You’ve probably never heard of the company doing the most robot deliveries: 5X more than all their American competitors combined. The company is Starship, and it has already completed 9 million autonomous deliveries, crossed roads over 200 million times, and operates thousands of sidewalk delivery robots across Europe and the U.S.
Now they’re scaling into American cities … and they say they’re ready to change your world
In this episode of TechFirst, I speak with Ahti Heinla, co-founder and CEO of Starship (and former co-founder of Skype) about:
- How Starship’s robots navigate without GPS
- What makes sidewalk delivery better than drones
- Solving the last-mile problem in snow, darkness, and dense cities
- How Starship is already profitable and fully autonomous
- What it all means for the future of commerce and city life
Check out our conversation below. As you hit play, do me a favor and …
- Subscribe to my Substack
I’ll unpack our conversation, summarize the insights, and contextualize the findings on Substack. I’ll also (soon) start releasing shows on Substack first. - Subscribe to my YouTube channel
It may start coming later than Substack, but you’ll always be able to catch TechFirsts on YouTube
Transcript: Starship’s 9 million robot deliveries
Here’s a gently edited version of our conversation. Check the video for a complete and possibly more accurate version.
You’re right — that earlier version was missing the very beginning and had some choices you’ve since updated. Here’s a fully re-cleaned transcript of everything you just pasted, with all your preferences applied:
-
“Tech First” → TechFirst
-
Fix obvious name errors (John Koetsier, Waymo, Grubhub, etc.)
-
No bolding of city or company names
-
Remove or reduce “mm-hmm / yeah / uh-huh / wow / mmm” backchannel noise
-
Don’t repeat the speaker’s name when they’re clearly just continuing
-
Fix clear typos and transcription glitches while keeping meaning, tone, and order
John Koetsier (00:01.599)
What changes when robots deliver everything? Hello and welcome to TechFirst. My name is John Koetsier. We’re hearing a lot about delivery robots lately in just about every city imaginable. Robots with wheels, some robots with wings, maybe even a few robots with legs. There’s a ton of innovation here, but I bet you haven’t heard about perhaps the biggest player in the space, with nine million deliveries under their robotic belts. It’s called Starship.
They just raised $50 million to bring robots to your city. And we’re chatting with the CEO today. Welcome.
Ahti Heinla (01:37.471)
Thank you for having me. Excited to be talking about this exciting topic here together with you.
John Koetsier (01:44.334)
As am I, as am I. Let’s start maybe super broad. Nine million deliveries. What’s that taught you about robotic logistics?
Ahti Heinla (01:54.123)
All right.
So Starship is a company that is developing robots that deliver stuff — delivering food and groceries to your doorstep in the cities — and we also operate in college campuses. We have done nine million deliveries with our robots. We have thousands of robots operating in hundreds of different locations in both Europe and the US. And we have been at it for essentially a decade.
So we started about 10 years ago. And that’s the time that it takes to actually mature a product like this. If you think of it — for example, autonomous vehicles — when did you first start hearing about it? I think Google, for Waymo’s parent company, announced their plans in 2009.
They had already, I think in the first year of development, driven like a thousand miles with the vehicles, right? But it still took a very long time after that — more than a decade — to actually start rolling out commercially. So there’s a very big difference between having a prototype, having something that can do some deliveries or drive around somewhere, and actually having a commercial reality which is reliable, safe, works in any weather, is effectively more reliable than a human courier, and also is cost-effective — because it can’t be more expensive than a human courier, otherwise people would not use it.
So that’s what we have learned. Ten years ago we had a prototype. Now we have a commercial product that is doing millions of deliveries and being rolled out in tens and tens of different locations and cities.
John Koetsier (03:50.695)
It’s a really good point. I took a Waymo a month ago in San Francisco. It just felt normal. It just felt completely normal, right? And the new models that they’re coming out with, they won’t have steering wheels. They don’t need them anymore — gas pedals, brake pedals, all that stuff. That’s actually interesting. We’ll get to that too.
I want to talk about how your robots work, how they sense what they do, what Level 4 autonomy is, what it means, weather, snow, any remaining technical challenges if they exist, and what this all changes about commerce. But I think I’ll go right there.
We all kind of know, I guess, the science fiction author William Gibson’s famous line, right? That the future is not evenly distributed. When will this just be normal? I mean, it probably is for you. It probably is for kids on college campuses that are using this regularly, right? And there are places where Wing, the drone delivery company, is using it. People are getting their coffee and that’s their normal.
But for, let’s say, 99% of people, it’s not normal. Of course, you raised a big round to kind of make it normal. When do you think it will be kind of just normal, default, assumed life in a city? You want something, boom, it’s delivered by a robot almost instantly.
Ahti Heinla (05:03.627)
Great question. In the places where we operate, this actually is completely normal right now, and it gets that way very quickly after we launch somewhere. We launched our first commercial-quality operation in 2018. That was seven years ago. That was not in the US, that was actually in the United Kingdom in a city called Milton Keynes.
That’s the name of the city, Milton Keynes. And we started delivering groceries there and have been doing so every day since then for seven years. So this city just completely gets their stuff using our robots and that has been the case for seven years. There have been kids who have been born during that time who don’t remember, they don’t see, they don’t have any recognition of the time where there were human couriers delivering stuff. It’s robots.
We actually take up a significant percentage of all deliveries in the cities where we have deployed, and people very quickly get completely used to robots — completely used to robots. You see the robot. It’s not the case in these cities that there are robots just completely everywhere and it’s not as if there are as many robots as there are people in the city, right? But there could be hundreds of robots in a city, and you could see them maybe about as often as you see somebody with a pram or with a baby walking. It’s not as frequent as you see an adult walking, but maybe as frequent as you see a small child walking with an adult, for example. So you see them quite often.
And we have some customers that have ordered with our robots thousands of times. This is completely commonplace. The only reason why it is not completely commonplace everywhere is that we haven’t expanded all over the world yet. We are in tens of cities in Europe right now. We are in tens and tens of college campuses in the US. We are launching soon in US cities as well, doing food and grocery delivery. And during the 10 years that we have been at it, we have solved all of the problems that there are to solve.
So this is a completely commercial-quality solution that lots of people are using, that is coming to your city probably quite soon. People think this is going to be a reality because if you just look at science fiction movies, you don’t see people knocking on doors like couriers; you see things coming to you, things flying to you, things being created out of thin air. That’s the reality that we are creating and that’s actually the reality that exists in a lot of cities and a lot of places right now. It’s just not everywhere yet, but it will be everywhere. We have solved everything there is to solve.
John Koetsier (08:30.343)
That’s a big statement: “We’ve solved everything.” Because cities are big, noisy, dirty, sometimes crowded places. Sometimes they’re working on the street. Sometimes a sidewalk is chopped up and they’re working on the sewer or something like that. Or there’s an accident and there’s a car in the way or something like that. Talk about the technology that you’ve built in and what capabilities your robots have to navigate all this.
Ahti Heinla (08:59.797)
These are the things that take a decade to solve. It’s a decade of collecting data, gathering miles. We have driven more than 12 million miles with our robots.
And we have learned so much with it. Our robots go on the sidewalk. They also go on bike lanes and drive on the roads sometimes. They cross roads using crosswalks and intersections. How many times are we doing this? We are actually doing this on average multiple times per second. So during the time we have talked here right now, our robots have probably performed already hundreds or thousands of crosswalks or intersections all around the world. We do a million of these per week. We have done a total of around 200 million road crossings.
I mean, that’s the amount of data we have, and that’s what it takes to learn from all of these situations — to train the AI, to train the machine learning, to see all of the edge cases. What happens if somebody with a bicycle comes from the right-hand side and turns unexpectedly? What does the robot do in this case? What should the robot do in this case? It takes a lot of experience and a lot of learning to actually make this work and make it safe.
It is much easier to just have this one robot driving a little bit and it kind of looks fine. But if you do 200 million road crossings, you’re going to have some issues, and it takes a decade to solve these issues.
John Koetsier (10:56.948)
Yes. Huh. Weather as well. Snow. I mean, you’re from Estonia, relatively northern Europe. There’s going to be some weather. There’s going to be some winter. There’s going to be some snow. You’re going to have that in American cities, Canadian cities as well. Does it work in that?
Ahti Heinla (11:20.523)
Great question. Weather is a big part of it — a big part of what separates a prototype from a commercial-quality operation. We operate in the heat of the southern US. We also operate in Wisconsin. We get a lot of snow in Wisconsin. We also operate in many cities in Finland.
Finland has lots of snow and sometimes very poor weather. In fact, we cover a very significant portion of the territory of Finland, like a double-digit percentage. We are approaching nationwide coverage in the country of Finland. We get a lot of snow and we have gotten a lot of experience on how to operate in that. So snow, rain, and obviously even the most obvious thing — darkness versus light.
It all looks different during the night, right? Generally, when autonomous driving companies start their first trials, they start in sunny climates. Why? Exactly for this reason. We have graduated well beyond that. We operate in all sorts of climates and that’s what is actually super important. Our customers are delivery apps and grocery retailers. These are the businesses that do not really tolerate failures.
If, for example, we dispatch a robot to do a delivery and the robot only gets there 90% of the time, 10% of the customers would be super unhappy and they would switch to a competitor — a different delivery app or different retailer. That wouldn’t work. So it’s 99.9% reliability that is needed to create this as a commercial reality, not just as a prototype. And we have walked that path, we have done that, we have solved that.
And that’s also what Waymo has done with the car. Some time ago, it maybe wasn’t as commercial-quality a system as it is right now, but now it is a very fluid, very usable commercial-quality service, just like Starship. There’s only a couple of companies in the world that have actually made autonomous driving this sort of commercial-quality, commercially mature autonomous driving. It’s Waymo, it’s Starship, it’s Baidu in China — they’re also driving with cars — but actually almost nobody else.
Almost everybody else is operating at a much, much smaller scale where they’re not yet at the point where they’re just going city by city, deploying thousands and thousands of robots and operating in a safe and also cost-effective way. Economics is also a big part of it. We are saving money for our customers because we are charging less per delivery than it takes for a human courier to do it.
John Koetsier (14:27.853)
What’s the savings?
Ahti Heinla (14:32.669)
We are giving cost savings to our customers, obviously subject to commercial negotiations, but the cost savings are a significant part of why our customers are interested in this. Affordability is very important. As we all see, recently things start getting more expensive for everybody. Affordability of delivery, tipping — there’s no tipping with robots, right?
That’s really important. The unit economics of it absolutely need to work. This is not some sort of luxury service or something that is like a premium-priced service where the unit economics do not matter. Unit economics absolutely do matter. We are doing it purely for high scale. We are offering cost savings to our customers. We are offering commercial-quality, reliable service where we can scale to city after city after city, and the world of delivery will not be the same.
John Koetsier (15:48.173)
Super cool. So I guess you’re not disclosing exactly how much you save. I’m guessing it’s at least half or something like that. I think I saw at one point, for winged delivery — drone delivery — the cost was like 10% of sending out a car or a van or a vehicle with a human. So there’s a lot of costs there with bigger vehicles.
Of course, you’re an EV as well. You’re using electricity, not gas. So it’s saving on the environment and probably much cheaper to operate in that way. What technology is on the robot? I’m assuming there’s cameras, I’m assuming there’s LiDAR, I’m assuming there’s a few other things.
Ahti Heinla (16:23.627)
Yes, there are cameras, there are radars, there’s lots of different kinds of sensors. And in fact, having been at it for a decade, the sensor evolution that has happened is amazing. Back then, when we started, fewer sensors were available. Right now, there’s much more available in terms of sensors.
When you combine these new sensors with the amount of driving data that we have — with the 200 million road crossings we have done and so forth — you can really apply AI and machine learning in a way that makes the robots just work, and work, work, work very nicely. So the sensor evolution has been very important. We have radars, we have cameras, we have many other types of sensors as well.
Interestingly, we are actually not using GPS all that much, because GPS, for our purposes, is just not accurate enough, not by a long shot.
John Koetsier (17:30.272)
Also, it’s problematic in cities, right? You’re in the shadow of a large building or trees or something like that and you lose the signal.
Ahti Heinla (17:37.759)
Yeah, exactly. GPS can get you a rough bearing of where you are, but our robots navigate with a precision of about one inch, effectively. So GPS is not actually very useful for us. We actually understand where the robots are visually. We look around at what the robot sees with its 10 cameras and we compare it with the pre-built map of the area.
John Koetsier (18:03.938)
Cool. Update your map in real time — there’s construction here, it’s going to be for a few days maybe, the other robots know they can route around. All that stuff, super interesting. Your robots can also climb a little bit, right? I mean, they need to climb up when there isn’t a ramp from the street to the sidewalk. They can climb up that little incline. That’s correct, right?
Ahti Heinla (18:22.729)
Yeah, they climb up inclines, do steps as well, they can obviously climb curbs.
So they can get to pretty much anywhere, and this reachability of various areas is important. They do stay outside, but they can get right up to the door, depending on where exactly the door setup is. We drive as close as we can get.
And that’s actually an advantage of specifically sidewalk robots, right? If you think of how close to your front door an autonomous vehicle on the road can get, or how often and how quickly, or how close a flying drone can get, then a robot can actually get the closest.
John Koetsier (19:22.716)
Which is actually really important in a dense city. The space up to a place may not even have a road that cars can go on. Certainly in Europe you have that, and in some of the denser American cities you have that as well. College campuses — I can imagine some of that as well.
What’s the range? I’m assuming when you go in to service a city, you’ve got to build a hub, maybe a couple of hubs where the robots come in, they get serviced, they get a fresh battery or recharged or whatever. What’s the range that you can typically deliver?
Ahti Heinla (19:54.859)
Charging is actually fully automated by now. So we do not have any facilities where the robots drive to a facility to get charged. They actually live outdoors. And they don’t regularly return to some sort of a facility. We do have a servicing setup. Sometimes the robots need some maintenance or repair and we do have people to do that repair. But for example, in the country of Finland — it’s as big as a US state, so it’s not a massive country — we have only two maintenance facilities, and we are covering tens of cities.
We are covering tens of cities where we don’t have any people or any of our own employees in these cities. Robots go on just by themselves. And that’s actually really important. That’s, again, something that we did not have seven years ago, but we do have right now. And that is absolutely necessary to make this commercial-quality system work.
We have built a whole system for autonomous logistics at scale. It’s not this one robot that requires a lot of human care, like engineers to calibrate the sensors every morning. That was the case seven years ago. It’s no longer the case right now. And that’s necessary. This evolution is completely necessary to make it a commercial reality. If engineers needed to calibrate the robot every morning, it wouldn’t work in terms of commercial quality and commercial maturity.
John Koetsier (21:38.559)
So what’s the range?
Ahti Heinla (21:42.859)
The battery actually lasts for the whole day, so it could drive tens of miles on a charge. But generally, when we do grocery delivery or food delivery, it’s usually a couple of miles that our deliveries are, and the robots do many deliveries during each day.
John Koetsier (22:04.883)
Cool. So I want to ask a question that maybe you’re one of the few on the planet who can answer, because you’ve been operating continuously for seven years in places like Milton Keynes. What does this change about life in the city — in terms of commerce, how people live their lives, and how stuff gets where it needs to go? You get your groceries… what does it change?
Ahti Heinla (22:33.643)
It’s very interesting how quickly people actually accept robots in a city. It happens very, very quickly. When I, as an entrepreneur, started this 10 years ago, we had our first back-of-the-napkin calculations and first thinking: okay, would this work? Can this work? One of the risks that we identified was what if people just do not accept robots driving around in the city?
I don’t mean the consumers that are getting our deliveries — I’m sure they love it — but just random people. Are they accepting robots on the sidewalk? But they do. They actually accept it very quickly, even if they have not seen one of our robots before. They see, “Oh, what’s this thing driving down the sidewalk or the road? Oh, okay, it’s apparently a robot. Okay. Okay, so robots are a normal thing.”
People wave to our robots. Kids feed our robots bananas. It’s a little bit of a social phenomenon. People really develop an emotional attachment. For example, it happens quite often that when a robot stops somewhere for a second, somebody around thinks, “Does the robot need help? Can I help the robot? Can I give it a little bit of a push?”
It does, yes — it actually says “Thank you” when you do something like that. Robots actually do not really require help. They are six-wheel drive; they don’t really get stuck. But imagine the social experience or the emotional experience: “I helped a robot on the street.” That’s not something that in my childhood I could ever have said, but people today can do that.
And it’s also a change in terms of the uses of our autonomous logistics system. Decades ago, when people were booking flight tickets, they used to go to a travel agent. Remember that? They went to a travel agent face to face and they said, “I want to go here,” and the travel agent wrote them a ticket or printed them a ticket — it’s like a physical, you know, on paper. That used to actually happen decades ago.
And now it’s all done online and people are booking their flight tickets while in their bed or anywhere they are. That is a little bit of the transition that has happened right now. Just like nobody really goes to travel agents anymore — maybe some people somewhere do, but most people don’t — in the same way, people have really accepted that you get your deliveries with a robot.
A robot shows up at your door. A robot says, “Thank you for your delivery,” and you say hello to a robot effectively. It becomes a commonplace thing.
John Koetsier (25:53.822)
What about for retailers that use it and regular customers? Does it increase the number of orders? Does it increase order velocity? Does it decrease average order size because I can just order what I need today and get it delivered? What does it change about commerce there?
Ahti Heinla (26:05.848)
Yeah, revenue definitely increases. It increases adoption. When we work with grocery stores, for example, we generally work with grocery chains that have hundreds or thousands of stores, and we equip hundreds of stores with robots. We then deliver in a couple-of-mile radius around each store.
We generally see — or the stores see — that their revenue, their orders, effectively go up by about 10% or so because people love robots. Now it’s a really unique way to get things delivered to you. So we’ve seen huge adoption for this.
John Koetsier (27:05.533)
Cool, cool. Is it also usable for a hot meal? Like you’re ordering a meal — a Grubhub type of thing or something, or Uber Eats or whatever — and it comes and it’s still hot. Is it quick enough for that?
Ahti Heinla (27:20.337)
We are actually cooperating with Grubhub in college campuses. You can order with the Grubhub app in many college campuses and you get this with a Starship robot. Yes, it absolutely works for food. It works for grocery. We have worked on things like temperature insulation for the hot food so it doesn’t get cold and all of these things. These also take years to figure out and perfect in a commercial-quality setting.
So food and grocery — these are the things that we are doing right now. We could also deliver other things in the future. There’s a whole revolution happening, but we have started from these two segments of the market because these are the biggest.
John Koetsier (28:08.679)
Sure. Will you go to the air as well? Will you do drones? Or will you partner with a drone delivery company, maybe for large, longer distances, more rural spaces or not-as-dense urban spaces? Or is that just not your lane and you’ll let others do that?
Ahti Heinla (28:26.923)
Right now we are not using flying drones, but I’m sure we will also fan out to do other types of deliveries and other types of robotics in the future. We have started out with this one product, this one robot, that can do the majority of grocery and food deliveries.
But yes, there is some specialization where different-sized robots, different-speed robots will specialize in certain types of deliveries or certain types of areas. In the future, I do absolutely imagine that there is a multimodal combination of various delivery methods. I’m sure also people will continue to do deliveries. It’s not going to be 100% robots.
John Koetsier (29:18.652)
Sure, big heavy things require a large vehicle. Absolutely, yeah.
Cool. This has been super interesting. Thank you for taking this time. I really appreciate having this conversation. And what cities will you be coming to first — new cities?
Ahti Heinla (29:38.603)
I can’t tell any specific names right now, but we are working with a number of major delivery apps and major retailers, and there will actually be some announcements coming soon.
John Koetsier (29:52.453)
Very good, looking forward to it.
Ahti Heinla (29:55.374)
Thank you.