Really good overview and write-up. The issue is that learning curves require standardization and investment. There is no entity in the current project-based real estate development model that has the control or incentive to invest. That's why the study from 1965 is still current.
Way back when shopping malls were a big thing, I almost got involved with a group that specialized in building them. They were highly organized and had all sorts of internal procedures to keep construction on track. PCs were a new thing, and they wanted PC based software for the tracking. I considered putting in a bid to supply some but didn't. They had a site south of Boston and were all gung-ho. A few years later, I read that they were abandoning their plans for that site having run into regulatory issues. They were building it in a swamp and wanted access to already overcrowded main roads, that and a recession, killed the project.
Still, they clearly HAD experienced some kind of learning curve. They had built over a dozen malls, and presumably had gotten to the point where their experience prevented a complete learning curve reset.
Other manufacturing is of standard products. We are not close to that in residential and even further in commercial construction. A real, though minor issue, is that other manufacturers do their production inside and do not have to deal with varying site conditions or weather. We once did identical GFRC pieces as part of the roof for identical office buildings. Because the construction sequence was not identical, the GFRC project became non-identical. Connections had to be revised. One of the mysteries in all this is what happened to all the cost and labor savings from things like factory made roof trusses, telescoping fork lifts, etc. Clearly some (but how much?) went into safety., not just of workers, but of occupants. 75 years ago steel was not fireproofed, doing so adds costs.
Also "doubling the size of a building just gets you twice as much building" is inaccurate. To go from 3 stories to 6 may add a lost of cost - elevators, a stiffer structure, etc. Other example abound that call a simplistic analysis into question. Non standard products require a non standard supply chain. It adds cost and complexity.
Have you read Azeem Azhar’s book ‘Exponential’? A similarly broad thesis on learning curves and other technology advancing (& productivity/economy accelerating) phenomena.
It isn't just nuclear power plants. My guess is that there is only a limited learning curve for building any type of process plant. There's a whole business of measuring and documenting such facilities as-built. The implication is that the facility is designed. During construction and early operations, things are changed so that they actually work properly. When the whole system is running, probably not at peak efficiency but well enough, it is documented so that people can look at the plans for troubleshooting and modifications. This is an expensive but vital process, but the fact that it exists at all suggests that each new facility has a reset and a learning curve.
Hi Brian - this is another awesome post. Could you clarify one thing for me? Is there any overlap in these figures between economies of scale and manufacturing expertise? I would expect as volumes increase the costs of goods decrease. But the efficiencies you are highlighting are isolated from lower cost of goods. Am I understanding that correctly?
I know you've discussed this in previous posts, but a I think big reason could be the cost of buildings is more than just the quantity/sq ft. Also, are safety and/or environmental requirements taking down that efficiency change? And lastly, is it the suppliers to the construction industry that have earned the gains and they've managed to keep those for themselves?
We need to reduce the geometric variation in single family homes down to (4) types and limit aesthetic options to paint colors. Structural systems, cladding, windows, cabinetry, fixtures, and finishes will be identical and envelopes will be calibrated to the most extreme climate zone. Site selection will be limited to areas with no topographical features that can impede construction efficiency. With this criteria we will be able to automate assembly and maximize productivity after a few thousand units.
Your implicit goal is standardization. In addition to the variables you mention, add size, or are all the 4 types coming in 4 sizes? Other anti-standardization factors include that building codes are local and may affect design, ditto for climate (why would you put the same house in Tucson as in Buffalo - one deals with a lot of heat and sun, the other large snow loads).
Great article as usual, Brian. Is a simple solution here to "keep the team together"? Try (as much as possible) to reduce team turnover from project to project and work on similar types of buildings/infrastructure.
Fantastic article. Anyone who has worked in software is keenly aware of the "reset" effect you describe (though I've never seen it examined so clearly!). This is why freelancers/software consultants spend so much up-front effort getting a spec and demanding that the client pay penalties for deviation. Too many intrinsic dependencies to easily pivot.
Do you know of any learning curve data in reference to 3D printed home technology?
I don’t know about learning curves in printed buildings, but it’s worth stressing that it’s early, early days for 3D printed buildings. Current examples, like Austin’s Icon (https://www.iconbuild.com/) require lots of setup and infrastructure just to automate the cheapest & easiest parts of construction, rough wall framing. There’s still a tremendous amount of labor that goes into plumbing, electricity, flooring & roofing. It’s distinctly unimpressive… except that only by working through these awkward stages can we move on to buildings that embody the idea of truly automated building. We’re definitely in the top-left of the chart now, and I expect another 10 years or so before there’s significant headway.
"The US had approximately 45 million housing units in 1950, and has since built over 100 million more - about 1.66 doublings "
The learning curve applies to the rate of production, not the cumulative output. Housing starts have fluctuated around the 1M/year mark for the last 70 years so learning curve theory would predict no durable productivity improvement because there was no durable increase in production quantity.
Really good overview and write-up. The issue is that learning curves require standardization and investment. There is no entity in the current project-based real estate development model that has the control or incentive to invest. That's why the study from 1965 is still current.
Way back when shopping malls were a big thing, I almost got involved with a group that specialized in building them. They were highly organized and had all sorts of internal procedures to keep construction on track. PCs were a new thing, and they wanted PC based software for the tracking. I considered putting in a bid to supply some but didn't. They had a site south of Boston and were all gung-ho. A few years later, I read that they were abandoning their plans for that site having run into regulatory issues. They were building it in a swamp and wanted access to already overcrowded main roads, that and a recession, killed the project.
Still, they clearly HAD experienced some kind of learning curve. They had built over a dozen malls, and presumably had gotten to the point where their experience prevented a complete learning curve reset.
Other manufacturing is of standard products. We are not close to that in residential and even further in commercial construction. A real, though minor issue, is that other manufacturers do their production inside and do not have to deal with varying site conditions or weather. We once did identical GFRC pieces as part of the roof for identical office buildings. Because the construction sequence was not identical, the GFRC project became non-identical. Connections had to be revised. One of the mysteries in all this is what happened to all the cost and labor savings from things like factory made roof trusses, telescoping fork lifts, etc. Clearly some (but how much?) went into safety., not just of workers, but of occupants. 75 years ago steel was not fireproofed, doing so adds costs.
(see https://en.wikipedia.org/wiki/Our_Lady_of_the_Angels_School_fire).
Also "doubling the size of a building just gets you twice as much building" is inaccurate. To go from 3 stories to 6 may add a lost of cost - elevators, a stiffer structure, etc. Other example abound that call a simplistic analysis into question. Non standard products require a non standard supply chain. It adds cost and complexity.
Lots of “forgetting” involved from project to project. Hinders maximizing learning curve effects
Outstanding work yet again.
Have you read Azeem Azhar’s book ‘Exponential’? A similarly broad thesis on learning curves and other technology advancing (& productivity/economy accelerating) phenomena.
It isn't just nuclear power plants. My guess is that there is only a limited learning curve for building any type of process plant. There's a whole business of measuring and documenting such facilities as-built. The implication is that the facility is designed. During construction and early operations, things are changed so that they actually work properly. When the whole system is running, probably not at peak efficiency but well enough, it is documented so that people can look at the plans for troubleshooting and modifications. This is an expensive but vital process, but the fact that it exists at all suggests that each new facility has a reset and a learning curve.
This article is far too optimistic:
https://www.archdaily.com/971874/off-site-construction-is-radically-changing-the-rules-of-architectural-design
Hi Brian - this is another awesome post. Could you clarify one thing for me? Is there any overlap in these figures between economies of scale and manufacturing expertise? I would expect as volumes increase the costs of goods decrease. But the efficiencies you are highlighting are isolated from lower cost of goods. Am I understanding that correctly?
I know you've discussed this in previous posts, but a I think big reason could be the cost of buildings is more than just the quantity/sq ft. Also, are safety and/or environmental requirements taking down that efficiency change? And lastly, is it the suppliers to the construction industry that have earned the gains and they've managed to keep those for themselves?
We need to reduce the geometric variation in single family homes down to (4) types and limit aesthetic options to paint colors. Structural systems, cladding, windows, cabinetry, fixtures, and finishes will be identical and envelopes will be calibrated to the most extreme climate zone. Site selection will be limited to areas with no topographical features that can impede construction efficiency. With this criteria we will be able to automate assembly and maximize productivity after a few thousand units.
Your implicit goal is standardization. In addition to the variables you mention, add size, or are all the 4 types coming in 4 sizes? Other anti-standardization factors include that building codes are local and may affect design, ditto for climate (why would you put the same house in Tucson as in Buffalo - one deals with a lot of heat and sun, the other large snow loads).
Great article as usual, Brian. Is a simple solution here to "keep the team together"? Try (as much as possible) to reduce team turnover from project to project and work on similar types of buildings/infrastructure.
Fantastic article. Anyone who has worked in software is keenly aware of the "reset" effect you describe (though I've never seen it examined so clearly!). This is why freelancers/software consultants spend so much up-front effort getting a spec and demanding that the client pay penalties for deviation. Too many intrinsic dependencies to easily pivot.
Do you know of any learning curve data in reference to 3D printed home technology?
I don’t know about learning curves in printed buildings, but it’s worth stressing that it’s early, early days for 3D printed buildings. Current examples, like Austin’s Icon (https://www.iconbuild.com/) require lots of setup and infrastructure just to automate the cheapest & easiest parts of construction, rough wall framing. There’s still a tremendous amount of labor that goes into plumbing, electricity, flooring & roofing. It’s distinctly unimpressive… except that only by working through these awkward stages can we move on to buildings that embody the idea of truly automated building. We’re definitely in the top-left of the chart now, and I expect another 10 years or so before there’s significant headway.
"The US had approximately 45 million housing units in 1950, and has since built over 100 million more - about 1.66 doublings "
The learning curve applies to the rate of production, not the cumulative output. Housing starts have fluctuated around the 1M/year mark for the last 70 years so learning curve theory would predict no durable productivity improvement because there was no durable increase in production quantity.
This is incorrect. See, for instance, the wikipedia page on experience curve effects https://en.wikipedia.org/wiki/Experience_curve_effects
"The learning curve model posits that for each doubling of the total quantity of items produced, costs decrease by a fixed proportion"
Oops, you're right. My bad!
This is really great.