This article is part of a series called Future Space, which focuses on innovations in design, technology, and other fields.
When architect Yorgos Kapourniotis started at WeWork, one of the simplest tasks—deciding how to arrange desks in each private office—was causing him some big headaches.
“I was covering projects for other architects, and placing desks was slowing things down,” says Kapourniotis. “It’s usually a simple process, but when there are dozens of offices per floor it can take a lot of time.”
Kapourniotis knew that there had to be a way to automate the process. So did WeWork’s Fundamental Research team, which was already putting together a tool that does just that.
The tool—too new to even have a name—is officially a “suite of procedural algorithms for space planning in commercial offices.” In layman’s terms, it’s a tool to calculate the most efficient ways to arrange desks in offices holding up to 20 people.
The tool is smart and fast—completing its task as efficiently as a designer and in a fraction of the time.
“It’s amazing the speed it does things,” says Kapourniotis, who’s based at London’s WeWork Hammersmith: Brook Green. “When you’re designing an entire building, it could save half a day or even a full day of work.”
But the tool does more than save time. It frees up architects to use their creativity in other ways, such as designing an eye-catching central staircase or covered courtyard where members can mix and mingle.
“None of our architects went to design school so they could place desks in offices,” says Daniel Davis, director of research at WeWork. “Now they have more time to do the really cool stuff.”
‘It’s not theoretical, it’s real’
Automated office design has been around since the 1960s, when experts started using algorithms to help devise floor plans. While these researchers laid the theoretical foundation for WeWork’s later work, few of their innovations made it into practice given the difficulty of developing tools that handle all the oddly shaped floors, errant columns, and other constraints on architecture projects.
But at WeWork, all of its buildings around the world are created with the same set of building blocks. Although its locations are unique—from the 19th-century grandeur of WeWork Champs-Élysées in Paris to the strikingly modern lines of WeWork Oskar Von Miller in Munich—inside each of them are private offices filled with desks.
The size and shape of the offices and the types of furnishings inside them are consistent across all WeWork locations. And WeWork’s design guidelines—the amount of space a person needs when getting up or sitting down at a desk, for instance—ensure that every WeWork office is a comfortable place to work.
For WeWork’s researchers, this meant that a tool programmed to design offices in New York City could do the same quality of work in offices anywhere in the world. There was no need to reinvent the wheel for every building.
Because human designers have already designed thousands of offices, WeWork’s research team had a way to measure their success. Simply ask the tool to design the same set of offices and see if it can match, or even exceed, the efficiency and design quality of desks placed by humans.
“That’s why this project is so exciting,” says Davis, whose team combines data science and social science to better understand how spaces can enhance people’s happiness, productivity, and connection to their community. “For previous studies of building layout tools, these other companies didn’t have data sets to compare their results against. But WeWork does. The benchmark isn’t theoretical, it’s real.”
Andrew Heumann, a senior researcher on the Fundamental Research team, explains how the layout tool works. An architect selects all the private offices on a floor plan, then activates the tool. It responds by generating all the possible desk layouts for each office. For a small office it might be two or three layouts, while for a larger one it might be a dozen or more. The architect clicks one, and the tool adds desks to the floor plan.
The floor plan Heumann was using as an example had 34 offices. Using the algorithm, he was able to place all the desks in about two minutes. For a human, the same task could take an hour or more.
“There’s no question that this tool could save a lot of time,” says Heumann. “And it’s just the beginning of what we can do.”
Letting computers be more human
When Davis’s team was creating the desk layout tool, the goal was to have it perform as well or better than architects 99 percent of the time. While it sounds like an impossible goal, Davis points out that unless it performs reliably and consistently, it won’t become a go-to tool for architects.
“Our team has to be confident that this tool can do the job,” says Davis. “If it works only 80 percent of the time, nobody is going to trust it.”
To access the tool’s effectiveness, they had it go head to head with architects in terms of designing the most efficient space. They compared the tool’s results on 13,211 offices designed by architects since 2016.
The results? The tool performed as well as architects 77 percent of the time and did better another 6 percent of the time. But while 83 percent was impressive, it was still far from the goal.
“It means that the tool was underperforming almost 20 percent of the time,” says Davis. “We had to figure out why, if it was using the same parameters as our architects, it was falling short.”
Davis and his team pored over the data they had from those 13,211 offices. Why were the humans so easily beating the computers? How were they coming up with layouts that seemed out of reach for their digital counterparts?
The answer turned out to be simple: The humans were cheating.
WeWork’s guidelines determine how much space each desk requires, which makes the design process pretty straightforward when the office is a simple rectangle. But if one wall sits at an angle, or a window is off-center, or a column juts out into the space, things get a little more complicated.
Humans deal with these problems by making tiny adjustments. Seeing a potential problem, they can move the placement of a desk by an inch or two. Crunching the data, the researchers found that human designers made these kinds of adjustments 28.4 percent of the time.
But a computer can’t make a decision like this unless programmers give them permission. They have to know where, when, and how they can bend the rules—essentially, the programmers need to allow them to act a little more human.
“We found out that it’s quite an art to translate what a human thinks about when designing an office,” says Heumann. “ A lot of thought and intuition goes into to making a lot of those decisions.”
The research team looked closely at the specific reasons human designers were bending the rules and discovered it was almost always due to small variations in the space around a particular desk. They gave the tool permission to do the same, and the tool has since performed as well or better than humans 97 percent of the time.
To find the most efficient layout for an office, humans were bending the rules 28.4 percent of the time. The tool ending up bending the rules just 17 percent of the time.
‘Just scratching the surface’
The desk layout tool—which the Fundamental Research team wrote about in last month’s International Journal of Architectural Computing—isn’t the only one in WeWork’s toolbox. Heumann points out that architects are consistently using another tool to help them calculate the right amount of common spaces—such as lounges, kitchens, and conference rooms—for any location.
“It’s especially helpful for designers who’ve not been with us for long,” says Heumann, “because it helps them adhere to WeWork standards.”
In the future the space calculation took will be able to automatically adjust to regional differences, such as members in China preferring large conference rooms or safety codes requiring wider hallways in Japan.
Heumann says these tools have one thing in common: They allow architects and designers to eliminate the drudge work.
“We’re looking at what tasks can be automated, and what tasks should be automated,” he says. “The tasks that are most tedious and repetitious are probably the best place to start.”
Experts believe that one of the last fields that will be completely automated is design. But the repetitive tasks can easily be automated—a recent study showed that 18 percent of design tasks can be handled by technology that currently exists.
“At WeWork, we’re just scratching surface of what’s possible,” says Heumann, who was one of the designers of the new tool. “We’re coming to the end of having to do the same things over and over again.”