Laura Bliss is CityLab’s west coast bureau chief. She also authors MapLab, a biweekly newsletter about maps (subscribe here). Her work has appeared in the New York Times, The Atlantic, Los Angeles magazine, and beyond.
The co-working startup is hatching plans to deploy data to reimagine urban problems. In the past, it has profiled neighborhoods based on class indicators.
The We Company, the all-encompassing life-services platform formerly known as WeWork, is entering the booming business commonly known as “smart cities.” Di-Ann Eisnor, the former Google executive who helped grow Waze into a traffic-data juggernaut with 90 million monthly users, will lead the recently rebranded We Company’s efforts to build data-driven products and partnerships with cities and community groups, aimed at tackling barriers to jobs, housing, education, and other problems related to urbanization.
It sounds like a characteristically ambitious move for the startup, which began by renting desks in a Manhattan storefront in 2010. With its signature free-flowing beer taps, ping-pong tables, and “Thank God It’s Monday” ethos, the We Company has since grown into an empire of co-working spaces in more than 100 cities around the globe, with a valuation of $42 billion. It has also built out a host of new services, including dormitory housing (WeLive), Montessori-style early childhood education (WeGrow), fitness (WeRise), and social outings (Meetup, acquired in 2017).
Details are scant on how its latest move to sculpt the future of cities will contribute to its efforts to turn a profit. (Like many unicorns in this stage of capitalism—including ones now going public, like Uber and Lyft—the We Company loses hundreds of millions of dollars per quarter.) There likely won’t be much information available until Eisnor is settled with a team of software engineers, architects, economists, and (apparently) biologists.
But the initiative is likely to include a survey of the firm’s many data sources to see how they might be applied in the service of entire communities. Quartz reported that Eisnor is charged with taking “what We has already done inside the building, take it outside, and reimagine a sort of connective tissue for 21st-century cities.”
Given Eisnor’s history, there may well be thoughtful offerings in store. She’s known around Silicon Valley for her sense of a social contract with customers and communities. One of her major projects at Waze is a good example: She helped launch its Connected Citizens program, whereby the app shared some data with city governments looking to fix recurring transportation problems. In an interview with CityLab, Eisnor emphasized that what We Company builds will be in concert with existing community efforts. “The smartest thing about a city is the humans on the ground,” she said.
But the We Company isn’t a novice in the urban data department, so it may be useful to look at what it has tinkered with in the past. At one point, the company experimented with applying machine learning to information about neighborhood attributes and demographics, in order to inform its real-estate leasing decisions. Such tools could be useful to city builders in the public and private sectors. But they may also raise some eyebrows.
The We Company has long been building a robust spatial analytics infrastructure. Besides having rich datapoints filtered from its members-only network, which includes the professional profiles of more than 400,000 WeWorkers and active messaging boards, We also has 18 years worth of information about how folks socialize offline, thanks to its purchase of Meetup. It also has been using sensors and cameras to test how its employees use common areas, conference rooms, and desk setups, and is planning to experiment with capturing smartphone movements via a wifi beam popular with shopping malls.
And since the company has a major stake in understanding cities and neighborhoods—after all, it wants the highest possible occupancy rates in the buildings it leases—the firm has dabbled in urban data science. As a 2017 article in Entrepreneur explained, it’s not always easy for the company to know which neighborhoods are most likely to attract WeWork members, especially outside major hubs like New York City, Washington, D.C., and San Francisco. In secondary and tertiary markets, like a Nashville or Kansas City, or in cities where WeWorks do not exist yet, pinpointing the right corner for a new location might be about a particular mix of amenities, or a hard-to-pin-down vibe. Is there a nice new gym in the area? A certain mix of bars and higher-end restaurants? Then there might be a future WeWork member—a youngish, college-educated, white-collar worker—interested in coming.
Such inferences might be intuitive to people in the real estate business, or indeed, anyone who’s ever walked around a gentrifying neighborhood. But for WeWork to utterly saturate the planet with kombucha taps and plush phone booths—and make good on the billions that SoftBank, its Japanese investment patron, has pumped into it—it seems that decisions about where and when to open new locations have to happen faster, more systematically, and from a remove.
“We need a systematic and scientific way of making assessments,” said Aaron Fritsch, formerly the company’s head of product systems and operations (“SimCity in real life,” reads the description he gave that job on his LinkedIn profile) and the current chief of staff to We’s Chief Product Officer, in Entrepreneur. “Anything that we can do to give people tools to synthesize more information in less time immediately correlates to more locations open.”
To make its real-estate decision-making less about gut feelings and more about hard data, the We Company has turned to artificial intelligence. One of their researchers talked about how this worked in late 2017, at an industry event held at the Brooklyn offices of Carto, a mapping firm. “We have fleets of brokers, real estate experts, who are going out to look at buildings, stand on street corners, and assess the ‘vibe and energy’ of a location,” Carl Anderson, then a senior research scientist at WeWork who left the company in 2018, told the audience. “They do a good job, but they don’t do a good job of explaining why the vibe is good. What are the features that contribute to this feeling?”
To determine what those attributes might be, Anderson explained, WeWork was testing out machine-learning processes that draw in several types of data about a given neighborhood—first, general indicators of its “feel,” such as its daytime population, its median home value, and its Walk Score, and second, all of its storefront businesses and major points of interest, sorted into identifiers as granular as “dog parks,” “Chinese restaurants,” and “steakhouses” by the location data services company Factual. The algorithm analyzed these data points, generating a “thumbs up” for good buildings to investigate, and “thumbs down” for unattractive locales.
“So maybe we learn that having a Blue Bottle Coffee nearby is a good thing,” Anderson said. “And a Western Union might be an indicator that it’s not such a great block. Maybe it’s proximity to rivers and parks. Or whatever. But we’re trying to work that out.” Around the same time, the company claimed that its ability to scope locations in roughly this fashion sped up its already fast-growing footprint.
What Anderson described, in other words, was a kind of automated taste-tracker for a specific set of affluent people. A city planner or housing developer could find such a tool handy, especially if it was more up-to-the-minute and detailed than, say, the census and business registration data that tends to show how wealth and amenities spread around cities. An automated neighborhood-profiling widget could very easily go awry, however, if it isn’t attuned to who and what is being screened out. Imagine, for example, a local government basing planning and investment decisions on the ratio of check-cashing places to upscale cashless coffee shops in a certain neighborhood. One could see this particular “smart city” innovation leading to algorithmic redlining.
The We Company is no longer actively pursuing these types of applications for machine learning. According to a spokesperson, it still pays WalkScore for various data-points about the surroundings of its potential locations, including proximity to transit. The company has largely refocused its data-driven efforts toward assessing the attributes that give office interiors good “vibes”—think algorithmically optimized arrangements of desks, chairs, and lighting, and the like. And it hopes to distance itself from the “smart city” label, which has become fraught with suspicion as other tech companies have attempted to turn urban data into dollars (see: Sidewalk Labs’ efforts to develop a Toronto neighborhood “from the internet up.”) Notably, it is referring to the work that Eisnor will spearhead as a “future cities initiative.”
But we don’t know yet what that will look like in practice. And plenty is known about the risks of applying machine-learning processes—or really any “smart cities” technology—to huge social challenges. Information technology doesn’t solve problems by itself. If the data is biased towards certain groups and away from others, high-tech tools can simply reinforce existing inequalities.