🖤 Donations of all sizes power our public service journalism Give Now
How an algorithm helps convert empty offices into housing
Jun 2, 2023

How an algorithm helps convert empty offices into housing

In the wake of the shift to working at home, many cities want to transform depopulated buildings into residences. Steven Paynter of Gensler is helping to automate the assessment of each property’s livability and cost-effectiveness.

Many cities in this country are suffering from a serious lack of housing.

But thanks to remote work, cities also have a lot of empty office space. Converting one into the other seems like the obvious solution, but it’s not always simple to accomplish.

Many office buildings just aren’t suitable for residential use. The cost of figuring out which ones are can be prohibitive for developers.

That’s where artificial intelligence comes in, says Steven Paynter, a Toronto-based studio director for the design firm Gensler. Marketplace’s Meghan McCarty Carino spoke to Paynter about an algorithm he created to identify the best conversion prospects.

The following is an edited transcript of their conversation.

Steven Paynter: The big things that start to become issues are really elements like the quarter window depth. So how far is from the elevators to the glazing, in an office building, that can often be 50, 60 feet. Imagine walking through the front door of your apartment and the window is 60 feet away. All of a sudden, it’s not so great. And you see that in a lot of buildings, especially kind of ’80s, ’90s buildings. A big quarter window depth really prevents them from becoming residential. It’s just too far, and the units end up too big or too skinny. You end up with a unit that’s 12 feet wide and 50 feet deep. It’s not very pleasant. So you have to work pretty hard to identify the buildings that have the right characteristics physically to actually work like a residential building.

Meghan McCarty Carino: And how much does your algorithm streamline that analysis?

Paynter: Before we did this, so pre COVID really, if you wanted to assess a building, you would select the building, work with the developer, and you’d probably spend three, four months doing a full analysis of it Look at how to plan out, do the design work, do your due diligence, and actually kind of 7 or 8 times out of 10, you do all of that. The developer put a financial model together and they just say, “Oh, it didn’t work.” We developed the algorithm to try and streamline that process and make sure that, you know, 10 times out of 10, we could find the right building, not 2 times out of 10. Because you can spend a lot of money in that first three months trying to work out the design and work out where it’s viable. And it’s also the frustration factor. A lot of these developers will look at one or two of them and just be like, “Well, we’ve spent a fortune. We’re not going to ever do this again.” And that really stops a lot of good projects from moving ahead.

McCarty Carino: So what factors does the algorithm actually weigh?

Paynter: Yeah, so it takes everything that we would normally do as designers. So it’s taken all of that design thinking and turned it into numbers. And that works well for residential because it really is a numbers game, you know. Units have a certain size, they have a certain proportion, you need a certain number of elevators to get people to those units. In different cities, you have a different number of parking spaces per unit, etc. So it’s, it’s all this kind of metric. So when we sat down to design a residential building, traditionally you start with the drawings, and you start with a spreadsheet, and you try and work out all of the maps at the same time. So what the algorithm does is do that automatically. It takes away the drawing portion and just runs the numbers portion first. So if you took a typical floor plate, for example, in those the average unit size for your location or your city, it would divide that unit size by the quarter window depth, so very quickly say OK, your unit ends up being 25 feet wide and 30 feet deep, really nice. You can get the bedroom and the living room or on the glass, or it will say, Oh, it’s 60 feet deep and 10 feet wide. That’s horrible. Let’s not go ahead. So starts to look at all of those data points and say, Is this good? Is it bad because it’s either not viable, like people won’t want to live there? Or it’s not viable from a cost point of view? And it just then aggregates all of that to give you an overall sense of yes or no.

McCarty Carino: So tell me a little bit about, you know, how this works in practice. How have you used it?

Paynter: Yeah, so there’s a couple of different things we’ve been doing. The first has been working directly with cities. So we can go in, we can analyze a whole city. We can analyze the vacancy. And then we can start to say, “You know, if you create a program of incentives, or you create a program of tax abatement or some policy changes around change of use, then you could start to bring this many units to these neighborhoods.” And that’s obviously very important from a housing supply point of view. But it’s also increasingly important from a tax revenue point of view. So using Calgary as the example, because it was, they’ve been moving the fastest on this, they realized they’d lost $17 billion of tax base in the last five years, and actually these conversions to residential could bring that back if they could do enough of them. So when we start to study that, we can go back to the city and say, “OK, you need to change this policy. And you need to put in this much money in order to make it viable for, for developers.” The other thing we’ve been doing is actually working directly with developers and landlords to look at the buildings they own and say, yes or no, is it viable? Or in a lot of cases, working with residential developers and helping them select or pinpoint the buildings in the market they should go and make, you know, make offers on or try and buy.

McCarty Carino: Something we always ask about when we’re talking about algorithmic decision-making on the show, is are there any ways that an algorithm could magnify bias, you know, reinforce problematic patterns that exist in the inputs in the real world? I mean, I think in this case, probably the stakes are a lot lower than in, you know, algorithms deciding medical questions and things of that nature. But I mean, have you observed any outcomes that are not optimal from kind of an equity human-decision-making lens that, you know, have you evolved the system in any way?

Paynter: And that’s actually a great question. And we were worried about that when we first started. So what we’ve done to try and counteract that is really just refine the algorithm so it only looks at physical characteristics. So there is no bias, there’s no opinion and there is literally pure math on measurements on ceiling heights, on that kind of thing. The one area where it then starts to stray into something that isn’t quite like that is when we start to look at locations. So it does look at which neighborhoods are more desirable. For example, which ones are better connected to transit and which ones are better, have better amenities and more walkable, etc. So it does then start to say, you know, if you did this building in this neighborhood, it’s going to be more valuable than if you did it somewhere else. And of course, that is then bias towards the best places to develop. However, that starts to get counteracted by the fact that the buildings in other neighborhoods are less expensive. And what we’re actually seeing, then is that kind of evening out of, OK, I can buy a building there less expensively. And I can start to do that work. And cities are working pretty hard to do that as well. They’re trying to create, in a lot of instances, two or three buildings that could, can get converted together. So you can actually then help bring services into areas that didn’t have them before. And you’ve seen in a lot of cities and incentivize that to try and get rid of those kind of food deserts or education deserts that you start to see in downtowns. Because bringing people there actually helped solve that problem.

McCarty Carino: Can you give me any examples of how this is being used? Any sort of projects that are, you know, in process now, and, and how it’s going?

Paynter: As I said, we’ve actually reviewed and scored, using this system, over 800 buildings now. And between 25% and 30% of those actually look viable for conversion. So those projects then move into test fitting. So we start to look at really what the units would look like. Are they too small, too big, or are they, they really kind of nice? And then quite a lot of those projects are now moving into development. A couple of really good examples that we have done over the last few years is Franklin Tower, which is in Philadelphia. Now that’s a full conversion of what was a kind of ’70s office building, reclad it, put new units in, created new layouts, new amenities in there. And that is now a really excellent example of this type of work.

We’ve got more on the push to convert offices into housing at our website, Marketplace Tech.

One city that has been a big focus for adaptive reuse of office buildings is San Francisco. The high share of tech workers going remote has hollowed out the downtown, and housing construction has lagged for decades.

Last year, Gensler, which is headquartered in San Francisco, conducted an analysis of dozens of office buildings in the city and found about a third of them were good candidates for conversion.

Marketplace’s Janet Nguyen reported on a bill being considered by the California Legislature that could clear some permitting and zoning barriers for these kinds of projects. And I reported on a public-private program in Chicago that offers incentives to developers to undertake conversions in a historic downtown office district.

One monumental project has already been completed there, and it required no incentives. The famous Tribune Tower was transformed into stunning and quite expensive luxury condos.

The future of this podcast starts with you.

Every day, the “Marketplace Tech” team demystifies the digital economy with stories that explore more than just Big Tech. We’re committed to covering topics that matter to you and the world around us, diving deep into how technology intersects with climate change, inequity, and disinformation.

As part of a nonprofit newsroom, we’re counting on listeners like you to keep this public service paywall-free and available to all.

Support “Marketplace Tech” in any amount today and become a partner in our mission.

The team

Daisy Palacios Senior Producer
Daniel Shin Producer
Jesús Alvarado Associate Producer