Everyone wants to invest in robotics

At Foundamental, we invest globally in businesses redefining project-based industries. We make no investments outside of the project-based sectors we feel dedicated to.
Robotics has been my main focus ever since joining the firm, and I wanted this article to be an extreme synthesis of the mental model my brain invokes every time I meet a robotics founder/operator or I talk about the space with a fellow investor.
I’m going to write much more in depth about business models, markets, development styles and manufacturing.
This below is “just” the 30,000ft view.

From a sharp stone tied to a stick with buffalo sinew, to autonomous haul truck used in large scale mining operations, a timeless way to see tools is a way to turn OpEx (recurring cost, tied to human labor) into CapEx (one off expense, outcome lasts for some time):

• Spears made men more efficient at killing the prey: two men with two spears are much stronger and mightier than four men with sticks
• Autonomous haul trucks reduce the headcount needed to run mining operations. Mines are remote, long commutes are the norm (in this remote Australian mine, most workers fly in and out) and hauling material in burning hot weather is not a trendy job last time I checked

Tools make humans more productive at doing things, and the pace at which happens increases when the thing is something nobody wants to do (the good old Dull, Dirty and Dangerous) and per economics 101 those that want to do it are just very expensive to hire.

An example of modern tooling pioneer is Black & Decker, the century-plus-old American behemoth (now called Stanley Black and Decker after the 2010 merger). The company patented the first portable drill in 1917 and the first electric screwdriver in 1923. It is impossible to even fathom the labor productivity efficiency that these two inventions alone brought to the world (both on the jobsite and while struggling to assembling new IKEA furniture).

Tools + computers = robots

Perhaps the mightiest tool (not the most important, as it could be fire, modern agriculture or penicillin) humanity has recently invented is the computer. In broad brushes:

• It enables perfect replication: Unlike physical tools that degrade with use, digital information can be copied infinitely without loss, allowing knowledge to scale without degradation.
• it automates repetitive processes: By encoding instructions as algorithms, computers can execute precise, consistent operations billions of times without fatigue or error

Even before the invention of the transistor, we humans understood that computers are perfect "brains" for physical tooling that can greatly enhance our productivity.

In the late 1940s, Numerical Control (NC) machining allowed for the first time to precisely and repeatedly control a Cincinnati Milling machine by specifying instructions through punch cards to a quite primordial vacuum tube computer system. NC was a precursor of CNC, that was patented at MIT about a decade later and became a backbone piece of equipment of the human industrial manufacturing machine.

In the context of physical world tooling, the great things about computers is that if you pair them with the right pieces of hardware to intake instructions from a human (user interface) and physical data (sensors) as well as exercise physical action (actuators), you get a powerful combination of technologies that can augment, and even replace, human labor. The combination above is commonly called a robot, and it's yet another tool that humanity is developing in the quest to turn more of its OpEx into CapEx.

Robots' adoption in physical industries (on Earth) has followed an easy-to-summarize yet incredibly nuanced evolution that goes something like:

1. Robots appear on assembly lines: between the 1960s and the 1970s, industrial robots started to appear in large automotive manufacturing plants. General Motors ended up being the pioneer of this wave with Unimation's Unimate robot famously automating the unloading of die casting presses and spot welding (now one of the most widely automated tasks in high-volume manufacturing plants).

2. They expand into Warehouses: At the dawn of e-commerce, robotics started permeating the logistics sector with automated guided vehicles (AGVs) and sophisticated picking systems composing the orchestra of efficient movement of modern fulfillment centers. In 2012, with the acquisition of Kiva Systems, Amazon set out to become the biggest global deployer of robotics for logistics.

3. Deploy at decent scale in select use cases in food production: picking up in the late 2000s, milking robots are among the most widely deployed robotic systems in the world, around which many of the largest intensive cattle farms in the world are being designed. Drones and other UAVs are also changing the way farmers collect data and manage treatments.

Note: The big one that's missing from the list above is robotics designed for space missions. Arguably this is the true edge of robotics (machines that need to operate in extreme environmental conditions and humans are light-minutes away, unable to react and command from a distance). In case you missed it, Perseverance and Ingenuity (the first flying robot to cruise through the skies of another planets) have redefined our understanding of Mars' surface and atmosphere.

Building and servicing with robots

One obvious area for robotics to make a significant impact is that of construction and its many trades. This is something that we spend a lot of time thinking about at Foundamental (and from first principles, it's quite easy to understand why):

• Safety: construction it's the mother of Dirty, Dull and Dangerous industries (sitting on a two-seat throne with mining): globally, construction is cursed with high rates of injuries and fatalities from falls, struck-by incidents, and repetitive stress. Robotics may just be the enabler for higher human safety.
• Throughput: robots are fast, don't unionize and don't take breaks. With most construction projects suffering from cost overruns are delays, this is music to developers' ears. Although it may not be possible to substitute entire crews with robots, robots can become part of a crew's standard tooling and increase its productivity massively.
• Consistency: by design, robots are not only moving material and completing physical tasks, but also verifying the quality of their work through built-in QA/QC. This is invaluable for construction, where expensive quality control needs to be omnipresent to reduce rework and delays.

Despite this first principles thinking, however, robotics has not yet taken over the world of construction. Judging the progress of construction robotics through the number of deployments of the most common robots is a disappointing exercise. Respective to the massive size of the industry very few construction(-adjacent) robots deployed on jobsites to do physical work.

Two notable exceptions may explain the reason behind this slow progress: robotics for data collection has picked up faster than any other use case in the space.

• Drone-based reality capture is an increasingly common way to track progress in construction sites, inspect assets and certify outcomes. Companies like DroneDeploy are powering the software glue that lets users track and understand insights from this new data, while hardware developers like Boston Dynamics, Gecko Robotics and Field AI build (or repurpose) physical platforms to improve autonomy and the quality of data collected.
• Automated layout robots print exact digital floor plans directly onto construction surfaces, eliminating manual layout errors and accelerating the critical but error-prone process of transferring designs from blueprint to building site. The best product in the market is Dusty Robotics', but players like HP (who is leveraging decades long printing expertise) have entered the space. Rugged Robotics and CivRobotics are among the co-pioneers of this product space.

What both of these workflows have in common is that they involve little-to-no physical action. Drones are intelligent scanners, and Dusty's robots very conveniently print on concrete, but they don't pour/break/grind/polish/drill it.

For robotics that serves the purpose of exerting some physical action to produce a certain physical outcome for a specific trade, adoption has been incredibly slow with very few robots crossing the 100 deployments threshold.

Workflows and business models

After meeting dozens of construction robotics companies, as well as some of their customers, key reasons behind the strong headwinds to adoption seem to be:

1. High variance of the workflow (and its outcome’s specs and requirements) that the robot has to solve for pushes robotics teams to optimize for their early design partners’ needs only to later find out that substantial modifications are needed to serve each successive customer.
2. Somewhat related to the point above, a lot of these robots are betting on a fundamental change in the way customers operate, as opposed to seamlessly integrate within their workflows. A big example is that of robots that 3D print concrete trying to sell into traditional General Contractors.
3. Long sales cycles, where the bottleneck is often driven by the robotics firm’s shipping velocity (how quickly and cheaply can I iterate on my product to ship the features that this customer needs?) and manufacturing velocity (how quickly can I get a production ready unit in the customer’s hands after a short pilot?)

For a robotics firm that raises venture capital to fund its initial development, the sum (or sometimes even just one) of the above sources of drags, can be a killer, which brings us to the condensed insight on what makes or breaks such a firm in this market: choice of workflow, speed of development and manufacturing velocity are the main factors that coalesce into a robotics company’ capital efficiency and its ability to get out of the R&D valley of death to either (1) show real traction to raise subsequent rounds or (2) hit profitability.

However, the above is just a prerequisite to avoid early death. As early stage investors hunt for generational outcomes, the spikes to look for become more specific: in construction, track record is the most sought-after asset and robotics businesses are not exempt from having to compound on it:

• Narrow, well-defined workflows are easier to compound one’s track record in. Overlooked and labor-intensive tasks are the best to start with.
• A perfect task for robotics has a clear P&L item in the customers’ books. Drilling holes or hammering nails do not have dedicated budget lines. Installing curtain walls (like Raise Robotics does) or replacing roof shingles (Renovate Robotics) do.
• Very few narrow tasks can single-handedly support very large outcomes, especially in markets where we expect competition and the regional nature of construction creates barrier to entry (one of the few in that category is Monumental). We look for robotics that can recombine in a product suite that, over time, grows our penetration rate within the same (or adjacent) customer base across a variety of tasks.

Physically, not all workflows are primed for robotics to be successful. As Patric codified it those that we think are the best fit have two main features:

• Buy-readiness, as a measure of the market fit of the work that:
  • is already bought as-a-service or via contractors regularly with established processes
  • has a clear budget line (and isn’t nested in other costs)
  • does not raise questions about norms, standards and durability
• Success-readiness as a measure of feasibility of getting an outcome that:
  • needs to work functionally, but not deliver perfect optics
  • has a high ratio of homogeneous surface to edges/corners
  • requires a repetitive sequence of functions
  • has at least 100% value uplift from un-assembled to assembled components (or a massive recurring ops/maintenance cost)

The choice of workflow and the market dynamics associated with it inform the business model.

If the workflow to automate is one that our customers already purchase equipment for, they will likely want to purchase our robot to replace old equipment with our software-driven new-gen version of it. Concrete power tools (grinding/polishing machines) and compact excavators are a prime example of this category.

If, instead, our customer’s ROI comes entirely from replacing labor costs with our robot’s work, we will more easily fit into their OpEx bucket and leasing the robot becomes the best option for them. Placing PV modules in utility solar parks was something solely done by humans before players like Luminous and Cosmic came along, leasing their robots to Solar EPCs.

Leasing is a great option in other two macro-cases as well:

1. When CapEx tied to that piece of equipment is a source of financial stress for our customer. It is the case of autonomous tractors and Sabanto, who is relieving farmers from the heavy upfront expense of purchasing large tractors by leasing autonomous versions. In this case, it is not necessary to fit within replacement cycles of this equipment, but ROI can be proven on a variable cost basis.
2. When customers already lease that equipment from leasing companies like United Rentals. This is likely to be the case for low utilization (or seasonal high utilization) equipment like aerial boom lifts (or asphalt paving machines in northern regions). In this case, it can look like an easy swap for the customer: traditional equipment for (semi-)autonomous equipment.

Hidden labor-intensive workflows are a great fit to develop robotics because chances are that the legacy equipment used for these tasks is old, clunky and hated by its users.
However, hidden workflows often times coincide with small markets. And even if the size of the market allows for a large outcome, misalignment with purchasing processes and the customers’ structurally low margins may be an extra source of friction to adopt robots within a trade crew.

These, I believe, are opportunities for starting Outcome-as-a-Service firms leveraging robotics in the back-end. Patric and Shub, GPs at Foundamental, have talked about this extensively.

The playbook is quite straightforward:

• Take a fragmented, low NPS, poor-CX trade in construction
• Automate as much labor-intensive tasks within that workflow
• Stand up a humans+robots crew with much higher labor productivity than a human-only crew
• Deliver a great outcome, at best-in-class margins, with never-seen-before customer experience

If you want to simplify the concept even further: don’t sell bricklaying robots, just lay bricks with robots.
As the construction industry is deeply project-based and hungry for services (more than it will ever be for products) this is a way to substantially reduce static friction when going to market and to faster compound track record.
Another great feature of this model is that you don’t have to ship a perfect robot to start: you can progressively automate more and more of the workflow while still being able to deliver projects with progressively less human input.
You can even leverage robot simulation data for precise quoting and avoid costly mispricing.

Becoming a subcontractor, however, has a cost and its operational complexity, but I’m going to cover the entire playbook in a dedicated article.

Summarizing the thesis so far:

Some core beliefs make us incredibly interested in construction robotics:

• Robotics continues humanity's millennia-long pattern of turning recurring labor costs (OpEx) into one-time capital investments (CapEx), especially for "Dull, Dirty and Dangerous" work. Successful robots will be eventually perceived as simple tools.
• Construction is prime for robotics disruption due to safety issues, throughput needs, and consistency requirements

Yet, adoptions remains low for robots handling physical tasks and we think we know why:

• High workflow variance forces companies to over-optimize for early customers, making scaling difficult
• Companies try to change customer operations instead of integrating seamlessly into existing workflows
• Long sales cycles bottlenecked by slow product iteration and manufacturing velocity

We think that :

• “buy-ready" workflows: already purchased as services, have established budgets, don't challenge industry norms
• “success-ready" tasks: functional over perfect aesthetics, high homogeneous surface ratio, repetitive sequences, high value-add

Once chosen the trade, market understanding informs the business model:

• Equipment replacement → sell robots (eg. concrete tools, excavators)
• Pure labor replacement → lease robots (eg. solar panel installation)
• Financial stress relief → lease (eg. autonomous tractors)
• Already-leased equipment → lease swap (eg. boom lifts, seasonal equipment)
• A jolly strategy remains becoming a subcontractor (Outcome-as-a-Service) delivering high quality outcomes at high margins with unparalleled customer experience

This is a radical work in progress. Each new encounter with brilliant entrepreneurs and customers has the potential to develop our taste, sometimes marginally, sometimes drastically.

What’s next?

If you made it this far, there’s a chance you would want to know what I’ll write about next. There is so much to unpack when it comes to robotics for construction, so if you subscribe, you’ll soon find in your inbox my rants about:

• Development styles
• Manufacturing and financing a robot fleet
• Dealerships and after sales
• The importance of inventory for GTM
• Becoming a robot-powered subcontractor
• The art of building deployment teams
• and much more…

‍