In almost all discussions we had at Foundamental in 2019 with corporate partners and our newest LPs, the topic of robotics and automation in construction came up this year. Not a surprise, since the message that the construction sector is lagging behind in productivity has sunk in fairly well.
‘It is comparatively easy to make computers exhibit adult level performance on intelligence test or playing checkers, and difficult or impossible to give them the [adaptive] skills of a one-year-old when it comes to perception and mobility.’
— The Hans Moravec paradox, in his Stanford PhD thesis on ‘Obstacle Avoidance and Navigation in the Real World by a Seeing Robot Rover’, 1980
When we recently dug into our deal flow of the past 18 months specifically looking for patterns of what tasks tech companies in construction automate first, we couldn’t help but observe one thing: there doesn’t seem to be a clear pattern. Some go after highly labor intensive tasks, some go after highly repetitive tasks, some go after a set of tasks with a high share of total project cost. Our conclusion from both sides (corporates and tech companies) was: there is a lack of understanding what tasks are best suited for automation in construction. We needed a better framework.
1/ Automation of tasks has been a widespread public topic for the past 5 years. The general simplistic analytical framework widely used by many analysts has been that repetitive tasks have a higher probability of automation while problem-solving or creative tasks have a lower probability. Examples of such a simplistic analysis can be found across public media:
2/ And: we can observe the automation of physical tasks in many industries. In 2018, the global operational stock of industrial robots was at its peak (2’440’000 installed industrial robots). The largest industries represented in this data set are automotive, electronics, metal and machinery, plastic and chemical products, and food. In South Korea 2’435 multipurpose industrial robots in the automotive sector are installed per 10 000 employees. (In Canada: 1’354, US: 1’200, Germany 1’162, Japan 1’158). This data confirms that automation of physical tasks is not just a mirage, but it has happened in several industries.
3/ And: in construction technology, automation of physical tasks is a hot issue for GCs. Shortage of skilled labor is a significant problem in practically all mature economies. Health & safety concerns are top of the agenda — in 2018, the US construction industry accounted for 1,003 fatal injuries at 12.2 fatalities per 100’000 construction workers. In 2018, construction worker was unfortunately the deadliest job in the US private industry (bear in mind: US = mature economy).
4/ But: we don’t see anywhere the same degree of task automation in the construction sector as we see in other industrial sectors. Proof: Agriculture and manufacturing gained 240+% of labor productivity since 1990. Construction: dropped by -23% in labor productivity. So what might be the reasons, although the opportunity is in plain sight for everybody ?
5/ We found: applying the above simplistic framework of “task repetitiveness” alone does not accurately represent the automation of tasks in construction. By that logic, hammering nails into walls should have been automated first. Instead…
6/ We see that a second key automation criterion is the control over dimensions. By that we mean
(a) how many components/parts of the exact same dimensions do you repeat a task on and
(b) how repeatable and unobstructed are the surrounding dimensions around the part/component.
7/ Example: In automotive, punching sheet metal parts for a car is a repetitive task in itself. But: also (a) the dimensions of the part itself are repeatable as is (b) the surrounding environment of the tool/die you punch on. Meaning: the environment is standardized as well. This unlocks a high degree automatability of this specific task without requiring adaptive skills.
8/ In construction, most tasks you will find will be applied to parts/components or surroundings with a high variability of dimensions and low control of dimensions. “Every project is a prototype” is a famous saying in construction. All the more surprised are we about how many companies are first going after the automation of tasks with highly variable dimensions in construction. This is the most difficult automation to achieve. The problem with this approach is that you have to engineer robotics that are able to gain control over the environment, which is exponentially harder than to automate for tasks where you already control the dimensions and environment.
9/ Ergo: it appears more fruitful in construction not to blindly follow a principle of “we fully automate the highest cost tasks” (which, honestly, the majority of robotics players we see in our deal flow seem to be following) but instead follow a principle of “we partially automate moderately labor-intensive repetitive tasks with highly repetitive dimensions”. This principle leads to augmentation of labor rather than replacement of labor, and reduces the skills needed by workers. This can also help solve the skilled labor shortage. A good example of this approach is our portfolio company Mighty Buildings.
10/ Applying the above framework, we see several Founder Opportunities in the space of automating construction tasks we would like to fund more:
As always, credits go to the data wizards in our Foundamental Insights group.