S01 E01
How data helped WeWork exit bankruptcy
Podcast available on
Data powers everything at WeWork—from enhancing member experiences to optimizing office layouts. Tom and his team analyze usage patterns to refine services, adjust spaces, and host events that resonate with every member.
Dodson’s leadership is all about collaboration. His data team is split into two: one group focuses on data modeling, while the other builds the tools that keep WeWork running smoothly. They work globally but adapt to local needs, integrating data across operations.
WeWork’s data stack includes Fivetran, Snowflake, Tableau, and Airflow. They’re also exploring advanced tools like DBT for data modeling. The setup may be lean, but it’s highly effective in driving results.
Dodson’s team played a critical role in WeWork’s successful exit from bankruptcy, providing insights that helped renegotiate leases and streamline operations. Their data-driven approach guided decisions on which spaces to keep and which to exit—ensuring WeWork stayed efficient and member-focused.
The team is currently migrating legacy systems to a new platform, tackling data accuracy and integration across platforms. Next up is leveraging AI and machine learning to elevate WeWork’s operations and member experience.
Tarush Aggarwal (00:00)
Welcome to episode one of People of Data, where we get to highlight the wonderful people of data. These are our data leaders from various industries, geographies, and disciplines, discover how companies are using data, their achievements, and the challenges they face. Today on the show, we have Tom Dodson, a technology leader with over 16 years of experience, who's currently heading data at WeWork. Tom has previous stints as an engineering leader at N26 and Capital One. For those of you who don't know, WeWork has just emerged like a phoenix from bankruptcy, ready to make some new headlines. This episode is especially special to me because I spent four years at WeWork before starting 5X, and it was one of the most positive experiences of my life. Tom, welcome to the show.
Thomas Dodson (00:49)
Thank you. Thanks for having me. Excited to be here.
Tarush Aggarwal (00:51)
Are you ready to get into it?
Thomas Dodson (00:54)
Yeah, I'm ready.
Tarush Aggarwal (00:55)
Let's do it. So, what does WeWork do? How do you make money?
Thomas Dodson (00:59)
Sure. So at WeWork, we provide flexible workspace solutions for businesses of all different sizes. So from freelancers and startups to large enterprises. Our core offering includes beautifully designed office space, co-working environments, meeting rooms equipped with amenities and different services. We focus on creating a community-centric environment that fosters collaboration and productivity for our members. Additionally, we offer enterprise solutions with customized office space to meet the unique needs of larger organizations.
Tarush Aggarwal (01:34)
And you know so I'm assuming the business model is more paying for access and think the space or do you also monetize in any more sort of in any in any digital avenues?
Thomas Dodson (01:43)
Yes. There are some digital avenues as well. We've launched a product called WeWork Workplace, kind of a post-COVID product where maybe you have 200 people, but not all 200 are going to the office at the same day, on the same time, right? So maybe you only need 100 desks instead of 200 desks. And having this ability to have flexible space in that way has been a new product that we've launched. That's more on the digital side.
Tarush Aggarwal (02:12)
And are there any sort of products you're moving towards, more digital products coming soon?
Thomas Dodson (02:17)
We're looking at some things as to how can we let some of the customers have some of their data on what's going on in their spaces, who's coming in, how often, that type of stuff.
Tarush Aggarwal (02:28)
Exciting. How does data play a part in helping WeWork with the vision and make money?
Thomas Dodson (02:37)
Yeah, so data is a real central piece for WeWork's operation and decision making. We leverage data to enhance the member experience, optimize space utilization, improve operational efficiencies. By analyzing the data on our spaces and how they're used, we can tailor the services to better meet the needs of our members, from adjusting layouts and amenities to the planning of events and community activities.
Tarush Aggarwal (03:01)
And are there any, if you look at the next six months, what are some of the big data projects and how are they going to make a material difference in the business?
Thomas Dodson (03:11)
Yeah, so a lot of the stuff we're doing now is to really understand our footprint and where we should be, what buildings we should be in, what types of services we should be providing in those buildings, while also really looking at what types of companies are in those buildings. What would be more useful for them based on the industry they are, the size that they are.
Tarush Aggarwal (03:31)
And how do you do this, you know, across, you know, we work obviously operating across, you know, 30 countries and, you know, multiple hundred offices and, you know, a large number of cities. What percentage of this is sort of centralized? What percentage of this is, is regionalized? How do you look at this across different regions of the world?
Thomas Dodson (03:54)
Yeah, so we operate a global WeWork and then we have separate franchises. The franchises will do things more specific and tailored for themselves. However, we as a global piece will do things tailored in the different markets and we'll look at what makes sense in New York versus California versus Barcelona.
Tarush Aggarwal (04:15)
Got it. In terms of your footprint, what does the data footprint look like? You know, team, infrastructure, any tools you're using on a daily basis.
Thomas Dodson (04:26)
Sure, so we have two lean but very mighty data teams that they're split up into. One is focused on modeling and the other is focused on the tooling. We have a pretty standard stack, I feel, utilizing Fivetran for ETL for the majority of our systems, Snowflake for our data warehouse, and then downstream Tableau for most of our reporting.
Tarush Aggarwal (04:47)
Gotcha. And how about anything on the product analytics side or orchestration? Anything around data modeling?
Thomas Dodson (04:56)
Yeah, so we're using a bunch of different things. Most of the systems we're building will be in Airflow kind of on top of Snowflake to utilize that data that we have there.
Tarush Aggarwal (05:07)
And anything inside product analytics or are you using some modeling tools potentially like DBT?
Thomas Dodson (05:13)
Yeah, so we're actually in the process of kind of redoing all of our modeling. So that's something we're looking at, especially with how we have the different franchises. We want to be able to provide data to them in a way where it's not all these constant requests that are all different. And we can provide it in a way that they can then utilize it the way that they need.
Tarush Aggarwal (05:32)
Yeah, and you mentioned you have two teams, you know, one focuses obviously on the modeling piece. You mentioned the other one focuses on tooling. Is this BI? You know, is there any sort of data science involved? And if you can kind of go into these two teams, you know, what would the split be?
Thomas Dodson (05:48)
Yeah, so there is BI, separate team from the data segment. Those sit kind of within the different functions.
Tarush Aggarwal (05:55)
Mm. got it. So you have two teams, one focuses on modeling and that's sort of presumably a lot of your Airflow stuff. You have some BI developers which are embedded into the business and then the second team when you mentioned it focuses on other tooling. What does that look like?
Thomas Dodson (06:08)
Correct. Yeah, so they're the ones kind of setting up a lot of these different systems, making sure everything in the infrastructure is running and optimal at that point. So something's going on wrong with Tableau, right? That's the team we're going to call to make sure that everything is up and running. The wrong data is in the model. That's when we'll go to that modeling team.
Tarush Aggarwal (06:37)
Interesting. And when you think about infrastructure, is it bundled in with, you know, sort of software engineering infrastructure or the data infrastructure group sort of sits sort of differently?
Thomas Dodson (06:50)
So they definitely sit differently right now, but it is something that I'm looking to try and bring a little bit closer together and try and marry the two, bringing a little bit more data into our software systems.
Tarush Aggarwal (06:56)
Got it. Yeah. And obviously the question on everyone's mind right now, especially to our audience of data practitioners is, what's happening in AI? How does AI play a part in the data strategy? And are any of these two teams focused on AI?
Thomas Dodson (07:22)
They are. I would say both of those data teams as well as the software team is focused on AI and they're looking at, or looking at, they're currently building systems that help internally as well as ones for members.
Tarush Aggarwal (07:37)
Got it. And any AI products you want to talk about today?
Thomas Dodson (07:41)
I'm trying to think of which ones I can talk about that are not still in development. We do have one that we've recently started using that has shown a lot of promise and being able to kind of tell us if a member is more likely to churn and how we can help to prevent that and what types of things would they need, how can we better service them.
Tarush Aggarwal (08:03)
And how would you measure as a business the impact this is having? As a data team, one of the things we're always thinking about is what is the ROI we produce? This is sort of churn is the sort of quintessential sort of day one data team problem. How are you looking at measuring this?
Thomas Dodson (08:20)
Yeah. Yeah, so. Yeah, we're kind of just looking at the numbers from the business standpoint as to how much churn is going on before and after and how we're able to actually impact the ones that we are predicting would churn.
Tarush Aggarwal (08:36)
And would you have any numbers for us in terms of having deployed this model? Has there been a meaningful decrease in churn or even being able to identify this beforehand? Or how many cases then is the business teams able to go prevent churn later on?
Thomas Dodson (08:55)
Yeah, so I could say that right now, the way that we're using it, we're looking to see how we can better service those members and then looking at ratings from them on that. And that has gone up over 100% in terms of their feeling of how they feel that we work. So that's what I can say about how we're modeling that right now or measuring that, I should say.
Tarush Aggarwal (09:16)
And would you have any numbers for us in terms of having deployed this model? Has there been a meaningful decrease in churn or even being able to identify this beforehand? Or how many cases then is the business teams able to go prevent churn later on?
Thomas Dodson (09:29)
Sure. So for this, I would probably have to say a lot of the work with the bankruptcy. So as you know, when you mentioned earlier, we worked successfully, exited bankruptcy last month. When we filed for bankruptcy, seven, eight months ago, our primary focus was renegotiating leases for all of the buildings. It's a monumental task that the real estate team had to take on, but they couldn't do it without leveraging data that's provided by the team. Using data analytics, the real estate team was able to identify which buildings to exit, where to reduce our footprint, where to maintain a presence, but really try to restructure that lease term. The strategic use of the data played a crucial role in having a successful turnaround and being able to exit bankruptcy so quickly.
Tarush Aggarwal (10:17)
Very, very interesting. We'll have to poke into that a little bit. So as data supported the bankruptcy filings, talk to me about what is this more on the BI side of it where you were supporting the legal and real estate teams? Are they looking at, my guess is, things like what is the sort of market rate on cost per square foot and you know sort of what did we get ourselves into or you know what sort of data at the end of the day was really you know is needed to basically sort of go deeper into this. Were any of these, you have data and very typically this would be, I'm guessing inside of your Tableau and you have some dashboards and occupancy and the buildings sort of around and cost per square foot and all of these, as you mentioned, a broad range of sort of metrics. Legal and the business teams got into particular leases, was this just you know BI support or was the you know analytical you know you're as an analyst actually you know diving into things with the business sort of you know getting into you know the analytical use cases and then again the obvious question on anyone else's mind is Did ML and AI have any sort of role to play in some of these recommendations, particularly naming to potentially what should we be paying or where can some of these members move?
Thomas Dodson (13:21)
Yeah, so definitely a lot of other teams helping out there. I would say the stuff that my team is producing is more of a foundational piece. Definitely don't want to take anything away from the real estate team and the legal team and actually being able to accomplish all of this because that's where the real great stuff lies. At least from the things that came from my team, there was not anything AI related for this piece.
Tarush Aggarwal (13:46)
Got it. Awesome. So moving on to our last question is, what's one challenge the data team is currently facing? We're seeing the data ecosystem evolve very, very rapidly. one of the things which we see data teams always on fire for is cost optimization. And now with the new AI world coming in, we'll have to understand what is one challenge you're currently thinking about which you're interested in solving.
Thomas Dodson (14:12)
Yeah, so I would say probably the biggest challenge and the most immediate pressing one. We're currently in the process of migrating many of our legacy systems to a partner system. These systems include things like, I'm sorry.
Tarush Aggarwal (14:27)
I'm guessing this is Yardi, who we were announced as the... Would you want to just give us a little bit of context on Yardi?
Thomas Dodson (14:29)
Yes, yes. Yeah, yeah. Sure, so we were working with them as a partner with a lot of their software systems that do similar things to what we did, which is why we're migrating over to those. But now them being a financial backer and kind of running things a bit more there on the board. So that's the two different kind of relationships. But for the piece that the data team is... is focused on moving those legacy software systems over. Things like billing, booking, inventory management, things that Yardi has done very well for a very long time. This migration requires the data teams to build new connections to these new systems, ensuring reporting accuracy, matching legacy data to the new systems for locations that have not yet migrated because we're doing it over time over different locations. And as you probably know, no two systems that are supposed to do the same type of thing operate the same way. So being able to identify where there is different problems, data definitions, creation, modification, and how those work within the two systems. Being able to pull all of that data is really where the team, I feel, shines. Because they're coming from all different systems, right? There's, I don't know, maybe 30 different software systems that we're running that have all this. Being able to collect that, have it in the warehouse in a nice way where the model could actually pull that out is the key secret sauce.
Tarush Aggarwal (16:03)
Wow, and you know, if I remember correctly, and please refresh my memory, coming out of bankruptcy, you were able to get rid of a few hundred leases. Am I on the money over there? And when you think about getting out of a few hundred leases, what, you know... Were any of these, you have data and very typically this would be, I'm guessing inside of your Tableau and you have some dashboards and occupancy and the buildings sort of around and cost per square foot and all of these, as you mentioned, a broad range of sort of metrics. Legal and the business teams got into particular leases, was this just you know BI support or was the you know analytical you know you're as an analyst actually you know diving into things with the business sort of you know getting into you know the analytical use cases and then again the obvious question on anyone else's mind is Did ML and AI have any sort of role to play in some of these recommendations, particularly naming to potentially what should we be paying or where can some of these members move?
Thomas Dodson (13:21)
Yeah, so definitely a lot of other teams helping out there. I would say the stuff that my team is producing is more of a foundational piece. Definitely don't want to take anything away from the real estate team and the legal team and actually being able to accomplish all of this because that's where the real great stuff lies. At least from the things that came from my team, there was not anything AI related for this piece.
Tarush Aggarwal (17:04)
And something I really like to focus on and we start focusing on is really developing the data product part. Because for me, building a data product, you really showcase how strong is your platform, right? Because when you build that, you answer a business use case, you have ROI, but then you are leveraging all the capabilities the team is already building and where all the stack is on top. So I think something that I really like to foster is to make sure that we
Thomas Dodson (17:35)
to make sure we build more data products, so with a concrete use case. And also, I'd like to make the user more and more autonomous on our stack. So meaning you're coming tomorrow, you're like, okay, I'd like to automate this task. I'd like to use the products you've done on AI because I want, let's say I'm a sales, I want to customize my sales workflow. I'd like these persons to come, pick the components on self-service part and try test.
Tarush Aggarwal (18:04)
more data products, so with a concrete use case. And also I like to make the user more and more autonomous on our stack. So meaning you're coming to more, you're like, okay, I'd like to automate this task, I'd like to use the product you've done on AI because I want, let's say I'm a sales, I want to customize my sales workflow. I like these persons to come.
Tarush Aggarwal (18:53)
adoption of this.
Thomas Dodson (18:56)
So looking at the product we built that is enabling everybody to use Gen AI where I told you we have a knowledge focus, which is on different departments in the organization, and another one which is on how do we build automation with that, with the customer care example I gave you.
Tarush Aggarwal (18:56)
So looking at the product we built that is enabling everybody to use the AI, where I told you we have a knowledge focus, which is on different departments in the organization, and another one which is on how do we build automation with that, with the customer care example I you. I think a next step of that is really to see what are the other workflows we can simplify. And I can think of one. We know we in
Anaïs Ghelfi (19:15)
I think the next step of that is really to see what are the other workflows we can simplify. And I can think of one. We know we are on multi-duet transactions. So we have a team that is focusing on everything related to payment delay, problems in financing and so on. So one of the next steps would be to say, okay, can we work with them and empower them as we did with the customer care?
Tarush Aggarwal (19:25)
So we have a team that is focusing on everything related to payment delay, problems in financing and so on. So one of the next step would be to say, can we work with them and empower them as we did with the customer care to really make sure we reduce efficiency? And then it's really easy to find ROI behind that because you look at the number of requests handled, complexity of requests and stuff like that. So that will be another use case. And of course you can expand it.
Thomas Dodson (19:43)
to really make sure we reduce efficiency. And then it's really easy to find ROI behind that because you look at the number of requests handled, complexity of requests and stuff like that. So that will be another use case. And of course, you can expand it to many, many aspects of the processes.
Tarush Aggarwal (19:55)
to many many aspects of the processes. That makes a ton of sense. Anais, thank you so much for being on the show and adding value to our listeners today. Thank you for having me. Of course.
Thomas Dodson (20:06)
Thanks for having me. Thanks. It's been a pleasure.
