Investing in his audacious vision for a robot-powered world
Seth Winterroth is a Partner at Eclipse, a team of operators and investors building companies that transform physical industries critical to people’s lives. When he’s not coaching the next generation of revolutionary robotics teams – companies like 6 River Systems, Third Wave, and Wayve – he’s hanging out with his family, hiking, or skiing in Lake Tahoe.
I've been a Partner at Eclipse since the inception of the firm in 2015, but prior to that, I worked at General Electric (GE) for three years. It was there that I had a front row seat to how this large industrial conglomerate – with hands in different physical industries responsible for a vast majority of global GDP – was adopting exciting new digital capabilities to evolve into the 21st century.
During my stint at GE, I worked in the robotics space, in precision and metal additive manufacturing, and had some exposure to industrial IoT efforts. It not only shaped my understanding of these nascent technologies, but also made me aware of their broad implications.
Seeing this wave of connected compute systems proliferating into the physical world and believing that these new systems would redefine how our world operates made me really excited to zero in on robotics. I saw that this nascent technology could enable legacy industries to be more efficient, more precise, and more transparent – and do more with less capital.
In my role at GE, I was exposed to projects that leaned heavily on the startup ecosystem to deliver the capabilities we needed. I became fascinated with small teams working in concert to solve really hard and audacious engineering challenges. Ultimately, these partnerships helped GE shift its thinking about the industries these startups had been playing in for quite some time.
In parallel, a lot of the companies that I thought were interesting were really struggling to raise institutional capital. Great founders were tackling really hard engineering challenges and going after big market opportunities, but couldn’t get investors past the perceived risk around this space. Robotics didn’t have the traditional Sand Hill Road model that consumer-facing social and mobile ventures at the time did. This was the early 2010s – everyone's looking for the next Snapchat, Salesforce, or hot enterprise SaaS company. These industrial tech teams were too off the beaten path, and they were selling into industries that were perceived as too hard to sell to – there were hardware elements and complicated product designs, they needed to navigate supply chain challenges, sales cycles were too long, or customers were too difficult to convert.
And I just didn't think that any of these rebuttals were true. I didn't think the perceived risk around those companies was as acute as the rest of the allocators of institutional capital did. My experience at GE helped me see that there was a tremendous amount of opportunity here to completely transform these legacy industrial markets.
That’s what ultimately inspired the inception of Eclipse – knowing that there was something missing in this ecosystem. With Eclipse, we decided to blaze a path and say, “Hey, we're gonna invest significant amounts of Early-Stage institutional capital in teams solving these types of problems.”
Zooming out, I think most technology categories follow a similar cycle where it labors in obscurity or academia or research-oriented settings for decades. And then, some enterprising founders put together a team and get enough capital and early customers to build the first iteration of a successful private enterprise that breaks the mold. This is the pivotal event that demonstrates that this technology can break out of the historic academia research setting.
I think robotics is no different. For me, the instantiation for modern robotics was the Kiva Systems acquisition in 2012. It was the acquisition that launched a thousand robotics ventures, and really demonstrated that this field isn't just about research. This isn’t just the philosophy degree of the engineering sciences – it has very practical applications and a real impact on enterprise. I felt very inspired by that event.
Image courtesy of The Verge.
We've seen many of those companies that launched after the Kiva acquisition fail over the last 10 years. Because it's hard. To launch a new robotics company, you have to do everything yourself – from silicon through the application layer. You need a really broad array of functional engineering, operations, and sales capabilities within your organization. It’s just really not easy to do everything at a really high level – to get a high-quality, performative, safe, and reliable solution inside customer environments that really moves the needle for customers.
But on the flip side, you saw some companies break through those barriers, and that inspired more companies. 10 years later, there’s a bit of a groundswell. It's getting easier to build these companies now – the customers are aware of and understand what the value of these solutions are. In fact, they're clamoring for more of them. Hardware elements have gotten higher in performance and lower in cost. Industrial-grade open source solutions are becoming available, making it possible for small teams to develop more effectively.
This progression is no different from any other technology category – we gradually get more customer awareness and interest, and have access to tools that are both better and cheaper. That's where we're at now in this category. There are real companies today delivering delightful robotics solutions to customers and generating tens and hundreds of millions of dollars of revenue. There’s a sea of problems across these industries that make up the vast majority of global GDP. There’s a lot of opportunity for autonomy solutions to have material impact across these industries.
I see companies trying to commercialize too early. Oftentimes, companies don’t do enough work on system design to really understand whether what they're endeavoring to build is a complete solution for their customers. They get tripped up at the commercialization or product introduction stage by system design choices that they could have avoided if they'd done more preparation.
Another common mistake that I see from seed and Series A companies in this ecosystem is not engaging with customers to really understand their problem and build the right solution. It's not like software where you can go back and just write some new code or push an update. You really have to get the granular system design characteristics correct before you start building and certainly before you start implementing with customers. It’s important to nail that and spend a lot of time on that, on the front end. With hardware, mistakes are just costlier – both in terms of time and dollars.
So many companies just don't spend enough time in the customers’ environment. They think they get it – they think, “This is the engineering spec that we're gonna go execute. On the other side of this development effort, there's going to be a product that delights the customer.” And so they go pencils down and design, and buy parts and spend money on components, and manufacture a system. Then, they show up on the customer site with key characteristics of the system design that don't meet the needs of the customer – it's not fast enough, it’s not big enough, it can't do xyz thing, it doesn't have the right means of integrating with the human operator, etc.
I'm biased, but there are a bunch of companies in our portfolio that I think are doing quite interesting work.
Third Wave is a great example – the forklift is the universal workhorse in supply chain logistics distribution centers. It's not a new device. And 80% of the operating cost is the human operator – human operators who get paid anywhere from $60,000 to $80,000 per year, and work across multiple shifts. Not only is that a lot of money, it’s very difficult to find these operators, the turnover is really high, and the work is both monotonous and unsafe. And so it's just a really great opportunity for automation.
I’m also excited about Foxglove, and the customers the team has brought in. The roboticists working on cutting-edge perception, control prediction, planning, navigation problems, and trying to push the state-of-the-art, shouldn't also have to stop what they're doing and piece together the infrastructure that enables them to develop more effectively on an incremental basis within their teams. One of the things that pushes product development is cycles of learning and how fast your cycles of learning are.
If your organization's development cadence can only move at the rate of your infrastructure — and the infrastructure's all homegrown and built from scratch by people that have never really seen scaled infrastructure because they're so specialized in these specific areas of engineering — that's not a really great recipe for fast cycles of learning and iteration. We need better infrastructure, better tooling that expands across the development requirements of complex, highly diverse, functional organizations, in order to accelerate the cycles of learning.
One of the companies I work with in the AV space spent a whole month trying to debug this one problem they were having, and finally figured out that it was a camera on a single vehicle that wasn't calibrated correctly. Leveraging Foxglove, that’s a problem they could have solved much more effectively. Multiply that by infinity around the types of problems, edge cases, and little things that you encounter when you're developing these types of companies – enabling teams to work more cohesively with a tool like Foxglove is a huge driver of more rapid engineering execution.
At a high-level, I think the industry will focus on streamlining the movement of goods, and then the movement of people, in a variety of environments. Supply chain robots, autonomous aircraft, and self-driving cars are all good examples of this. How can we automate and streamline the way we move products, in various unit sizes, from point A to point B? How can a company like Target or Walmart hold significantly less inventory, because they can trust that they have perfect visibility and efficient movement across their supply chain? How can people get to a destination more efficiently in a way that is not so carbon-emitting? I think you'll continue to see more answers to that question come to bear across the industry.
In line with that, I think manufacturing is going to experience the biggest upside. There's a huge trend around reshoring, and you can't bring manufacturing closer to the point of consumption without solving the very obvious labor challenges associated with it.
Photo courtesy of Toyota.
All these improvements – across manufacturing and autonomous transportation and supply chain – will reduce friction in people's lives by allowing them to do what they want to do, have the service done they want to have done, transport themselves and their goods in a way that's efficient, effective, low-cost, reliable, safe, and carbon-neutral.
Also, my definition of a robot is pretty broad. It's anything that sees the world, computes to understand, and then either takes action or informs action. In my mind, anything that takes data in through a multimodal sensor, runs software on top to extrapolate information from that sensor data, comes to conclusions based on that data, and then acts on those conclusions is a robot. This includes edge devices like smart cameras and IoT devices, so I think there are a lot of cool applications there as well.
Focus on the highest-value applications possible. I think a lot of people fall into the trap of going after incremental value generation solutions, but the big potential upside of your idea – the amount of consequential value that your application can deliver to real-life customers – will convince people to join your company and convince investors to invest. Oftentimes in this category, you need to raise multiple rounds of funding before you even have something that operates with a production-level SLA, so make sure you're pointing at an application that's got some heft to it.
Also, don't fall into the trap of trying to pilot with too many customers too soon. Pick the best one to three customers whose environment or operational domain is indicative of as much of the rest of the market as possible and just get it right with them.
