's completely exhausting. It feels so bad.

Especially for a company that's trying to go really fast and do something very difficult, nine out of 10 things you try on any given day are going to fail. Zipline is very used to failure at this point. We are constantly trying new things, failing in a wide variety of ways, having to rapidly iterate from there. So you definitely learn to, instead of craving success, if what you are sustained by and what motivates you is success, success is actually few and far between. The vast majority of the experience is failure. So I think that you have to be very zen about actually being in love with the process of being in a painful, uncomfortable situation and having to rapidly iterate your way out of it and learning by doing. The first 10 times you try, you are going to fail. So you have to be motivated by failure and motivated by that learning and growth, rather than by having some big success or getting a bunch of praise. Those things don't come that often, especially for the first five or 10 years of a hardware company.

There were times in Zipline's history where the company's survival 100% hung in the balance and every day we needed to be attacking core engineering challenges or core operational challenges in a very relentless, disciplined and focused way. Something we actually learned from one of our biggest investors, Valor. Valor was the largest investor in, I think, one of the largest investors in Tesla and SpaceX for many years.

We jokingly called them WIFO moments at Zipline. WIFO, 'We're It's Over.' Like, okay, here's another WIFO moment that we're encountering. I can certainly remember one in the early days when we were actually launching in Rwanda. The president was coming to the distribution center for the launch. We had been working for months to get the system ready. The day before the launch, we were doing a final test flight and the aircraft crashed. It was a complete disaster. We had to work through the night to fix it. We managed to get it working just in time for the president's visit. That was a WIFO moment.

We had a distribution center to inaugurate and do our first commercial delivery. We had been soft launching for a couple weeks before the president showed up, and it was a disaster. The system was not operating successfully. We had used a contractor to design the launcher, which would catapult the aircraft into the air, but it was terribly designed and would break every third launch. We were trying to fix it at 7 p.m. the night before the president was coming, with the BBC and New York Times showing up. They built a big tent for the president to give a speech and launch the first aircraft. At 7 p.m., the launcher broke. We tried to fix it until 10 p.m., then 1 a.m., still not fixed. Abdul and I were on our backs in the gravel, holding flashlights in our mouths, taking it apart and putting it back together. We were Skyping the team in the US, like Apollo 13, saying 'Houston, we have a problem.' They asked what we had to fix it, and we had duct tape and styrofoam. They walked us through it. Meanwhile, the president's security team, five Navy SEAL-looking guys with submachine guns, watched us work at 1 or 2 a.m.

Yeah. They were looking at us like, 'You guys are totally... what are you doing? The president is showing up in seven hours and you're taking this launcher apart.' We put it back together, tested it once, and thought, what more can we do? The media and ministers showed up, the president arrived by helicopter, gave a big speech, came to the distribution center, and launched the aircraft. It worked perfectly. We were as amazed as everyone else. The zip flew off, made a delivery, came back, landed, and we acted like it was totally normal. The president spent two hours with us talking about the future, then left. The system broke again for the next nine months. So luck was involved. Hardware companies, especially with each new generation, face such higher scope and scale that every cycle nearly bankrupts the company. The Model 3 nearly bankrupted Tesla, even 16 years after founding. When you're doing big hardware launches, getting it reliable and launching into the real world is incredibly hard. You have a complicated global supply chain, and it only takes one unreliable supplier to screw you. Your performance is the worst of your 10,000 suppliers. If the company is growing fast and building a big product, a couple months of not producing can mean bankruptcy. We went through that with Platform 2. It served a market 100 to 1,000 times larger than our original focus on healthcare logistics in rural areas. Platform 2, operating behind us, is designed for teleportation logistics from any business to homes, 10 times faster and dramatically less expensive. But it was way more complicated, and we were six months late shipping the product. Had we gone another six months, Zipline would not have survived. We went into hardcore mode, burned the boats, no one was going to give us more money. We needed to launch. That was about a year and a half ago, even less. The team worked through Christmas and New Year's, and we launched on January 15th.

That's the main trick. In this industry, automated logistics and drone delivery, no one knows what it looks like at scale. We're figuring everything out from scratch. There are no precedents. Tesla knew they were building a car with four wheels and a steering wheel, but no one knows what this looks like. How should charging work? What's the business model? Maintenance? Regulatory strategy? Reliability? Pricing? No one has any idea. One of Zipline's advantages was that we were the smallest team. Amazon, with Jeff Bezos on 60 Minutes saying they'd do drone delivery to every home in the US within two years in 2013, was spending hundreds of millions or billions a year. We had a tiny fraction of that and a tiny team. We weren't the biggest or richest, but we were the most practical. That was in our DNA. We lived in construction trailers, worked nights and weekends, and moved super fast to get things into the real world. The trick with hardware systems like this is that safety is paramount. Look at Uber ATG or Cruise—a bad safety event can impact a company's trajectory. It was especially important because Zipline was delivering life-saving products. We needed to move fast, so we became extraordinarily good at testing. We invested in a large software-in-the-loop testing team and built powerful hardware-in-the-loop (HIL) systems, learning from SpaceX, the best at HIL. These systems let us deploy new software onto the aircraft's avionics connected to a simulator, so the plane thinks it's flying while sitting on a desk. That catches 95% of software bugs without crashing an aircraft.

Yes. But even so, 5% of problems can only be caught at the test site. That's why Zipline invested in large test sites operating 24/7 in the US. We have three major test sites, one just opened last week. They give us access to real-world conditions impossible to replicate in simulation. Some stuff only breaks in the real world—weird edge cases like a bird nesting in hardware or solar weather, which became a big consideration for GPS reliability. Solar weather is something most people don't know about but is really important for these systems.

We were complete idiots and had no idea. We launched in Rwanda with zero context on what we were getting into. Had we had any common sense, we would have looked at the weather forecast before signing the contract. Rwanda is a high-elevation country with some of the most volatile weather on Earth and challenging terrain. It was hypergravity training. We had to build a system that 10 million plus people could rely on with their lives. Failure was not an option. We had to harden the system to extreme weather events. That work became immensely valuable because Zipline is now the only autonomy company operating at this scale in all those gnarly weather conditions 24/7. We were forced into that by the original use case. Antonio mentioned we collected 40 to 50 million autonomous miles before starting commercial operations in the US.

The big vision was to build an automated logistics system for Earth. Just as the internet moves information efficiently, logistics is built on humans, gas combustion vehicles, and unions—an antiquated system. We felt it would all be automated, dramatically less expensive and faster. The advantage is that logistics today serves maybe the golden billion people. The other seven billion have poor or non-existent access. The most exciting thing was to build a system that serves all people equally. That was the promise of automation and robotics. In 2019, big US companies and health systems saw what we were doing and wanted similar solutions for telepresence, home delivery, and home healthcare. Companies like Walmart, Chipotle, Sweetgreen, and health systems like Cleveland Clinic were excited to launch. It became a chicken-and-egg problem with the FAA. Drone delivery companies asked for permission, but the FAA said prove it's safe. Zipline had 40 to 50 million commercial autonomous miles outside the US with zero safety incidents, which we provided to the FAA. That was really valuable. The FAA moved pretty quickly. We've always been an amazing partner to them. The FAA wants to innovate, but they're constrained by legal bureaucracy and processes like NEPA. We created a productive partnership over five years, pushing the envelope year after year. The FAA has been a partner every step. The narrative that the FAA is impossible or too bureaucratic has been the exact opposite over the last three to four years. There are no regulatory constraints on Zipline's business right now. Once we showed it was safe, the FAA accelerated this.

We've expanded from Rwanda to Ghana, Nigeria, Côte d'Ivoire, Kenya, and the US, and we'll announce new countries next year. In all these countries, you meet people who are alive because of Zipline or have loved ones who are. You hear stories all the time. People at Zipline will be in cabs and hear about someone's sister, mom, son, or grandpa. Global public health institutions estimate that Zipline has reduced maternal mortality in Rwanda by 51%. The minister of health of Rwanda was named to the Times 100 list this year for that. Many thousands of lives saved. Other studies estimate Zipline saves about 17,000 lives a year. We're about to triple or quadruple the scope of our platform one infrastructure with a US State Department contract, going from 17,000 to 60,000 lives a year. That's a huge part of why we do what we do. Logistics is often seen as a luxury for the rich, but it's a necessity for the poor. Instant delivery is a necessity. The less time you have to drive yourself or get to a good hospital, being able to meet people where they are is incredibly important economically and for healthcare. As we've expanded, that's critical for the brand and community acceptance. People often say, 'Our drones save lives. Our drones deliver blood transfusions.' There's a strong sense of local ownership because Zipline hired entirely local teams, created high-paying jobs, and promoted people into global leadership. It's not an American or African company; it's a global company. Anyone can rise super fast if they're extraordinarily talented. In San Francisco, the places with the most access to technology appreciate it the least and have strong anti-tech, anti-progress strains. You see the coning of autonomous cars in some cities. Waymo hit a cat, and there were calls to ban it, but Waymo is 30 times less likely to hit a cat than a human driver. Do we care about the cats, or is something else going on? Some cities have had it so good from technology that they've become negative, thinking the golden goose will always be there. But 99% of the world is dying for innovation, for a tech startup to help solve problems. They're dying for entrepreneurship, technology, and high-paying jobs. There are entrepreneurs in all these countries and parts of the US dying for an opportunity. You only have to open the door an inch, and they'll do the rest. That's been the story of Zipline. We went to parts of the world people ignore, outside the US and inside the US. We launched in Bentonville, Arkansas, and Dallas, and we'll announce other metros that aren't seen as tech hubs. The cool thing about this next generation of AI and robotics is that it requires operating in the real world, in partnership with government, regulators, and cities. That will lead more companies to find government partners as aggressive and startup-oriented as they are. They exist and might be the majority, just not the metros people think of for advanced technology.

It's hard to know the counterfactual. Zipline got incredibly lucky at key moments. The biggest mistake is around hiring. Everyone has to learn these hard lessons themselves. The biggest mistake I made again and again was falling in love with specific experience over innate characteristics. Whenever we hired for specific experience over innate characteristics, it was a huge mistake. In a fast-moving startup, you hire someone for a role, but two months later everything changes. If you hired someone because they were deep in one thing, they become irrelevant when the company changes direction. We realized there were four innate characteristics that, when we hired for them, those people always ended up winning at Zipline: practical problem solver, fast learner, low ego, mission-driven. Those are hard to teach. If we hired people with those, they might fail a couple times but eventually succeed. It's crazy how many times I made that mistake.

Every bad hire is extremely memorable. Firing people is no fun. When you're leading a startup, you constantly ask everyone to believe in you. You say, 'I have this great idea, we're going to hire this person, they'll be amazing.' Then three or six months later, you say, 'Bad news, that didn't work out. I was an idiot, but now I've learned my lesson.' Leading a startup is shameless. You make stupid mistakes and constantly ask the team to believe in you again. There's a line: 'Winning is going from loss to loss without a loss of enthusiasm.' There was a lot of losing. We hired many people who didn't work out. Looking back, the mistakes are about how we could have bet bigger on people who stayed for a decade and became executives. You want to hire as many of those as possible and treat hiring as a sacred duty. No amount of time with a candidate is too much because the cost of getting it wrong is so high, and the leverage of getting it right is so high. Those are the biggest things. Could we have moved faster to build the team? Maybe everyone has to learn those lessons themselves.

We've largely listened to the market. For five or six years, it was clear the US wasn't open from a regulatory perspective. We needed to prove what we were doing and expand our healthcare logistics business outside the US. We had our hands full. In 2020, we could see three or four years into the future that automated logistics would dramatically accelerate. If we could build a product that competes against cars and is dramatically better, with enough regulatory progress, it would work in every country, including the US. We had a good understanding of the problems, customers telling us what they were, and big partners like Walmart willing to sign big contracts. At every stage, you're 10xing or 100xing, building hardware that's more cost-effective, reliable, and higher performance. Each product is more complicated than the last. SpaceX went from Falcon 1 to Falcon 9 to Starlink to Starship. Platform 2 is 10 times as complicated as Platform 1. It almost killed us getting it into the real world, but the market on the other side is incredibly vast. Instant delivery is one of the biggest markets on Earth. We now have data showing that if you make it 10 times faster, less expensive, more reliable, and safe, customers consume 10 times as much. The instant delivery market in the US today is 5.5 billion deliveries a year, but there's demand for 50 billion. There aren't enough roads, cars, or humans to do that job. We need to use technology that takes cars off the road, gives humans high-paying next-gen technology jobs, and is quiet, serene, and beautiful for neighborhoods, reducing pollution. That's the vision. It's only this year that it's become clear, looking at how Zipline has been scaling in the US. It's going to happen over the next three to five years.

Last week. Maybe I haven't had that day yet. You live in a state of constant paranoia. I don't ever tell myself it's definitely going to work, but there have been eyebrow-raising moments over the last couple weeks. We're hitting new record numbers of deliveries practically every day. Platform 2 deliveries in the US have been growing about 15% week over week since April. On a graph, it's super vertical. When you're getting into thousands of deliveries a day, even though it's still small compared to UPS or FedEx, it's something that will grow a lot and become completely normal. I visited a grandma in Bentonville, 82 years old, who ordered from Zipline 350 times in the last year. The service has a net promoter score of 95. Normal delivery platforms have net promoter scores of negative 10 or 20. Many have scaled fast with net detractor customer bases. It's a testament to the opportunity and market scale that people are willing to use a service that's expensive, unreliable, and sometimes unsafe, with bad product experience. But even so, those platforms have scaled to extraordinary scale. It's obvious these systems need to be automated. We shouldn't be using 100-year-old technology to solve a problem we've only understood in the last 5 or 10 years. AI and robotics will transform instant delivery and global logistics. That's the simple idea Zipline has pursued for 12 years. Now we see customers adopting the service, ordering everything through the Zipline app. It's crazy to see user behaviors. People order at offices, universities, apartments, townhomes, hotels, even public parks. Realizing you can pull out your phone, press a few buttons, and have something delivered quietly, precisely, and safely 10 to 15 minutes later, no matter your GPS coordinates, changes how people live. People change their grocery shopping behaviors. Families say they grocery shop once a week and do three to four fill-in orders with Zipline. It enables them to eat healthier, fresher food, waste less, and save three to five hours a week they don't have to battle traffic, park, or drag kids into a store. Everything is delivered to you, giving you time to enjoy with your kids or loved ones.