Sketch of a Theory of Construction Productivity, and a Research Road map
This newsletter is, broadly, about how the construction industry fits together, and why it’s so hard to improve its efficiency. The reason for this focus is that while many industries, such as agriculture or manufacturing, have seen sustained productivity improvements over long periods of time, construction has not. The amount of productivity growth in construction is difficult to pin down due to measurement difficulties, but it has been very low, possibly negative, for many years.
These measurements tend to be US focused, but US construction costs appear to compare relatively favorably to other western countries for most project types, suggesting it’s not simply a quirk of the US construction industry.
Lack of productivity growth is important - it means buildings haven’t gotten cheaper the way manufactured goods have. Being able to construct buildings, especially homes, could make them far more affordable and be a huge gain for society.
So far this newsletter has been a series of loosely connected articles about various aspects of construction efficiency, which mostly stand alone. Below is my initial attempt to tie these together into a more coherent theory of how construction productivity works.
Construction and Productivity
When we say productivity, we mean the amount of resources it takes to build something. A system getting more productive means that over time it takes fewer and fewer inputs - labor, money, materials, etc. - to produce a given volume of output. For instance, to produce a given volume of corn requires about 1/8th the land today as it did in 1930:
When we talk about productivity we often specifically mean labor productivity - how much output is produced per hour of work. Labor is expensive, and finding ways to use less of it allows things to be made cheaper. Most other industries have seen increasing labor productivity over time, but construction has not:
In fact, one of the defining features of construction is how labor intensive it is - up to 50% of the construction cost of building a single family home is on-site labor, for instance. This is partly due to the structure of the industry - construction in many ways resembles a service industry, where a great deal of time and effort is spent on providing a tailored solution to the client. Labor intensive industries tend to see steady cost increases over time, seemingly due to the Baumol Effect - increasing labor productivity in one industry pushes up wages not only in that industry, but in all other industries.
Construction vs manufacturing
The solution to this is generally thought to be changing the construction process to be less like a service and more like a product - in essence, to be more like manufacturing. This would theoretically allow a given amount of labor to produce a much larger volume of output, and perhaps enable the kinds of cost reductions that manufacturing has been able to achieve.
People have tried many times to do this - to build buildings in factories like other manufactured goods. It has seen some success in the sense that it can be a profitable way to build, but it has not dramatically reduced the costs of construction (except through labor arbitrage). Efforts to deliberately bring down costs by applying manufacturing best practices or lessons from mass production have generally failed. A few examples:
Lustron - A post WWII factory-built housing company in the US, made houses from enameled steel panels. Ended in bankruptcy
Stirling Homex - A late 60s factory-built housing startup, founder declared “The Henry Ford of construction”. Declared bankruptcy, and was later found to have engaged in widespread sales fraud.
Katerra - Late 2010s factory-built construction builder that was highly vertically integrated. Declared bankruptcy.
Emergency Factory Made Housing Program - Post WWII program in England to repurpose munitions factories to build needed housing. Successfully built over 150,000 homes, but did not fundamentally change construction methods.
Toyota Home - Homebuilding arm established by Toyota to apply their manufacturing expertise to houses. Continues to build what appear to be advanced, high-end prefab houses, but has not been able to bring down construction costs.
Operation Breakthrough - A late 60s HUD program to kickstart a factory-built housing industry. Funded many innovative factory construction efforts, provided a market for them, and established performance-based codes to make it easier to use a factory-built system. None of the funded systems are currently used to build homes, and most stopped production after just a few years.
This is just a small cross section of attempts at factory-based construction - I have books with hundreds of different industrialized building systems, dating back 100 years. Many of these systems were used successfully for many years. Many used advanced production techniques such as assembly lines, machine-built components, innovative materials, etc. But they don’t seem to have altered the fundamental economics of erecting a building, and construction in the US remains relatively unchanged - in fact, it mostly resembles the way cars were built over 100 years ago, prior to modern methods of manufacturing that made them inexpensive and ubiquitous. From “The Machine that Changed the World”:
In 1894, the Honorable Evelyn Henry Ellis, a wealthy member of the English Parliament, set out to buy a car. He didn’t go to a car dealer - there weren’t any...Instead, he visited the noted Paris machine-tool company of Panhard et. Levassor and commissioned an automobile…in 1894 it was the world’s leading car company. By the early 1890s, P&L was building several hundred automobiles a year. When Ellis arrived at P&L...he found in place the classic craft-production system. P&L’s workforce was overwhelmingly composed of skilled craftspeople who carefully hand-built cars in small numbers. These workers thoroughly understood mechanical design principles and the materials with which they worked. What’s more, many were their own bosses often serving as independent contractors within the P&L plant, or more frequently as independent machine-shop owners with whom the company contracted for specific parts or components. The two company founders, Panhard and Levassor, and their immediate associates were responsible for talking to customers to determine the vehicle’s exact specifications, ordering the necessary parts, and assembling the final production. Much of the work, though, including design and engineering, took place in individual craft shops scattered throughout Paris.
Manufacturing and process improvement
Why has the manufacturing playbook not worked in construction? To understand this, we need to better understand what is happening when a manufacturing process gets more productive. One way to do this is to look at the specific mechanisms that allow an improvement to happen. Making a process more efficient means removing some of the resources required to do it, and there’s only so many different ways that can be done. Some of the possible ways of doing this:
Economies of scale
There are a number of different effects that all allow a process to get more efficient as it gets larger:
Geometrical scaling - a process will often have costs proportional to surface area (which rises with the square of some dimension), but output proportional to volume (which rises with the cube of some dimension). This often shows up in chemical manufacturing or other bulk material production, where larger and larger tanks or pipes increase the efficiency of a process.
Fixed cost scaling - A process will often have parts of it that have the same cost regardless of how much output is produced. If I’m printing a book, the costs of editing, proofreading, type layout, etc. are all the same regardless of how many copies I print. The more books that get printed, the more those fixed costs can be spread out. Vertical integration can be thought of as a form of fixed cost scaling - the larger you get, the more it’s worthwhile to bring some process in-house so you can optimize it for your particular needs.
Statistical scaling - Variance in a process degrades its performance (more on this below). A larger process can often have a lower variance due to uncorrelated sources of variation averaging out. If you have 50 of a given type of machine that breaks down every so often, you’ll need a proportionally smaller number of replacement parts than you would if you only had 5 machines.
Learning curve effects - Production efficiency tends to increase with cumulative production volume, a phenomenon known as Wright’s Law. The more you do something, the better you get at it. Specialization can be thought of as a learning curve effect where a person or business does something enough that they become an expert at it.
Bulk purchasing - The larger you get, the more you can take advantage of scale effects in your suppliers, which make it cheaper for them to sell to you larger quantities.
There are also diseconomies of scale - things that make a process less efficient as it gets larger. Coordination costs are a common one - every new person hired is someone everyone else might need to talk to, so the nth person adds n-1 new communication channels. Improving processes are often able to grow for a surprisingly long time before diseconomies of scale overwhelm them (more on this below).
Replacing a human in a process with a machine decouples your process from a human’s physical limits, letting a process go faster or do more at once than a human would be capable of. It also reduces your labor costs (at the expense of some increase in capital cost). Automation can be thought of as mechanization that includes mechanizing some decision or information-based process.
Mechanization doesn’t seem quite as fundamental as some of these other effects (it seems like it can be thought of in terms of both removing extraneous elements and economies of scale, for instance), but it’s been such an important factor for cost reductions in other industries that it seems important to include.
Removal of extraneous elements
Improvements to a process often take the form of removing some element of the process that wasn’t necessary. This can be organizational (arranging workers such that they don’t have dead time between tasks), to part of the process itself (limiting work-in-process to minimize the amount of inventory accumulated), or to the product itself (combining two parts into one thus removing an assembly step).
Reduction in variation
Variation in a process degrades performance. A process with a lot of variation will sometimes have equipment or labor going unused (and thus wasted), and sometimes get backed up with large queues (and thus increasing cycle time and inventory). As we saw in Construction, Efficiency, and Production Systems, a process can be made substantially more efficient purely by reducing the variation, without making any other changes.
Sometimes a change in technology allows a given function to be achieved in a way that requires fewer resources to produce. The Haber-Bosch process, for instance, is a chemical reaction for producing nitrogen that consumes less energy than previous processes:
This breakdown in improvement mechanisms should be regarded as a work in progress. For one, it doesn’t yet seem like it’s exhaustive. For another, these effects interact in complicated ways, and in fact many things can be in different buckets depending on how you look at them. For instance, in precast concrete when designing a building you generally try to minimize the number of pieces required, and aim for a small number of large pieces rather than a large number of small pieces. This looks like “removing extraneous elements”, but it also looks like fixed cost scaling - there are fixed costs associated with producing a single piece, and the fewer pieces you have the greater the volume of building each fixed cost can be spread over. Learning curve effects are a function of production volume, but they’re also an abstraction over a bunch of specific improvements. So I don’t yet understand how all this fits together.
But this gives us a good starting point to get a better sense of how specifically a process improves - most manufacturing innovations can be thought of in terms of one or more of these effects. Interchangeable parts, for instance, leveraged mechanization (since the interchangeable parts were machine made), to remove extraneous elements (by removing the time and effort required to get the highly variable parts to fit together), as well as reduce variance (identical parts meant the assembly process could be the same every time). Ford’s assembly line took these benefits a step further, combining interchangeable parts with a moving workstation to cut out even more extraneous elements (in the form of time between assembly steps) from the process.
Construction and the manufacturing playbook
Looking at the problem through this lens, it becomes a little more clear why construction has struggled with productivity improvements. There’s only so many methods for pulling resources out of a process, and construction struggles with many of them.
Economies of scale are the most obvious culprit here. Factory-produced buildings have had difficulties producing in truly large volumes, which has made it hard for them to capture scale effects. Even the largest scale producers, such as Clayton Homes, produce a relatively small number of homes at each factory.
This seems to be the result of construction running up against diseconomies of scale sooner than other industries, specifically transportation - the more you produce, the farther and farther away you need to transport your product to reach the customer. For small, expensive things, this cost is easily absorbed (which is why iPhones can be transported halfway around the world), but for large, inexpensive things (things with low product-value density), shipping costs dominate. This makes it hard to achieve economies of scale for large building components, as transportation restrictions limit how much a given facility can produce. In some cases, such as with bricks (which have exceptionally low product value density), even mass production doesn’t seem to have achieved lower costs.
Construction also faces vertical transportation diseconomies - the taller a building gets, the more expensive each floor becomes. Taller buildings have stricter building code requirements, see proportionally higher lateral forces (which go up with the square of height), have more complex plumbing to deliver the required water pressure, require larger cranes to build, must budget more time for workers to reach the top, must devote a greater fraction of the floor area to elevators, etc. This screens off some potential within-project economies of scale.
Economies of scale are further impacted by significant regional variation in product requirements. Environmental requirements (different HVAC systems will be needed in Maine vs New Mexico), local building codes and permitting authorities, transportation regulations (different states have different rules for how wide a truck they will allow), site conditions (soil quality, slope, lot size) will all vary significantly and make it hard to offer a uniform product.
In addition to regional variation, construction must also contend with the variation imposed by the industry structure. Buildings are generally built by a collection of small, regional subcontractors, each one working on several different projects. The small number of projects means that any given contractor will have wide variation in how busy they are, which means they’ll sometimes have queues of work build up and have to push out deliverables (“sorry, busy next week, we can come and run the wiring the week after”). The sequential nature of construction (and the small size of most buildings) means that these delays can push out project timelines, driving up costs.
Construction has also had difficulty mechanizing. There seems to be a confluence of factors at work here. For one, it’s difficult to amortize the cost of equipment over a large production volume (scale effects strike again). This is partly due to equipment getting more expensive the larger it is and the greater the forces it needs to handle, so mechanization equipment will be especially expensive for building fabrication (this is especially important with very heavy building systems such as precast concrete - much of the optimization for a precast concrete building involves minimizing the overall crane cost). For another, the topology of a building (where much of the work needs to be done inside the building itself) constrains the types of equipment that can be used - your machinery generally needs to be movable, and be able to fit through a doorway (though there are attempts to get around this). Mechanization also often struggles with making adjustments based on environmental feedback, which makes it hard to use outside of a controlled environment such as a factory. Mechanization has occurred, but it’s been limited - more use of forklifts and cranes, more use of power tools, more use of low-level prefabricated components (such as windows or trusses).
Construction also struggles with new technology that allows doing the same thing with fewer resources. At a certain point it becomes difficult to pull any more material out of a building.
Reducing extraneous elements seems like it has the greatest potential for increasing construction efficiency, through any number of possible routes: reducing time spent on fixes and rework in the field by better information coordination, reducing upfront design time with better design software, reducing wasted equipment time with better on-site monitoring, etc.
One challenge here is that construction projects are done by teams with weak, short-term relationships. This makes it difficult to learn from experience and adjust the process accordingly.
Current state of construction
Because of these difficulties, construction has instead been gradually optimized for its current set of constraints. Work is done by many small subcontractors with low profit margins and low capital requirements/fixed costs. Building systems are designed around what a worker in the field is capable of doing. Design cost is minimized by leveraging existing “standard practices” which reduces how much information must be specified up front, and how much subcontractors must coordinate. Prefabricated components are mostly low level (doors, windows, trusses, outlets and switches etc.) that can be effectively mass produced without interfering with building customization. Mechanization mostly takes the form of better power tools.
Over time, this system gets increasingly entrenched. Even if these constraints were found to no longer apply, it would still be difficult to dislodge. Incentive problems prevent builders from experimenting with new technologies. A patchwork of local regulations makes it difficult to achieve economies of scale, and the idiosyncrasies of zoning board public hearings and building official opinions make it hard to move away from the service-based model. Standard building practices means any new building method will face an initial design and coordination cost penalty. Prescriptive building codes make it hard to use things outside a specific system, and make it hard to introduce a system that would be optimized for machine performance.
Research road map
This doesn’t give us the answers, but it does suggest gaps that need to be filled in:
We need a better understanding of the mechanisms at work in an improving process, and additional mechanisms/effects that might be taken advantage of.
We need a better understanding why construction has a hard time taking advantage of these mechanisms, and what points of leverage might be applied to address them.
We need a better understanding of what manufacturing or transportation technology might change the calculus of efficiency, such as small-scale or local fabrication technologies.
Since building cost is at least partly a function of material cost, we need ideas for how materials might be made cheaper, or how buildings might require less of them without degrading performance.
If you have ideas for any of these, or can suggest literature on them, please let me know!
So, to sum up a years with of posts on this topic:
Construction has failed to see the productivity gains of other industries, such as manufacturing or agriculture. Attempts to address this by making construction more like manufacturing have mostly failed. This seems to be related to the fact that there are only so many ways of pulling resources out of a process, and construction struggles with many of them. Understanding what these mechanisms are, how they interact, and the constraints on taking advantage of them, as well as how technology might change them, would aid in determining what interventions might be effective in solving these problems.
This is great.
The retail / restaurant industries have the most of the same diseconomies of scale you talk about, but they at least seem to do ok in terms of productivity growth. They have to produce everything on site in small quantities, lots of variation in local demand, lots of different local regulations. McDonalds centralizes a lot of the marketing and R&D but different restaurants in different countries have their own menu items and operate as autonomous franchises. Amazon is more centralized and has developed stores where you can skip checkout.
It does seem harder to use microchips to improve productivity in construction than in retail or restaurants, where you can replace people with cameras and kiosks. But maybe not.
Coming at this as an outsider with no familiarity with construction, my first reaction would be to ask why there is no McDonalds of construction that centralizes whatever can be centralized? Or maybe there is?
The transport costs issue makes me wonder if heavy lift airships could make a difference to the economics of construction if they ever actually go mainstream. There are a number of companies working on them but it's one of those technologies that always seems to be 5-10 years away from real commercialization.
This article talks about some of the companies developing these vehicles: https://www.bbc.com/future/article/20191107-how-airships-could-return-to-our-crowded-skies