How Matic Built an Intelligent Home Robot (While Others Failed) With Mehul Nariyawala

With hardware, you do not wanna be in a business of creating a market as a startup. It's insanely hard to do hardware itself. And then if you have to create a market and justify value for it, you are in a world of pain because it's extremely hard to predict supply and demand. So starting with a category that already existed where we can go in and solve the problem and earn customers' trust and then expanding into various use cases is the right way to go about doing it.

Hey everyone, Welcome to the Founders in Arms podcast with me, Imad Akhan, co-founder and CEO of Mercury.

And I'm Raj Suri, co-founder of Lima and Tribe. And today we have with us Mehul Nariyawala, who is the founder and CEO of Matic. Welcome, Mehul.

Thank you. Founder, co-founder and president, but not CEO. CEO is Navneet.

Oh, okay, okay.

He's the final boss, big boss.

Okay, okay. Co-founder and president.

Yes.

And it's Matic Robotics, right?

It's Matic Robots.

Yeah, yeah.

Matic, okay.

Yeah, I guess tell us a little bit about what you do.

Yeah, we are a home robotics company. We saw in 2017, we saw an opportunity to build Apple of home robotics, partly because back then there were 200+ self-driving car startups in Silicon Valley. And prior to this, both me and my co-founder were at Nest. And we sort of came out, we had been thinking about homes, and it just kind of surprised us that there are 200+ teams thinking that they can build self-driving cars, but not a single team believes that we can build a robot that doesn't need to bump in our homes. And that sort of piqued our curiosity.

Your main form factor is this vacuum, right?

That's correct. Our first robot is robot vacuum and mop that we build and ship. And this is— we think of it as sort of like Robotics 2.0, first actually intelligent robot that we're delivering into the world. It just felt like it was time to kind of put a nice ribbon and have a rest in peace thing to Disc Robots.

I guess you started this, but Roomba was already a thing. In hindsight, it's now obvious that they were beatable and you could build something way more superior. But what gave you confidence that like, hey, this established company that's pretty well-funded can be overtaken with a better idea?

That's a great question. Partly because I had a golden retriever back then who passed away, but I had a golden retriever. Our joke was that he sheds twice a year, 6 months each. So I had tried all disc robots. And one of the last ones I got was Dyson's 360i robotic vacuum. And turns out Dyson's robot, which is really, really innovative company, was the worst of them all. It was probably the dumbest robot, and nine out of ten times it would just get lost in my home, not even get back to dock. But the worst was that it got onto one of our really, really nice rugs, and Dyson has amazing suction, and because of that high suction and not enough torque in the wheels, it got stuck and then stayed at it for 45 minutes in the same exact spot. So when we We came home, picked up the robot, the entire patch of the rug was gone.

Wow.

So $1,000 robot, exactly ruined a really, really nice rug. And that's when it dawned on us that none of these guys are approaching any sort of robotics from any sort of vision or intelligence point of view, that Roombas, Disc robots, and Dyson robots were really just automation without intelligence. And then I'm like, wait a minute, it's kind of crazy. I don't think I've ever met anyone who said Roomba changed my life. that this is the greatest device in the world. Yet they've been around for 20— back then 15 years and now 23 years, and people just keep buying them. And so we started digging into data and turns out market was growing 16% year over year. That's as fast as smartphone growth after iPhone interruption. And entire category NPS score, even though it was growing 16% year over year, entire category NPS score is literally negative 1. It's a product that consistently gets disrecommended, yet people keep buying it.

Yeah, I had a Roomba and it always just fell down the stairs and it actually made it kind of useless because I was like, I had to like put stuff in front of my stairs every time I like switched it on, which I was like, how hard is it to like see the stairs? That seems like an obvious thing to like catch.

Exactly, exactly. So it just seemed like there was a huge opportunity penetration, like even though so many robots were sold, robot vacuums were sold on Leap. Back then, only 10% of the US households had robot vacuums. Now it's about 15% versus 100% have vacuums. And it just kind of— the framework, and I don't know where I heard this framework, but the framework is, you know, we had this 3-dimensional framework in mind, which is one, what is the intensity of a problem? What is the frequency of a problem? And how many people in the world have it? And that sort of kind of gives you an understanding of the potential opportunity. And turns out floor cleaning, when your floors are dirty in a home with families, is actually quite annoying. So that's a quite intense problem.

What was position number 2 and 3 in this framework? What was like a second or third product? Like was it dishwasher or laundry? Like what's like the one that comes after this in this framework?

Anything in the home. So home chores in general, we defined as a big problem and everything we felt was pretty intense and frequent and consistent. The other one that I would say which is bad would be, and this is why we didn't do it, is lawn moving robots. So lawn moving robots are all the rage, but if you think about intensity, maybe once a week, maybe twice a week, but it's outside of your homes. You don't really, uh, you don't really feel all that crazy about it. And it's not— frequency is also not that great. So, so those are not the product that we would go after, for example, in this framework. Um, yeah, but it wasn't like floor cleaning and home robotics was one framework. There were completely different ideas. Like one of the ideas we did think about and rejected 2017 was code that writes code. Turns out that was a pretty good idea as well, but we rejected it because we just weren't sure if consumer AI and consumer software can have a business model for long term. 'Cause ultimately, you know, free models come out and make everything cheaper and then it's a race to bottom. So how do you do that?

So Mehul, I'm a proud owner of the Matic and—

Thank you.

My kids love it and it was really impressive to like, I mean, the unboxing experience was super cool. You know, like, Iman, if you haven't tried it, I mean, if you're just a fan of robotics, you should get this product because it's a really cool unboxing experience and a cool— the advancement in robotics, you can clearly see it in the Matic. And I think it's a much more likely form factor than a lot of the humanoids. The humanoids are very complex. I just saw today, this morning, that Elon's announcing that The humanoids will be ready to sell by end of next year. And someone retweeted it and quote tweeted that he said the same thing a year ago. And there's other companies like the Sunday Robots and stuff, which are trying to do dishwashing and stuff. It's going to be hard, right? It's a very complex product. We'll probably get there at some point, but it's not—

Precisely.

But what you've done is you've taken a familiar form factor and you've modernized it in a big way using new technology. And yeah, so the unboxing experience is cool. Like the robot just slides out of the box, right? Like it presents itself. It's probably the best unboxing experience I've ever seen. I mean, there's no other example where the product just comes out of the box for you, right? So—

Like when you open the box, it like comes out and says hello or something?

Yeah. So we had this insight from day one that If you have to pick up object out of the box, then it becomes an object. It's not intelligence. So if it's an intelligent robot, like even as a kid, if you wanna see if there is a bug, we try to touch it and if it moves, we're like, oh, it's alive. So how do you make sure that the robot is alive?

That's a fun insight.

Yeah. So that's how we started thinking about it that, look, if it rolls out of the box and says hello, then it feels intelligent and also makes a connection with the family, puts a smile on a kid's face within 30 seconds. And that allows us to instantly become friendly.

That's fun. How'd you make sure the box is the right way around? Is there lots of instructions?

No, no. We worked with many, many iterations of the box, but ultimately found one. So the way it works is it's sort of almost like presenting a trophy. So you just pick it up from the box and there is a ramp that falls down. And then there is a giant sticker we had to put in because most people are so used to picking up their objects that they just would pick up. So we had to put a giant sticker saying that, no, no, don't pick up the robot, just press this power button. And as it turns on, it kind of comes out of, rolls out from that ramp. Yeah.

Oh, that's cool.

And so yeah, it is really cool. And the box was constructed in a very intentional way. I don't even want to know how long you spent on that experience. Probably a lot of time on that. But the, you know, and then what it does is it maps your house with computer vision, right? So it like, it goes around, it bumps into all the walls, it like builds a map that you can see on your phone. It's a pretty cool experience as well to see that. And it builds a very useful sort of understanding of your house, which I think is probably gonna be a standard for all home robots, right? They need to understand like the layout of your house, right?

That's correct. There are 2 fundamental things we understood back then. So one was this idea that if you're trying to build self-driving cars, which are getting much closer to reality now, they have 2 advantages. One is the Google Maps, which allows cars to know where the road is going. Another one is GPS, which allows the car to know where it's located on that road. But without those 2 information, no matter how smart the self-driving robot is, it would be lost, we'd be lost as well. But inside a home, how does robot know whether it's on the right side of the couch or the left side of the couch, whether it's in front of the TV or in front of a refrigerator? And that precise mapping and localization was the missing piece of the puzzle. But for us, this idea was that, look, indoor home is built by humans for humans for our visual perception system. So if we're gonna build a robot, we have to give it the same visual perception system that humans have. And that was one of the first insights. And then second thing is, I would give credit to Navneet on this, but Navneet has always had this belief. His background is in computer vision. His PhD thesis in computer vision was a foundational piece of classical computer vision with about 55,000 citations with single paper. And he has always had this thesis from day one that I don't wanna bet against nature. So we sort of thought about this idea that, hey, all of us, each of us are family men. All of us want Rosie the robot. All of us want our laundry and dishes and everything to be done. But how do we get there? Well, we don't necessarily— nature doesn't create a full-blown human being on day one. What nature does is babies are born and for first 5 years of their life, 0 to 5, all they do is learn how to perceive the homes and perceive the world. So maybe in the context of this floor cleaning robot, we can teach robot how to perceive the world in a very precise manner. And that's step one. And then you can ultimately think about adding manipulations to it and dexterity to it. And then eventually— the long horizon planning and long horizon execution of various tasks. So that was the thing. So we do believe that humanoids will happen. We just think there's much more steps to get there versus here. For sure, for sure.

You got the right approach, I think.

So it turns out I apparently talked to Mehul for his angel round, which I failed to invest in. I just looked up the email. I didn't even pass properly. I just like, we just talked and then I ghosted you. Which I feel bad about, but it's annoying because I don't have notes on it. But yeah, tell us about that angel round. Like, was it hard to get that first round? I mean, it's hardware. This is like pre-AI, you know.

Consumer hardware.

I'm guessing consumer hardware. Yeah, I'm guessing it was like a pretty challenging round to put together, right?

So I would never say that any funding round is easy. It's just all relative because we had a It wasn't as hard as we did for Flutter. So we prior— because we had an exit and we had been acquired twice by Google by that time. First one was this company called Flutter. Flutter was, I wanna say, 10x harder than the angel round we did for Matic. But it was hard. But what we were able to do, having been in Silicon Valley for a long, long time, is that one, it was led by Matt Rogers, who was co-founder of Nest. And then we had built a bunch of connections. And through that connections, we were able to just get enough money to get it out the door. But we had a working prototype, so that made a difference. I don't know if you remember or not. I think I brought that prototype and maybe even showed you a demo in that cafe, if I remember correctly, but I'm not 100% sure.

You know, honestly, I feel like I would remember a robot in that cafe. That was my go-to cafe back in that office. So I don't think you showed the robot, but—

Okay, so maybe that was it. Maybe that's why I wasn't able to convince you.

Uh, and well, so Flutter was also a hardware company, right? Like you, you're just like very into doing kind of hardware companies.

So no, no. Uh, so our entire prior experience, uh, till Flutter was all software. We came from computer vision background. Flutter was gesture detection.

Oh, really? What was Flutter?

Flutter was gesture detection via webcam. So think of it as a Microsoft Kinect. Oh, I see. Got it. Right. Uh, the idea was if you're sitting in front of a TV, instead of turning your finger or hand into a mouse, Why can't you just shush if you wanna mute, right? Which is, as computers get smarter, they should understand us. And we thought we could use these RGB cameras instead of time-of-flight cameras that Microsoft Kinect had and solve the problem. So we launched this app that allowed users to control iTunes, Spotify, YouTube using gestures. And gestures were play, pause, previous song, next song, thumbs up to like, and shush to mute. And that worked very well, but that was an app. And very quickly we realized that no one wakes up in the morning and says, today I'm gonna buy gestures. So it was really, really cool technology UX, but not a product. So when we left Nest, so, but luckily for us, Google and Apple were interested in acquiring us. So we got acquired by Google. But then when we got to Google, we realized that there were 2 things we could have done, which is we could have actually created an entire unit that you could plug into TV with this UI, but we didn't have a hardware background. So we didn't even think about that. The other thing was we were using these cameras with these cameras. I don't have access to frame rates per second. I don't have access to auto exposure. So that's like building software for eye that is a bit wavered. It's like, uh, so that was the second part that we were only— we only built half solutions. So, so we said that, okay, next time when we go and do something, we're going to solve a real problem, we're going to build a real product, and we don't want to arbitrarily limit ourselves between hardware or software. So let's go learn how to do hardware. And luckily for us, uh, Tony Fadell was investor in Flutter. So when Nest got acquired by Google, there was an option to move there, and then they acquired Dropcam. Which then it became very obvious that Dropcam, which became Nest Cam, they would have cameras in the living room and outside and there is a computer vision need to it. So we transition over and that's really— and then just as we were transitioning, both Dropcam co-founders left. So then I, out of that chaos a bit, I got a chance to lead the entire Nest Camera portfolio from a product perspective and Amit led the computer vision perspective. And that's where we really learned hardware. That's where we got our hands dirty in this particular thing.

What did you learn about hardware that was like not obvious? Like what were the things that you really took that like you applied to Matic?

Many, many things. So first I will start from the category perspective. So one, everything in hardware is obviously hard, but one of the product that I worked once was Nest cameras and Nest cameras became very hypercompetitive category very, very fast. But if you want to go and buy decent thermostat for your home, Nest is still it. It's been 15 years since they launched. Not a single new product in that market that even remotely comes close. Okay. So if you can tell me that I can build a product where I won't have a competition for 15 years or 15 years later without any doing any innovations on top of it, still remains the number one product. Sign me up today. Right. So that was really a curiosity there. And we kind of dug into why that was the case, because microeconomic principle says that wherever there is a profit, competition will come. Well, Nest cameras were emerging category. There were no profits, yet everyone was entering that category. And on a flip side, thermostat— you don't have to convince any homeowners to buy a thermostat. It's legally required. 130 million homes, clear-cut market, and no one is coming in this space. And as we dug into it, realized that turns out it's not just about hardware or software or UI. All of that stuff is hard, but hardest part about Nest thermostat is making it compatible with last 60 years of HVAC system. That's tedium that no one wants to deal with, right? Similarly, Nest Protect, which is smoke alarm and really the only smart smoke alarm on the market even today after 12 years, 500 to 800 pages of regulations depending on states and things. And that's just whack-a-mole kind of system that people don't want to do. So we realized that actually one of the reasons why I picked floor cleaning category is precisely that, because it was really, really unsexy, uncool category to go into and We knew that Amazon is not coming, Apple is not coming, Google is not coming for a long, long time in this category. And if we can actually build a genuinely great product, we would have a long opportunity to build a dominant company, so to speak. So that's how we started the event. So that was lesson number 1. Lesson number 2, which I think most people miss and we would've missed it too. So if you kind of think, if you, you know, as Raj knows, we made a bet on day 1 that we're gonna do the whole robot vision only. And vision only meant that we will only use cameras and not add any other sensors, very similar to Tesla's full self-driving approach. And the reason we did that is because inside Nest, we had a rule of thumb that a single sensor you add in a hardware, assume 3 software engineers on a flip side as a permanent cost. So more sensors you have, the bigger the team, more sensors, the more calibration, more sensors, the more failure points, more complex the supply chain, BOM cost, manufacturing, so on and so forth. So with each sensor, complexity actually rises exponentially. And we saw this in Kickstarter campaigns where Kickstarter campaigns after Kickstarter campaigns of consumer robots get funded, but by the time they're supposed to ship, either they're prohibitively expensive or they never ship. So we made this bet that, okay, we're only going to do it with, uh, with cameras and absorb all the complexity on the software side and computer vision side. If we can pull it off, that is the only way to make indoor robot, especially home robots. But I'm willing to bet that even in indoor robots, that is, the vision only is the only way make them economically viable. And so that was a second key learning piece of the puzzle. And then third one is with hardware, you do not want to be in a business of creating a market as a startup. It's insanely hard to do hardware itself. And then if you have to create a market and justify value for it, you're in a world of pain because it's extremely hard to predict supply and demand. So starting with a category that already existed, Where we can go in and solve the problem and earn customers' trust and then expanding into various use cases is the right way to go about doing it.

Yeah. And you don't have inventory risk, right? Because, you know, you can build on demand, right? So you don't have, it's not like really, it's not really the Christmas gift type of thing.

Exactly. So many, and there are so many lessons. Like one is if you go the other lessons on a business front and I'll talk about, there is obviously hardware-oriented lessons, but on business side, the typical answer we might be that, hey, let's build a hardware product, let's sell it on Amazon, Amazon, Best Buy, or any of these traditional channels. And the moment you do that, you have inventory risk. The moment you do that, you're looking at 9 to 10 month vertical integration cycle. That's probably not true. Then we also, when you, the other conventional wisdom for hardware companies is, hey, you should manufacture in China or you should go and have many CMs that does your work for you. But really, Any innovation or product is number of iterations. And as Elon says, if you want to innovate fast, you got to reduce the time between that iteration. But the moment you go to CM, now you've got long lead times. You have to make sure that you give them design 3 months, 6 months, 9 months in advance. They're only going to work with you if you scale it, scale it even from day one as well. So there are lots of challenges. So that's another lesson we learned from Nest. And we from day one have been building and iterating in-house and optimizing for the speed. So those were some of the things that They were really, really critical in terms of vertical integration and understanding how to go about thinking point of view.

So where's your factory?

It's actually right here in our office in Mountain View.

Really?

Wow.

Yeah. We build every single robot in-house at the moment. I mean, we obviously get parts from various parts of the world and some sub-assemblies as well.

So how many robots a month are you making in that factory?

We now have a capacity to do 4,000 robots a month.

Wow. Oh, so it's a real assembly line.

Yeah, it's a real assembly line.

Yeah. That's cool.

We pushed it really hard over the holidays and shipped 2,000 robots in December.

2,000 December. That's great.

What was like the hard part of setting up that factory? Like, was permitting really hard? Like, you know, because hardly anyone does this, like where they have their own factory in California. So like, well, you know, why don't other people do it?

It's just that people assume the conventional wisdom and just go with it. But if you go and keep asking why, so I'll give you, I'll walk you through this 3 whys in a row, right? It's like one of the conventional wisdom is you should go manufacturing in Asia or China. Well, why? Because the cost is low. Well, why is the cost low? Because economies of scale. Well, do we have scale as a startup? The answer is not, no. Then why does that framework matter at all? So once you start asking that principle of things, you get an answer and then you see Skydio. So Skydio was a good example of company that was manufacturing in one when we were doing it back then. Tesla was obviously one. So in our mind, we're like, okay, here are all these gigantic companies, uh, they're all, uh, doing things, uh, uh, outside of, uh, United States. But, but then when you look at startups, when startups try to follow, they almost always fail. And only 2 startups that I know who have sort of made a dent is Skydio and Tesla, and they're both manufacturing here. So then you kind of look at, okay, where, where, which companies are working and why are they working and dig into it. And then you So I get to answer a bunch of questions.

So it's the permitting and like, I don't know, all of that stuff wasn't a big deal?

Permitting and all that stuff is not a big deal. I think it's just finding labor, obviously finding contractors, interns to actually do the work because the Bay Area is quite expensive. Being able to get the equipment, set it up. Luckily we were able to find an industrial zone where we were able to do it, but it's just Yeah, I think the main thing, we got lucky as well. And what I mean by lucky is our production lead used to intern at Ford, and his dad worked at Ford on assembly line for a long, long time. So there was just knowledge that we were able to obtain from that. So sometimes you get lucky, and that turns out to be a good place to be.

Well, this conversation gives me hope for America, like actually bringing back manufacturing.

It's absolutely doable, I think. Um, it— and, and not only it's doable, but I think the best way to do it is through consumer products and consumer robotics. Because one of the things that in robotics most people don't understand— so, you know, iRobot, as we just talked about, not a great product in the sense that for the current world of AI, it's very inferior. And that's why— and they stopped innovating, and that's why they're bankrupt. But in their entire lifetime, I believe they've sold about 55 to 60 million robots.

Wow.

Which is gigantic amount of— yeah, robots.

Exactly.

Crazy. 21 million robot vacuums were sold in 2024. If you, even though iRobot, you know, filed for bankruptcy, I don't have data for 2025, but in 2024 they generated $700 million in revenue, which is 99.9% of startups in Silicon Valley would love to have that much revenue. And this is a company that hasn't changed product for 25 years, so to speak, right? So that was mind-boggling. Now, if you take a step back and say, what is the next highest category in robotics? Next highest category in robotics is Amazon Robotics for their warehouses at a million robots. Exactly. Beyond that, there is no scale. So we sort of a tongue-in-cheek joke that we are already the largest consumer robotics company in the United States now that iRobot is bankrupt. 'Cause Shark Ninja is still, it's public in US, but it's pretty much owned by a Chinese oligarch. So this is really it. There is no other companies out there. And I think scaling this way, which is when you get to 50 million, 60 million, 100 million robots, that's when the industry gets established. And there is absolutely a possibility of bringing it back.

Yeah, I would say home robot. I don't know how much Waymo is, but Waymo I would probably classify as a robot, right? Yes.

So that's why there's a— Waymo would classify, Tesla would classify as a robot. So specifically home robots is where we are, we're doing it, or consumer robots. I don't think there are any other scaled consumer robotics categories at all.

Yeah, it's such a wide open space and so early still. Uh, very exciting, you know, that you, you know, and I think when I think of— I mean, I, I talk to consumer hardware founders all the time and we have many on this podcast as well. You know, it's really, even though the markets can be big and like, you know, it's really wide open, very little competition, it's because so many companies die along the way, right? You know, the first 5, 10 years, this is actually the same with consumer apps, you know, which I'm working on with Tribe. There's so many of them just die. And, you know, people always get excited when they start and they're like, yeah, this is cool, this is gonna be amazing. And then like, you know, after a while the growth, you know, is there, but it's not maybe venture-scale growth or something. And then the financing drops out or they hit one of these like landmines, right? Like inventory landmines or, you know, they buy too much of this or they have a recall or whatever it is. And then, and it's just hard for startups to survive, you know, these types of things. So what's your strategy to like, you know, I mean, you keep making progress on the product, we can all see that. And What's your strategy on surviving those first 10 years? You know, because that's the biggest part of it, because you have everything else.

Thank you. I, I, I can tell you that there is no way we would've pulled off Matic if we hadn't gone through Nest experience and if we hadn't gone through Flutter experience first. Even then, raising funding even for us was really, really hard in 2024. And that's because, you know, we sort of, again, we joke internally, that we were going to ship. Our initial hope was that we will ship robot by, uh, 2020. Uh, for longest time, our, our Wi-Fi password in for a company was Vacuum 2020. And, and, you know, we shipped in 2024. So we joked that it's only off by one digit, but really that 4 years is a long, long time. And, and for a startup, that's, that's hard. So even though, you know, raising first couple of rounds was relatively easy, the third round was hard. And my learning in hardware as well as robotics is that it's the 3rd or 4th round where sort of it gets real because by that time you have to ship.

Why did it take you 4 years longer than you expected? I mean, you guys were experienced entrepreneurs. Like, yeah, that's really, you know, I feel better about Elon's estimates now.

So yeah, The honest answer is that now that I think I tweeted once that if hardware is 10x harder than software, robotics is 100x harder than software because you're doing everything from hardware, software to AI and putting it into robot and packing it in a way to deliver. But reality is still the accuracy of the systems and how to get there and how do you get to deliver that 99.99% or 5 nines of accuracy that robotics requires. And most people, even we started with this assumption that, hey, inside Mistakes are trivial. People would be willing to tolerate mistakes. Turns out that's not accurate. The reason it's not accurate is with today's AI, we collaborate. If it gives us 90% of email correct or 90% of the report correct, we're happy to take it to the last 10%. With robots, especially with floor cleaning robots or any trivial task, and mostly robotics are applied to mundane, boring, and trivial tasks. We want to delegate. You know, you don't want a cleaner that comes to your home and says, I'm going to clean every single part of your home but leave one corner dirty. If they do that, you'll give them instruction once. If they do it twice, they'll be fired. So in the same exact way, turns out that the bar for shipping the robot and getting to that accuracy was much, much higher. And productizing it is actually far harder. So in robotics, getting to a demo is 20% of the work and productizing is 80% of the work. In software, usually you can— 80% of the work is just the demo and 20% is kind of putting into nice finishing touches. So that flip really took us a long time. And I think Andrej Karpathy in a recent interview really explained very well on Dwarkesh's podcast where he said that in self-driving cars or even robotics, each 9 of accuracy takes the same amount of time, constant amount of time. So, at 90%, you have a demo. At 99%, you might have alpha. At 99.9%, you can start doing your beta, but you at least need 4 nines to ship, and then you can iterate. So, getting to that 4 nines and getting handle on that long tail is actually quite challenging. And I've been to 200 homes now, plus homes now. Homes are really, really, really variant and difficult and unique. And it's kind of fascinating how different people's tastes are.

No, that's super interesting. And you were, I mean, I feel like other people willing to like sell a product that does fail on the 3 nines, but you, you were very insistent that you wanted like really high quality.

We just thought that there was no way for us to enter this market unless we create a product that was order of magnitude better. Otherwise there was enough competition, enough noise that we weren't going to stand out. So, and, and you saw that happening where if you overpromise and underdeliver, with sort of Rabbit or some of those guys that came out in early 2024, then you— market immediately rejects you. So for us coming out and saying that this is actually smart robot, unlike the other guys, we had to show everyone that this is genuinely a step change in technology and deliver that promise. And ultimately, internally, we talk about this quite a bit, that no one actually wants robots. They want solution to their problems. And robots are just a means to an end. And the solution to the problem in this case is clean floors. So if it doesn't clean well and if it doesn't clean consistently day in, day out, we don't have a product no matter the technology inside it. So that's really the way we had pushed it.

So you mentioned current scale, right? You shipped 2,000 robots or you made 2,000 robots in December. Is that sort of your run rate right now? So 2,000 a month sort of run rate?

2,000 a month. Yeah. We are and we remain massively supply constrained. Unlike software where once you find PMF, you can just, product market fit, you can just really scale exponentially. For us, downside of hardware is that you make certain purchases and bets 8 months, 9 months in advance. And because you do that, even if you want to scale and even if you have crazy amounts of orders, you can't do it. So we literally had to sort of throttle the demand and not push really hard. after the first Cyber— after the Black Friday, Cyber Monday weekend, because everything we want, we could have built by December 25th or get to customers by December 25th. We sold during that weekend and more.

Yeah. So are you intentionally not marketing that much right now because you were trying to throttle demand a little bit to catch up, have your supply catch up?

Correct.

So it's both. We're not marketing intentionally, but that also puts us in a good position where most, I think 2/3 of our orders are all word of mouth. We don't have any CAC, we're not doing any marketing. And that was one of the principles as well, that if you look at Tesla as a successful company, they haven't done a lot of marketing and they just built a great product. So can we do that in this category to create a brand along those lines where there's just word of mouth, number one? And number two, I think Paul Graham back in the days and when we were in YC would always say that what's interesting about YC is not amount of companies that they fund, is that anyone can do that if they had a gigantic amount of fund, which SoftBank did. It's like, what's interesting is how little of a capital we deploy for the amount of companies we are funding. So it's like optimizing 2 opposite variables. So in the same exact way, we sort of realized that every single company in this category or even consumer hardware typically end up spending about 30 to 40% of their revenue in marketing. So if we play the same game, We are basically same as everyone else. So the way we need to be able to do it is can we reduce that cost down to maybe 5% or zero? And if we can do that, that's the game. And if then potentially we have a shot at winning, and if we do that, how do we do that? Well, that's a great product. So we just keep investing in great product.

Yeah, that's really cool. I think there's opportunities, I mean, there's opportunities to do a lot of sort of organic marketing, especially with your product on like short-form video. And I've been exploring this quite a bit, you know, for Tribe and, you know, I mean, it's gonna sell very well on short-form video. But even, you know, you don't have referral programs or anything like that? Like, you know—

We do. We do. We do. There are 3 pillars of marketing. One is the referral program. 2, which we haven't done very well, good job, is we have amazing set of customer testimonials. Most of them are purely unsolicited and we haven't done a good job of honestly Marketing them and putting that out there on social media. And then third one—

I see them out there. Like, I see people talking about MATIC all the time. I think like that kind of like purely organic, like people talking about it is actually amazing.

It's amazing. Yeah. So that we've seen, we haven't taken that and, you know, turned it into nice content for, let's say, Instagram or YouTube or few other platforms. And that's the second pillar that we really want to go after. And then third one is Honestly, I think content is king, right? So even for us, we need to produce our own content. So that's really, those are the 3 things we're investing in and trying to push the, get the marketing going from that perspective, but avoiding as much of an ad buys as possible.

Yeah, I think with both consumer and SMB, it's very hard to build a big business if you have to like rely on buying every user. I think having that discipline where like, you know, people have to recommend this thing and that has to grow organically is powerful. It's the same with Mercury. Even today, we have like 60% of our users are kind of organic and word of mouth, which in fintech is extremely rare, but I massively prefer it. I think then you can really focus on building that product, building an amazing customer experience. It's like everything aligns rather than like, yeah. I also actually think when you buy users, you get a worse kind of user. You get someone that cares about some deal. They're not as bought in to being real long-term customers.

Exactly. And especially this, this, that's a critical difference in the early days, right? In the early days, you do want someone who's going to be your early customers, who's going to buy into it, who's gonna have a chance and turn that into a community of really good followers and recommenders. And if you can do that, that's an opportunity. And for us, we, the easiest way to say it is it's not like we didn't think about and try some of this experimentation, but I kind of joke that we, we kind of put the worst of everything. So we picked consumer, we picked hardware, and we picked probably the unsexiest category in the robotics space that we can pick. And we launched first time in April of 2023, and frankly, no one gave a shit. And then we launched again in November of 2023, and again, no one cared. And the only time we sort of started seeing is when we started shipping and when people started experiencing it. And it really took off when Wired gave us a really good review, first 10 out of 10 in May 2nd, 2025. So about 9 months ago. So we literally had to work hard to make robot vacuums cool again. And that's really the honest answer.

I think it is because everyone was burned from the category. Like everyone kind of assumed it would be like the next shitty thing in the category.

Yes, exactly. There's so much apathy. There is so much, yeah, it's another robot vacuum. So much overpromise and underdelivering category that we had no choice.

We touched on this, but you seem kind of skeptical of humanoid robots. I mean, we've had like, I don't know how much in investment, maybe like $10 billion went into human robots like last year. Like, what's your— how do you think that plays out? Like, do you think they're just not gonna deliver or it's never gonna work? It's never gonna get to the Five Nines or like, what are your kind of like, kind of prognostications on this?

Yeah, I think it's really, it will follow, for lack of a better way of saying it, it will follow the same trajectory that self-driving cars are following, which is there is a lot of excitement, there is a lot of hope, But the gap between what we want to deliver and where we are is much larger, number one. Number two, even if you have a really great humanoid, what is the product? I think as a consumer, why would I buy it? Would I buy it for laundry? Would I buy it for dishwasher? Would I buy it for this? What is the right price point? Is $10,000 a right price point? Now, in my mind, and the reason we went with a camera-only version was also because, and I touched upon it, that we wanted to make sure that we have unit economics and make You know, it can make it profitable. So if you take a step back and think about it, there is literally zero ubiquitous consumer electronics device that are priced higher than $2,000.

Apart from cars.

Apart from cars, exactly. Beyond $2,000, there are prosumer categories, but cars are the only thing we pay for. Now, cars have been around for 100 years, 100-plus years, proven utility. No one questions why you need them. Yet a $10,000 car remains a considered purchase, and we do think twice, 3 4 times before making that purchase. It's not an impulse. And the amount of value that car can give you is gigantic. So what is that value equation on a flip side for that to happen? So there is a gap of making technology work, and then how do you productize it?

Yeah. So it can't just be like, it does my dishes. It needs to like do my dishes, do my laundry, do my cooking, and make the bed. And like, it needs to do like enough things that you can be like, wow, I'm, I'm willing to spend $10,000 or $20,000.

Yeah. And then if you buy a $20,000 robot, let's say it comes to your home and says, I'm gonna do all the dishes except for the utensils. I don't do utensils. Or I'm gonna do all the dishes except for cups. I don't do cups very well. I'm just giving an example, but there is a variance. And if it does one of those things, it's really hard. Let's say, you know, an example of just making a dinner table for Thanksgiving. If you have a family of 10 and you make that dinner thing, you're probably thinking about 100 different objects and variances. And if you break one of Once, them, like, you know, if you have a— and I just kind of think of this in my mind, that imagine a maid coming into my home, and if that maid, uh, if that person broke one glass every time they came, would I keep them around? And what is my patience on this?

It does seem like consumer humanoids are tricky. What do you feel about kind of industrial humanoids? Like, that seems like a simpler problem.

That seems— that's definitely more plausible, and that's where the opportunity— I think it will get started. It will have to start on an enterprise side, and, and I think that's where it's just a matter matter of what, what, um, use case you're going after and how much accuracy do you need for that. I don't think there is any proof at the moment that generalization or manipulation generalizes well yet. Um, so there's a lot of research to be done, fundamental research. Uh, and, and as an industry, you see this, right? Uh, in 2023, robotic foundation models were all the rage, and then, um, simulation was all the rage for a bit, and then, uh, last year was reinforcement learning, and Now a lot of robotic companies are moving to world models and then those are becoming the rage. And that's because the accuracy that's needed in order to ship anything, no one really knows where it's gonna come. So there's just a, they're just kind of going from one option to another option to another to see.

We're basically in like a research phase right now.

Very much in research phase.

So some feedback that I have on the, on the product.

Yeah.

One thing you could do to improve it.

Oh yeah, let's see live feedback.

Yes.

How does it suck, Raj?

Yes. No, no, it's more of a feedback on the experience. I think your, your post-sale follow-ups and like customer support, customer success could be a lot better. And that would improve your organic word of mouth. Like your follow-ups, like your onboarding. You know, you could look at Superhuman as someone who's like built a much better email product that does really careful onboarding, like really kind of tailor-made. Like we bought the Manage, we tried it a couple of times, and then we sort of didn't get into the habit of using it often.

That's great feedback, yeah.

You know, we have to get in the habit. Like every new consumer behavior behavior needs you to build a habit. Every consumer electronics in particular, right? So because it's not a Nest, Nest you have to use it. You have no choice. Once you install it, there's no other thermostat. But the Matic sits in our corner and I just haven't gotten the habit of pulling out my phone, pressing the app, and doing that. And it's just like, I know there's a power button I think I can press on there as well.

Yeah.

I didn't even get in the habit of doing that. And so we just didn't get in I just don't have the habit of using it, right? And we also have regular cleaners, but I would like to use it more. And so anyways, some kind of human-to-human onboarding I think would be good for this type of product and follow-up. If someone isn't using it, you guys have all the data, right? So if someone isn't using it, why?

Raj, I don't know how common your use case is. I feel like most people would go like, oh shit, I need to vacuum, and then they'll press the button. Whereas you're saying like, oh, I could just wait for the cleaner. Like that's—

but most people have cleaners, don't they? Like in the Bay Area?

Maybe in the Bay Area.

There is a— so you're spot on actually. One of the things we are doing, what we see is that when people add a schedule, which is the robot to just put on some sort of daily, weekly schedule, the moment they do that, usage gets quite a bit. And then when robot doesn't run, it's very easy to notice. It's very easy to know that, oh, my home hasn't been cleaned. And that sort of becomes a habit of just walking around in a clean floor all the time. So one thing we are doing is default schedule, default things. But really honest answer on this for a long time, and I think I mentioned a while ago on Twitter somewhere that we've only shipped about 50% of what we specced out as minimum lovable features back in 2019 that is still there. And the goal that we always had was that robots should actually just 3, 4 times a day take off, scout your home. If it finds dirt or dirty spots, just clean it. Over time, understand which part of your home gets frequently dirty, all the time, high traffic areas, for example, and just kind of add a suggested schedule or add it on its own. So even this habit of cognitive load, we don't even want to do it. So the goal that we've had, if the robots do a really, really good job, then they would be able to allow us to do 3 things, right? Which is live in a perpetually clean home with perpetually clean floors. Obviously not do it ourselves, that's the second thing. But ideally, third one is not even think about it. It just gets done for us. We don't even have to worry about or tell it through app or things like that. And then so default schedule is one, doing scouting and cleaning on its own, dirt detection, that's the second one. And third one is we are shipping voice interactions very, very soon where you can just say, hey, Matic, start cleaning. And then the friction is even lower instead of doing it. So that's the third one that we're shipping very, very soon. But all great feedback, you're correct.

Yeah, and you mentioned scouting. Scouting is really interesting to me. me, like actually just telling me about like, hey, my kids' rooms are pretty dirty, like probably could use a tidy up, you know, like that would be actually pretty helpful. So send, actually one thing I think you want to do better is like have more of a back and forth relationship with your consumers. Like, so like notifications on like the app and maybe emails, right? Like you have to like have this constant communication because otherwise Yeah, yeah. So yeah, the voice is great. I think the voice is great, but there needs to be more also push. I think you know best practices of using Matic, like you know Matic, and just like all these things. So anyways, it sounds like you're thinking about it already, which is great.

Yeah, all those things. We are just just we're just barely getting to that, and and and I don't know if you guys know the history, but by the time you probably got the role. We were already doing edges and toe kicks, but when we shipped about a year and a half ago, we had to ship in 2024. It wasn't even cleaning edges of your home rooms. It wasn't cleaning the space under kitchen cabinets. It was really the half-baked product that we actually ended up shipping. But luckily what it was doing was doing enough, good enough that people were happy with it and it worked out. But initially, so there was just basic features that we had to deliver first. So now we're just kind of getting into this place.

One thought I have, I mean, kind of inspired by the Matic, is like, you try to anthropomorphize or your robot, it looks like a pet, right? Like it's not anthro exactly, but yeah, it's meant to look like an animal or feel like an animal, a robot pet. What do you think? I mean, I think there's a whole market there for robot pets, like a fun robot that just walks around your house and does funny things. things occasionally and it can have a different form. Of course, when you have kids, I mean, I would rather have a robot pet than a real pet. I don't want to have cleaning up all the stuff, but I'd rather just have a fun little thing that just does random stuff. And I just wondered, that seems like there could be a market that exists that people just buy these fun pets.

Yeah, absolutely. I think the toys, AI toys and AI pets to entertain kids, those are all the places where robots are going. We had 2 insights. So the reason Matic looks the way it looks is, as I mentioned, I tried all the robot vacuums and I had a 3-year-old daughter back then and dog, and both of them would every time freak out with any of the disc robots. And it sort of dawned on us at some point that we as humans are genetically predisposed to being scared of anything that crawls on the So adults get over it, but kids and pets actually don't understand that pace, and they just remain scared for a long, long while. So that's where the robot being taller and kind of looking like a pet was this idea that it feels a little bit more friendlier. So there is an element of that. And then the second thing we— I fundamentally believe, and this is where AI needs to get much better, is that our pets and any sort of intelligent being, we're sort of— our reactions are affective. We do certain things repeatedly, but not 100% of times. Even your dog, you teach it to shake hands, and there will be a time it will say, nope, I just don't wanna shake hands, and I'm just not gonna follow that instruction you gave me. So you gotta have this personality, because if you can make any pet do same thing over and over again, it gets boring and it becomes a machine. So this entire engine where emotions and reactions and things are ephemeral and unpredictable, It's another AI work that I think people are just starting to get into. So when that happens, there will be potentially a huge market.

Yeah, I mean, I think they're— I mean, TikTok does this really well, actually. You open TikTok, it's a roulette. You have no idea what video is gonna be, right? And they do that on purpose. They use Pavlov's dog kind of thing. It's like they have random incentives, right?

Yes, yes.

But that can be programmed into a robot pet. Like, you could have totally random stuff. You can have software updates and like new things.

Exactly.

Coming in and like, and constant streaming actually of new actions. So anyways, I just think that if I think of home robots, you know, what you've built is very functional. All right. And, but there could be something that's actually fun and like just fun all the time, you know? So it's maybe a new product line for you.

Yeah. For us, as I did mention, we always believe that we need to actually earn customers' trust first. And we are better off doing that initially with the first product.

That would be a huge pivot for you. Vacuum cleaner to toy.

Vacuum cleaner to toy would be for sure. No, at the moment, at the moment we're just focused on this. There are second, I mean, with parents we do get lots of emails that, hey, can your robot do this? Or can you build a robot that does this? There is toys aspect of it, but we get more emails around, if you just, if you just gimme a toy cleaning robot.

Toy cleaning robot.

Yeah. I'll probably pay you thousands for it 'cause I'm just tired of picking up that Lego or picking up the doll or my—

Yeah.

Putting it all away. Yes, exactly. So there is an element of that that's happening, which is really interesting. And then we also get a lot of customers saying that, hey, what if I can just put this robot in my parents' home? They live in Florida and I don't even want them to control it. I'll control it remotely. But then can you just give me 90-second time-lapse of their day and just tell me that they're okay? If it's utility-oriented products—

No, that'd be great.

Yeah, and if it's utility-oriented product, they don't feel like it's a Big Brother thing. So there is interesting asks and things that are emerging out of it where— so our goal has always been that let's go where customers are pushing us to go.

Yeah, that's what I love about founding companies is that like you learn so much in the doing of it and everyone who like tells you like, oh, this is never gonna work, like, you know, like they're just missing of all the stuff you learn while you do it, you know? I mean, like it just, it's so superficial to tell somebody who's starting something, oh, that's not gonna work based on like the little that I know, you know? It's like, but that's their gut reaction when you hear of an idea.

Yeah, we just, we had a lot of ideas actually back in 2017. We did. And I wish more entrepreneurs would do that. We just kind of created our own set of rules of engagement on what we wanted out of it. And that helped us narrow down the idea a little bit better than anything else. Like we just wanted to make sure that it was an existing category, that there is a clear-cut business model. They were not educating customers, consumer. We didn't want to do SaaS or enterprise. My first job was back at salesforce.com in 2005, so I was very early into SaaS and realized that I just don't get excited about selling to other businesses. So those were some of the—

Same here.

Yeah, so those were some of the things we already knew and that helped us narrow down this problem.

I think when you're doing consumer, it's really important to not take that much risk. in areas you don't need to take risk in. So picking exactly what we did for Tribe, we picked an existing category, right? And existing business model. You want to really, really minimize your risk on other stuff because especially consumer habits are really hard to change.

So yeah.

Very hard. And even then I was optimistic. We tried to launch, even the first time we launched Matic, we tried to do hardware as a service or robotics as a service model. That was like an organ rejection. That was like, we got so much, such a negative feedback.

Like a subscription service?

Yeah, it was a subscription service. We're like, what if a robot comes to your home and it's just $75 a month and whatever, and it just cleans for you, you don't have to take any risk. And when we have a new version of the product, we'll just deliver the new version for you so you don't have to ever worry about it. And we thought that was such a great deal because why take a huge expense Massive risk, but turns out people want to own whatever is inside their home. There's this idea that I just don't want to have anything that is rented. So this idea that, what do you mean I can't own the robot? What do you, you know, why subscription?

Well, I also think that a lot of subscription services have kind of burnt consumers. So I think there's an element of like lack of trust in that business model.

100%. 100%. For us at a higher price point, what we did was we equated it to cars, that cars you can Yet most people still prefer to buy. And I don't know why that's the case. Maybe there's a deeper reasoning. I haven't looked into it, but at least in our side, we thought that there was just this desire. And—

I mean, whenever someone charges you something monthly, you just feel like you're being screwed. Like you feel like at the end of the day, like they are extracting margin from you somewhere. Like that's like, that's kind of your assumption versus like taking the one-time hit.

That's correct. That's actually, you know, you're 100% right. And we saw that. As I said, some lessons we learned. Like, one of the things we did know was on Nest cameras, the number one detractor was the NPS score for our product. But yeah, so subscription was not something that we were doing, but we were trying to still make it work for Matic, and turns out it didn't fly at all.

Awesome. I think we are at the end of our time. Really, really good chatting, Mehul. Congrats on the success. Seems like I need to go buy this robot. $1,245, no subscription fee. Sounds perfect.

And then both of you are welcome to come by and see how we're building the robots. It's actually quite fun.

Yeah, I kind of want to do like a family tour of a factory. I think that'll be kind of fun to let the kids You should do a thing, like get a bunch of families together and like, this is how you build—

Let's do it. I'm happy to have you guys over over the weekend if it's easier. But yeah, come over.

Thanks everyone for listening. Please leave us a review on whatever channel you're listening on. Makes a big difference to the algorithm and to the success of the show. And see you all next week.