Michael Mande

Clara (0:00): What if the most valuable thing in commercial real estate wasn't a building, it was the data about the building? My guest today built exactly that. CompStak started as a spreadsheet in 2011 and became the industry's most trusted crowdsourced database of lease and sales transaction data. 40,000 members going strong, covering every major US market now paired with AI. Michael Mandel started as a broker, lost a $2 million commission on a deal, and turned the pain into a platform. Let's get into it. Michael, welcome.

Michael (0:36): Thanks for having me. Sounds like this will be fun.

Clara (0:38): So, Michael, you were a commercial broker in New York. You went to Babson for entrepreneurship. And at one point, you worked for the company that created the Yahoo yodel. That's kind of a wild path. Where does CompStak fit into all that?

Michael (0:50): Well, you know, for me, I've always sort of intended to be an entrepreneur. That was always my goal. My career path hasn't been too crazy, I would say, to get there, but I did work in commercial music production for a year after I graduated from college. I went to Babson to study entrepreneurship. Then I went and got the opportunity to work with this founder of this commercial music production company. And so, it was more like I wanted the opportunity to work with a founder and have that exposure, which was cool. It was a really interesting opportunity, although not the right industry for me. From there, I got into commercial real estate brokerage because it's a super entrepreneurial profession and I really liked the tangibility of buildings, particularly working in New York City and Manhattan, huge valuable office buildings, and working as a broker. You eat what you kill, right? Make as much money as you can based off of the deals you do yourself. I've always had an issue with authority and I never really wanted to have a boss and this was kind of the closest thing to truly being an entrepreneur was getting into brokerage, but always intended to start a company and leverage that experience as a broker to identify an opportunity to start a company.

Clara (2:10): You know working purely on commission as well, the pressure is entirely on you.

Michael (2:16): Yeah. You know I've never been afraid to bet on myself. Maybe at times I've been too confident or cocky or whatever the case may be. And I'm maybe old enough now to admit that and to look at things in hindsight and be like, you know what, I probably could have spent some time learning an XYZ thing instead of just assuming I could jump into it. But I've never been afraid to bet on myself and that's always been sort of my mentality.

Clara (2:36): You were sitting in those Monday broker meetings where everyone swapped comps on paper. At what exact moment did you think, okay, this is broken and I'm going to arrange to fix this?

Michael (2:47): I actually had a bunch of different ideas of ways to sort of transform the industry and it was a matter of deciding which one to go after, and the challenge around data and around comps felt the most acute to me and maybe an example of the most antiquated thing. Right? So, we'd have these Monday morning meetings. Every brokerage firm would have these meetings. We're sitting around a big boardroom table for two or three hours just talking about deals in the market. I was frantically calling up other brokers I knew in the industry on Sunday night to find comps to share in the Monday morning meeting all so that we could try to understand what's happening in the market. But I was doing tech deals, I was doing data center deals, creative firms, and most of the deals being talked about were banks and hedge funds and insurance companies, and most of it was irrelevant to me. And so it was just sitting in one of these meetings realizing, why are we doing this? There's got to be a smarter way to share this information rather than wasting time in a meeting every week. And the idea wasn't to totally upend the industry and nor have we, right? It was really just to make that process more efficient and move it from offline to online.

Clara (4:01): Is commercial real estate even a big enough market? How do you answer that in numbers?

Michael (4:07): Well, you know, when we started the company, we're now in prop tech, right? When we started the company, the term prop tech didn't exist. CRE tech or commercial real estate tech didn't exist. Originally I was working as a broker trying to raise money to start this thing when it was mainly just an idea and that was 2011, you know, and everybody's just like, is this market large enough? Is the commercial real estate market large enough? Which is such a joke, right? I mean, largest asset class in the world, or second largest. But that was the mentality. The hot thing at that time was called SoLoMo, social-local-mobile, and everybody wanted to know how we could be more like Foursquare or Twitter, like that's what people were interested in. And so we had to convince the investors that commercial real estate is a market that can sustain a VC-backed business that can provide the kind of returns that VCs are looking for. And it was definitely a challenge. It was really a function of having to not just raise money, but also teach investors about why this matters and what this is all about. You know, make it clear to them that no, your mentality around this is wrong. You don't know this the way I know this. Let me explain it to you. Real estate is bigger than gold. It's bigger than bonds.

Clara (5:31): How is it so hard to having to convince someone, right? But there was also COVID right a couple years ago and then commercial real estate changed a lot and it was paused for a little bit. How do you think it's recovering?

Michael (5:44): You know, we found that our business took a big hit in the second half of 2020 into the beginning of 2021, and then second half of 2021, our business took off. And this more recent downturn, we're now four years into it and it's still a slog. Things are getting better. But, when I always say, and I experienced this as a broker as well, up markets, high markets are good. Down markets are also fine and they're good. Things are happening. There's opportunities in a down market. But a market that's in transition is the worst. It's the worst for everybody. Transactions need to happen for people to do business. And that's the situation that we've been in. I think finally there's been a bit of a reckoning now, which is great. But we had the Fed saying what's happening with interest rates is transitory. And we saw a lot of pretend and extend from the banks as it relates to the loans they have on these properties. And the sellers were unwilling to sell at discounted prices. The buyers were unwilling to buy at high prices. Just everybody did nothing. And that's not good for most people in the industry who rely on deals happening to move forward. And I think that's been what's been the challenge in this particular downturn. But yeah, things are getting better. We saw a huge boom in the industrial market because of COVID. It softened a bit. Now that's stabilized and is coming back up again. Retail's in a pretty good spot. So interest rates are still affecting the market. There's still a lot of deals that were done in an extremely low interest rate environment that are still impaired. But now because there's a bit of a reckoning, that's good. There's more defaults happening in CMBS than any time since the GFC. That's also good. Now people are finally able to take advantage of some of these distressed assets and these deals, and it means the market will be stabilizing. Yeah, it's definitely been a challenging but interesting time to navigate.

Clara (7:54): Anyways, now we're going to get a little bit more into the data engine, crowdsourcing, quality, and trust at scale. Let's talk about mechanics. So CompStak is built on a credit exchange. So brokers and appraisers give you comps, right? They earn credits. They spend credits to get comps back. 40,000 members contributing data. How do you make sure that the data is actually good? What's your quality filter?

Michael (8:21): Sure. Well, there's a lot to it and it's advanced over time, but effectively it's a human in the loop model. And so we've got a team of analysts, we have 60 people in Belgrade, Serbia, as well as people in New York. We have a team of analysts that review the data that comes in to make sure that it's high quality. And that team is supported by technology. They always have been. And then over time the technology has played a greater and greater role because you're able to train off of the decisions made by analysts and based off of the data itself to train the technology to become more sophisticated in its review. But there's several key components to it. Some of it is a function of statistical anomalies, right? 100 deals in this building, this most recent deal in this building is showing a rent that's X percent higher on this floor than the recent deal we received on the floor below it. That doesn't seem right. That seems too high or too low, or this free rent amount or TI allowance doesn't make sense with similar deals on similar floors of that building or the competitive set of that building. And so we're able to flag data for review by our analysts to dig into it. We also score our members. So, it's not unlike your Uber rating where every time you get in an Uber, your rating is impacted. Every time you submit a comp to CompStak, it updates your rating. And so someone who's been contributing data for 5 years where that data has been deemed reliable and hasn't had to be updated, their rating gets better and better. Somebody who's brand new, they're going to have their submission looked at more closely. And we actually get most of our data multiple times. On average, every transaction we get three times. And so every time we get it is another opportunity to validate the data. And we can compare the different users who submitted that same transaction and know who to trust more highly than others. So that impacts what gets reviewed as well. But there's also now a heavy amount of AI in the process that automates a lot of this review. But at the end of the day the best way that we know that our data is high quality is through our clients. When we go and pitch all of our prospective clients and our current clients, they look at their own deals and they know for sure if those deals are right or wrong, and we're able to pass that test over and over again. Sometimes we'll have a prospective client who will come to us and they'll say, how many data points do you have? And we're like, well, you're asking the wrong question. Or they're comparing us to some other data company that doesn't do what we do. And they're like, well, they have records on X number of million properties. And I'm like, great, go buy their garbage data from the public records around however many millions of properties, or you could buy our data around a smaller set that actually relates to the deals you actually do and that is high quality. You may understand that quality is generally speaking more important than quantity, but sometimes when you're dealing with a prospective client, they don't quite understand it. They think that they judge you by volume and that's not typically how we're interested in being evaluated.

Clara (11:34): So you said that they ask you the wrong questions about data before they actually buy or make a deal. What are the right questions to you about data?

Michael (11:41): I think it's all about looking at the granular data. Show me my deals in my building, right? Because if the data is accurate at the most granular level, you can be damn sure that when you roll it all up at a higher level, it's accurate, too. And that's just the nature of the way data works. You always want to have it as accurate as possible in its most granular form possible. There's plenty of really smart economists and data analysts that can extrapolate interesting trends from dirty data at a high level, but those same people can do even better analysis off of clean data.

Clara (12:20): Yeah, why not use that option, right?

Michael (12:22): So, yeah, that's what I would say. You want to look at the granular data. Also, not all data is created equal, right? Understanding what is a commodity and what is not a commodity. I often find that a prospective client will benchmark our pricing for instance against some other commercial real estate data set like public record data as I mentioned before, which is absolutely a commodity. There are a bunch of public record data providers and we work with some of them, and I don't want to criticize them but it's a commodity business and they have to beat each other up on price which drives price down. And we are not in the commodity data business and you just have to understand that that's not comparing apples to apples.

Clara (13:08): Yeah. No and just like we're saying it's not about chasing volume it's chasing accuracy. That's exactly what we're trying to do at Tech Accelerator in CRE where a single lease decision can be worth hundreds of millions of dollars. What does the data error actually cost someone? Can you give me a real scenario?

Michael (13:26): Yeah, you know, I think that at the end of the day, it's often about making a decision if you're going to buy an asset or pass on an asset. Versus sometimes it's about the mistake you don't make versus the mistake you do make. I think that, for instance, we were talking about lease transaction data and I'm literally doing the math right now, but let's say you've got a 50,000 square foot lease, and you underwrite it at $50 a square foot in tenant improvement allowance, right? That's $2.5 million. And let's say you actually need to give $80 a square foot in tenant improvement allowance to get that deal done with that tenant. Well, that's $4 million, and that's a $1.5 million discrepancy. And if you're underwriting buying a building that has 50,000 square feet vacant and you don't know how much you're going to have to give in tenant improvement allowance, that's a $1.5 million swing in what it costs you to lease that space, which meaningfully decreases the value of that property and how much you should be willing to pay for that property. There's so many relatively small things like that that add up to hundreds of millions of dollars easily. And at the end of the day, particularly when you're talking about leasing data and commercial real estate, there are three ways that you can determine the value of a commercial real estate asset. There's the income approach, the comparable sale approach, and the replacement value approach, and the cost approach. But far and away, commercial real estate is valued based off of an income approach. It's based off of how much income does that property produce, and the rental data drives that, and if you can't understand all the rents in the building and what the rents should be, you can't value the property properly and you don't know what your returns will be. Often what we hear from our clients is because of your data, we didn't do this XYZ deal that we thought we were going to do, or we did do this deal, because whoever underwrites the highest rents is willing to pay the most for the property. And anyone who underwrites rents that are lower will ultimately lose to the person who underwrites it higher. And if you're going to be the top bidder, you got to be damn sure that you're going to be able to get the returns that you think you can get.

Clara (16:12): Yep. Well, I really appreciate this point of view on deal sourcing and real estate and also the entrepreneurship behind all this conversation. You cover every major US market. You know, CompStak can tell me what's happening in Manhattan, San Francisco, Chicago, Atlanta right now, right? So, how do you actually build that kind of national coverage? What's the data collection infrastructure behind it?

Michael (16:37): Sure. Well, it's challenging. Our data relies on this large network of 40,000 brokers, appraisers, research people who give data to get data in exchange, but it has to happen on a market-by-market basis. So we have to build relationships with individuals in every single market who trade to get other data from other individuals in the market. There are some people who have access to national data sets but for the most part these people do deals locally. And so we had to launch one market at a time and I remember the order, right? I mean we started in New York and then we went to San Francisco and then the greater Bay Area and then we went to LA and then Chicago and then DC and then Atlanta eventually. I lose track then, I think it was Dallas and then Houston, but we had to go one by one and build those relationships and build the database in each market. And then when we had enough markets, I think once we got to like 35 or so, then we were able to open up the whole country because those 35 markets had enough people in them that could start filling in the gaps in between them to get to the national coverage. But it took blocking and tackling, individually one by one going to each market and building those relationships before we could get to that point.

Clara (17:52): Yeah. And you know this question that I asked you, we ask the same question every day at Tech Accelerator on a global level as well. You know, how do you maintain data freshness and coverage across dozens and hundreds of markets simultaneously. So it requires an obsessive data operations culture that we're building here and it sounds like CompStak built exactly that, which I find so interesting. So thank you for sharing that with me.

Michael (18:23): Absolutely. Yeah.

Clara (18:24): But anyway, so you know, brokers have been protective of comp data for decades. It was their competitive edge. CompStak essentially asked them to give that edge away. Did you ever get real push back, threats, people refusing to play ball with you? And how do you think about the ethics of all this?

Michael (18:42): Well, you know, the interesting thing was and is that brokers have always traded this data and appraisers have traded this data, too. And the landlords, for instance, they don't typically trade data, but they rely on and work with the brokers who trade the data. And they know it's happening. They know that the deals in their buildings are being traded, and they know that that needs to happen in order for them to have the market intelligence to do deals and to be advised appropriately. And so, we never set out to fundamentally disrupt the way the commercial real estate market works. In fact, I don't think that's a productive way to build most businesses. I think that if you're going to go into an industry, you need to understand the nature of the way that industry works and work with it to evolve it and to transform it and improve it, not disrupt it. Nobody wants to be disrupted. So I think that was the key thing was first of all building a model that isn't actually disruptive and then two really communicate to all of our stakeholders that think they're being disrupted and explain to them that they're not, right? And it was really just about explaining this and also showing people this is better than what you were doing before. You were sharing random information to get other random information in return. Now you can share random information and get exactly what you need in return.

Clara (20:09): AI companies have become a noticeable driver of recent office leasing, especially in markets like we were saying in San Francisco, New York. From what you're seeing in CompStak data, is AI actually changing the trajectory of urban office demand or is it too small, which I doubt, to move the market in a meaningful way?

Michael (20:29): It's kind of yes and yes or yes and no, I don't know. These TAMI, which is like technology advertising media information, which is kind of how the real estate industry talks about this sphere of companies, is growing in a way that it hasn't in several years, and the biggest thing driving TAMI is AI. But it's not astronomical, you know, and it is certainly concentrated. You bring up San Francisco. San Francisco is a really interesting market. First of all, it's still far below where it was pre-COVID. San Francisco got destroyed by COVID and by the interest rate challenges and is still in worse shape than most markets and is still below where it was pre-COVID. And San Francisco is not that big of a city, and then you've got the greater Bay Area, and a lot of the tech companies are in the greater Bay Area and in the peninsula. And what's interesting is that pre-COVID you were seeing a lot of the big tech companies and even the VCs that were in the peninsula moving and expanding in San Francisco proper. And with COVID first everybody was working from home, but then when they started to come back, a lot of people moved out of San Francisco into the suburbs to get more space, and they were happy in more space rather than in less space. And so the tech companies, like the big ones, Meta and Alphabet, that had the big campuses in Palo Alto, Menlo Park, Sunnyvale, Mountain View, they have brought people back to the office in the suburbs and they've been growing, but they have not brought back their San Francisco campuses and they're not seeing people go back to the office in San Francisco proper. They don't want to. A lot of these people have families and they're in the suburbs and they're happy and they don't want to. But on the flip side, all the AI companies are going into San Francisco proper and they tend to be very young and they work the 996, so they're working 9:00 a.m. to 9:00 p.m. 6 days per week, and that seems to be a real thing and they're in San Francisco proper. But from what I'm seeing, that has not reflected broadly beyond AI. So yeah, AI has been a big push in San Francisco, but because it's only the true AI companies and not broader tech that is really going crazy in San Francisco, it's not resulted in as big of a recovery as you might have seen in New York, for instance. So we'll see what happens, but a lot of other companies, I think, are reducing headcount because of AI, too. So you've got the AI companies hiring people, but in comparison to all the other companies that are reducing headcount because of AI, it's hard to say.

Clara (23:25): Now getting into a little bit more CompStak's AI and the future of data intelligence. The main goal here, in December of last year, you launched CompStak AI. You said it takes your decade plus of trusted data and pairs it with AI to deliver instant intelligence. That's the pitch, right? But what does it actually do that your platform didn't do before? What's the real unlock that this brought?

Michael (23:49): Well, there's like a world of things that we're doing, some on the back end, some on the front end. One of the things we released is CompScout, which is a chat interface within CompStak. It's a highly modified one that doesn't hallucinate off of just generic data that it finds on the internet. Maybe it could hallucinate in some capacity, I haven't seen it hallucinate per se, because it's very tuned to leverage CompStak data and to actually fall back on CompStak's core filtering capabilities and only use external LLM data when necessary. And so it just makes it easier to find data. Okay, no big deal, finding data. But what I think is really cool is when it does leverage the capabilities of both. So I always use this example around using CompScout to find fried chicken restaurants.

Clara (24:46): Why is that a good example?

Michael (24:48): That's a great example. In the past, our analysts would always have to manually code every record in CompStak for what it is. So we would label, okay, this is a retail location. Maybe the industry would be restaurant, maybe it's inline retail, or maybe it's a shopping center. Nobody has the time to have analysts go through and tag a million and a half tenant records for whether or not they are fried chicken restaurants. And so now we can throw all a million and a half tenants up against an LLM. The LLM knows that Popeyes is a fried chicken restaurant, even though fried chicken isn't in the name Popeyes. And compile all of those, then cross reference that against our standard filters in CompStak. So I could say, "Show me fried chicken restaurants on the East Coast and tell me how the rents have trended over the last 10 years." And it knows what East Coast is. It cross references that against all of our CompStak markets on the East Coast. It uses the semantic understanding of what are fried chicken restaurants and then it looks at our data and it pulls all the rent data and it averages it out and it tells you how it's trending and it's the best of both worlds. It's leveraging the data we have combined with the power of an LLM.

Clara (26:12): It is. Oh my goodness. It's changed completely how we spend our time, right? Tech Accelerator has also been investing heavily in AI augmented data pipelines like you were saying, and the idea that raw company and contact tech data becomes intelligent when paired with AI reasoning. So what CompStak is doing in commercial real estate we're doing in B2B commercial intelligence. So very cool here to have this intersection between the fields and there's the same underlying principle. It's identical. Clean data is the prerequisite and AI is the multiplier, right?

Michael (26:47): 100%. Yeah. I mean, I think we find that with a lot of our clients that so many of our clients are trying to adopt AI with their data and they're realizing that their data is just not ready to work with AI in the way they wanted it to. And in real estate you have data normalization challenges that are a little unique. There's cleanliness around the data itself, the rows, the columns, the cells. There's understanding the entities, buyers, sellers, tenants, lenders. So you've got buyers, sellers, tenants, you've got lenders, all these entities, and then you've got the geospatial challenge, right? Because you've got buildings with multiple addresses. You've got challenges in normalization there that are quite unique to real estate and it's something we have to tackle for our day-to-day, but our clients struggle with.

Clara (27:46): Going back to the hallucination that you were saying, you know how AI hallucinates sometimes, there's that risk always. If someone asks CompStak AI a question about a lease comp and it gets it slightly wrong, that could cost someone a deal. How do you guarantee that your AI doesn't invent facts?

Michael (28:10): Well, that was why we built this chat agent ourselves internally first. The next step is we have MCPs and we're allowing people to embed CompStak into their own... you could use the CompStak MCPs and put them into Claude as well, but by building it internally it gave us a higher level of control and that's how we wanted to start. And so for instance when we quote you comp data in CompScout, we actually reference the specific deals on CompStak within the chatbot. And we don't just return the data in text rows in there. We give you a table and that table is the same table you would see in CompStak and you can click into each one of those individual deals. It's not like giving you a summary only. Well, it'll give you a summary but it references the specific data and the AI cannot manipulate the specific data. The data is the data.

Clara (29:03): Yeah. Yeah. Definitely. I appreciate your answer. And now wrapping up, this has been such a great chat, but you've been in commercial real estate data for 13 plus years, maybe even more. Where does CompStak end and where do AI agents begin? In 5 years, does a broker still manually pull comps or does an AI agent do the whole deal analysis autonomously?

Michael (29:28): No, in five years, I think we're all sitting on the beach and the agents are just doing all of our work for us. It's a good question. I think that fundamentally we're all getting disrupted in a very big way by AI. That said, I do think agents are going to communicate and provide a lot of data off the bat, but many people have called for the death of brokers forever. I mean, in the residential space, as Zillow grew and Trulia at the time, lots of people said, "Oh, nobody's going to go to a residential real estate agent. You can find all the listings online." Well, as it turns out, there are far more residential real estate agents today than there were when Zillow and Trulia got started. The data is super valuable. Having a professional help you contextualize that data is critical. I'm not going to lie, I think more and more AI is proving that it can do things that we all never thought that it could. But I would be rash to call for the end of brokers right now.

Clara (30:34): Okay. We'll see where it goes. But Michael, this conversation is proof of something we believe deeply at our entities at Data Sharp and Accelerator that the companies that win in the next decade won't necessarily have the most data, but they'll have the best data paired with the intelligence to make it move fast, right? So CompStak built that for commercial real estate. What you're all doing with CompStak AI, it's exactly where the whole data industry is heading. So, thank you so much for being in Conversations with Data with me today.

Michael (31:02): Thanks.