Outcomes Over Workflows: Rethinking What AI Should Do in Pre-Construction

Podcast Transcript 

Steven Rossi – Zalmons: 

Hello there and welcome to the Construction Revolution Podcast. My name is Steven Rossi – Zalmons, and here on the show we explore the latest trends, technologies, people, and organizations that are revolutionizing and disrupting the construction industry.  

Today on the show I’m joined by not one but two fantastic guests. Mack Rush is Senior Associate of Emerging Technologies at Suffolk Technologies, and Zaid Kahn is Co-Founder and CEO of Neuron Factory. 

Mack came up in the field, running projects, and working in preconstruction and estimating some of the largest jobs in the country, before moving into Suffolk’s venture arm, where he now decides which new technologies get tested on live jobsites.  

Zaid arrived from the other direction entirely, having led the team at Microsoft that built the AI training systems behind partners like OpenAI, where the thing that kept delaying his data centres turned out to be construction itself. 

Join us as we discuss how AI and new technologies are revolutionizing the way that we think about preconstruction, and how they are working together to change the industry in the future. 

All right, welcome to the Construction Revolution Podcast. Thanks for joining, Zaid and Mack. How you doing today? 

Mack Rush: 

Really well. 

Zaid Kahn: 

Well. 

Steven Rossi – Zalmons: 

That’s great. I’m excited to learn more about both of you and your companies, and especially the partnership and what you’re building together. So to get us started, can you each tell me who you are and what you do, and what the company is that you work for? 

Zaid Kahn: 

Cool, I’ll start. I’m Zaid Kahn. I’m the co-founder and CEO of Neuron Factory, and Neuron Factory is a company that’s solving pre-construction challenges using a combination of AI and a specialized knowledge graph platform. What that really means is that we find construction does not really have a data problem — it’s really a context problem, especially in pre-construction. So you think about drawing specs, RFIs, scopes, revisions — it’s all fine on its own, but the problems that happen in between them is where we focus on solving, to see how the pieces relate to each other. And that’s really the difficult part. 

Mack Rush: 

And I’m Mack Rush. I’m a senior associate at Suffolk Technologies. I’m a member of our platform team there. What that means at Suffolk Technologies is I’m really the conduit between the field and the broader industry and the startups that we work with. At a lot of other venture capital firms, the platform function does a lot of things that help support portfolio companies, from marketing to recruiting to PR — all sorts of different activities like those. For us, we’re a little bit different because we’re sector-specific. We’re focused on the built world in general, and we really need to be able to walk the walk and talk the talk. I’m largely responsible for setting up those opportunities to get things out in the field, help people test and deploy and learn about how their products are being received on the job site, help people understand go-to-market dynamics, and try to be there with boots and a hard hat alongside all of our portfolio companies and the startups that we work with. 

Steven Rossi – Zalmons: 

Awesome. Thanks. So Mack, so Suffolk Technologies — obviously people are familiar with Suffolk Construction, it’s one of the biggest ones out there. So how does Suffolk Technologies differ from Suffolk Construction, and how do you work together? 

Mack Rush: 

Yeah, so this is a really unique partnership. And I think one of the things that can be confusing for a lot of folks is oftentimes we’re referred to as the corporate venture capital arm of Suffolk, but that’s actually not how we’re structured. We’re an independent sister company of Suffolk. We’re the venture capital affiliate. The way that Suffolk Technologies developed was in the mid-2010s, there was a handful of now quite large startups that came to Suffolk as an early adopter for product feedback. Some of those names are going to be widely recognized, like OpenSpace, EquipmentShare, Procore. And we found that we were enthusiastic about technology — it’s always been in our DNA at Suffolk Construction, but there was a moment when we started to look around and say, should we be a strategically focused corporate venture capital arm of Suffolk Construction and only invest in things that are relevant to Suffolk’s business? Or should we leverage this position and the brand ecosystem that comes along with being a general contractor at the center of the construction industry, at the center of a lot of these projects, as this connecting node to harness the rest of that network and really turn it into repeatable playbooks for startups to leverage those network effects, leverage the credibility of a large general contractor that has been able to put these technologies to the test and provide them with really authentic proving grounds. And we decided to go with this kind of hybrid model — a huge shout-out to John Fish, the CEO and founder of Suffolk Construction. I think it was a pretty non-traditional idea for him to still invest a material amount of the capital stack and take on a position as a GP, but then to allow other strategic firms, other wealthy individuals and private investors to join into this $110 million fund. And it really provided us an interesting piece of leverage where we could access the broader network of Suffolk Construction, thanks to our partnership with the general contractor, and really leverage a lot of its infrastructure. But we were also afforded this ability to move with the speed and autonomy of a standalone venture fund, which is something you often don’t see with venture capital functions within larger corporations. So everybody knows Suffolk as the roughly seven-to-ten-billion-dollar general contractor, and that has been a huge help for us in allowing us to not just grow our own firm and our capabilities, but really amplify our startups with that network as well. But to answer your question directly, what that really allows us to do is we have this really close access to experts. And that doesn’t necessarily mean that we have to trust what the experts say in every single situation, because — as we’ll talk a little bit about — this first-principles approach, questioning the way that we do things, has allowed us to harness this mix of independence and access and really provide startups an interesting place to explore how their products can change the industry. But fundamentally it’s just that access to the real world, the access to the job sites, that I think is a primary advantage for us. 

Steven Rossi – Zalmons: 

Yeah, absolutely. That’s great. So then for the next one, I’ll start with Zaid. How did you discover Suffolk Technologies, and how did you start working with them? 

Zaid Kahn: 

Yeah, I think when Salil and I co-founded the company back in February of last year, we had kind of a thesis that agents — and there’s going to be a ton of them, and they’re going to do different things, and various industries will take advantage of that. But the problem is going to be that there’s going to be a lack of context between them, right? And that’s just the way agents are — I could go into that in depth, but probably don’t have time today — but we realized that there’s a separate layer on top of it called the knowledge graph that we were starting to think about and how to build that. And what got us into construction really is because there’s not many places out there that are adaptable to the knowledge graph concept or technology. You have to have an ontology, you’ve got to have a repeatable pattern. So construction was one of those spaces. When we started looking at it, we came across Suffolk through the Boost program, and that’s how our relationship started. We realized that we needed a partner that could really embrace us, get us in front of experts and in front of operators in the ecosystem. We knew that a thesis is one thing, but actually getting it out there and trying it and learning how everyone is working in their daily lives in pre-construction is what we needed. And that’s really where Suffolk stepped up and took us in — like outsiders from the industry coming on, and said, let me show you how it’s done. And it was awesome. I mean, the experience we had at Boost — I would say for any founder it’s the most amazing experience you can have. And there are not that many good accelerator programs out there. I think Boost is probably one of the finest. 

Steven Rossi – Zalmons: 

Awesome. Mack, can you tell us a little bit more about what the Boost program is like? 

Mack Rush: 

Yeah, of course. So there’s a lot of differentiating factors that Suffolk Technologies has behind it, and the Boost program is one of our flagship differentiators, I think. The way that the Boost program is structured is — there are many accelerator programs out there where they will teach you more or less how to build a startup. It provides you generalist advice on putting together a pitch deck, setting up how you raise your fundraising round, how you organize your cap table and the way that you need to work with lawyers, etc. We have plenty of opinions on those things, but we also have really differentiated access. I think the thing that really differentiates it is this access to operators. The Boost program started in 2020, after we had launched Suffolk Technologies as a brand — it really started during COVID. All of the job sites got shut down, and we were looking for a way to energize our employee base at Suffolk. This originally started as an internal-only program. With everybody remote, we decided to search through our company for top performers and offered them up to the startup environment and the construction ecosystem to apply to come work with us during this program, where we would kind of tee up opportunities to test these technologies once our job sites reopened. But this would largely be kind of a brainstorming exercise for a period of, I think it was four weeks at the time, where we would just pair up our subject matter experts with founders and see where it led. And that program matured really, really quickly. And I mentioned before that the general contractor sits at the center of a very large network, and it was a fantastic place for us to be, because as startups started to apply to work with us, we started to realize that there were so many companies that needed access to these experts and to these job sites, but they just couldn’t find them. And oftentimes we were looking at these startups coming in, and some of these founders — we were so impressed, we just looked at ourselves and said, we don’t actually have the capability to help them with our own internal knowledge, and we can’t do this alone. So I have to shout out, we’ve developed this concept of operating partners of the Boost program. These are other construction stakeholders — we like to call it a vertical slice of the industry. We have 65 of them this year, in the seventh iteration of Boost. And they represent everything from institutional owners to building products manufacturers, architects, engineers, international builders as well as smaller regional builders, specialty trades, you name it. We want to bring this vertical slice to the industry to help founders solve these problems. And this program just took off like crazy as soon as we started to open up the doors and collaborate with a lot of our peers. So that has allowed Boost to gain a ton of credibility with founders. And when the Neuron Factory team came into the Boost program last year, it was obvious to us that they had world-class technological talent. And we started to look around at this roster of incredibly influential companies in the built environment that had this capability to open up their doors as well. And to Neuron Factory’s credit, they took full advantage of that. I mean, we got on calls nonstop. We got to job sites whenever we could during that program. It’s a short eight weeks, but one of the things that we like to do is we still pair our startups up with an individual — I was lucky to work alongside — well, I guess Neuron Factory got lucky, they got two of us. They got John Riggs, who was leading a data center project for us down in Texas at the time, and he was working on a very big project in New York. So John Riggs and I worked together to collaborate with Neuron Factory over those eight weeks, and after those eight weeks, that really turned into this organic partnership from there. But every single one of those eight weeks had a specific goal in mind, and we had to break it down week by week. It was like a season — this week’s game, we’ve got to go, here’s the plan, we’ve got to go execute it, right? And so that was what we did every week for those eight weeks. And where we ended up — I don’t know that we could have known that from the beginning, but we made that bet on this technical team, and I’m really excited about it. 

Steven Rossi – Zalmons: 

Awesome. Thanks. Yeah, we’ll touch on the partnership in a second, but before we get to that, I’m curious to get Zaid’s perspective — Mack mentioned it’s a pretty intense eight-week Boost program. From a founder perspective, how was that? Obviously everything in a startup is pretty intense and fast, but how was that? And then also I’m curious — the access that you got to Suffolk, and also, as Mack mentioned, the network of broader construction companies — how did that help you accelerate and evolve the business? 

Zaid Kahn: 

Yeah, I think, to sort of underscore Mack’s comment here — I think one of the things that is missing in this industry is that, when we step back, a lot of tools that were developed, or software that was developed, were like: “Hey, I’m a builder, I come from the construction world, I understand it, so I’m going to go build this piece of software.” As great as that is, you’re great at the construction side, but not necessarily at the software side. There’s a lot more complicated problems on the software side. And this is where we understood — we have a deep respect for people on the construction side, but I think there’s a turning point where the construction industry kind of needs to say, “Hey, how can we bring in talent that’s really good in software, or Silicon Valley, or whatever, to come in and help us?” And that’s the door-opening that I think Suffolk did — they merge the two, and we saw that as the welcome. I think the industry kind of needs that, because if we didn’t have Suffolk, we would build this software in isolation, without access to the people, the various personas, the construction sites, job sites, all that. So to me, that’s the most valuable thing I would advise for any software company who wants to get into construction to do, right — is getting, from day one, our philosophy has been “build with the customer.” That’s pretty much what we were doing with Suffolk at Boost — building with the customer. And that’s kind of where the construction industry needs to open up, so that you can welcome other people who bring in a great technology stack, because I think the industry needs it. 

Mack Rush: 

Yeah, and maybe just to hit on this again, because I think particularly in this world where AI has allowed people to build software really quickly, it’s actually a fairly common strategy for founders to find design partners early on. And we see companies apply to Boost all the time touting the different design partners that they’ve selected, right — oftentimes it’s one or two. And one of the easiest mistakes for a lot of those founders to make is just building exactly what the team tells you to build, and not having this first-principles approach to understand exactly the problem you’re trying to solve. And that’s also something that I think people in our industry suffer from as well — for those of us who are very operationally minded, we tend to say, “No, no, this is the right — I have the right process, I have the right way to do this.” And it’s almost militaristic, very, very hierarchical — follow the orders, just follow the process. And I have to give the team at Neuron Factory a ton of credit — they built trust with us so fast, because rather than leaping on every idea that we had and building it as fast as possible and showing us, “Hey, look at what I did, look at this product that I just built for you — isn’t this exactly what you wanted?” — they started at such a meta level of conversation. The level of depth that we got to around why the industry is structured the way that it is was really refreshing when we started the Boost program. And Zaid, I bring this up a lot when people talk to us about you guys — when we started the Boost program, I think the first conversation we had with one of our operating partners, who was a design firm, was: “What is a drawing?” That was the starting point. And for many people in our industry, you get asked that question and you’re like, “Why are you asking me this question?” But it was such an interesting way to start this, because we started to peel back all these layers. It’s actually a much more profound question than it seems at first glance, because a lot of people think this is the way that information should be communicated in construction, when in reality it’s a different form of contract, and it exists and looks the way that it does for very particular reasons that were developed over time — decisions that were made, in this case probably hundreds of years ago, as to what the right way to draw a blueprint is. And they didn’t come in with any assumptions about that stuff. They just said, “We need to figure out why this system works the way that it does. Why is 2D preferential to 3D, and in what situations? Why are people using these tools the way that they are?” And that pursuit of knowledge, that fundamental knowledge, gave them such a leg up so quickly on how they could build this product the right way. And I think we’re starting to really see the benefits of that. 

Zaid Kahn: 

Yeah, if I can just build on what Mack’s saying — I always say Salil and I, because we don’t come from construction, we like to say that we’re here to ask stupid questions, because we don’t understand it, and that actually is a superpower, because you get to ask, “Why? What is it you’re trying to do? Help me understand basic things.” And yes, that may be — you didn’t know that stirs up our questions of “why do you do it that way?” So for example, when we looked at drawing intelligence — everybody’s doing takeoffs and drawing stuff — and we sat down and said, “Well, what is a drawing? What is it supposed to do? What do you want the outcome to be?” And it made us — those conversations actually shaped how we tackle drawings differently. Whereas we don’t do object detection — we decided that object detection was really not what people are looking for, because one of the operating partners we talked to said, “I don’t need you to tell me how to count doors and windows, I can do that. I need to understand things.” So we came up with scene understanding. That means when you look at a drawing, you try to understand site, building, floor, door, and the relationship was more important, and the context. So that actually helped us build that feature of the product in a very different way than most people have, and that came from asking these really basic first-principles questions. 

Steven Rossi – Zalmons: 

Yeah, that’s great. I think it’s obviously a great approach for any business — building through your customer — but in construction specifically, and to your point about drawings, especially coming from the software side, you can see you need to understand what it does, so you can splice it up and get the data to put into the software, right? So it makes a lot of sense. So moving on to the partnership that you guys have developed — since the Boost program, obviously you had an initial partnership, but we just had an exciting announcement of your further partnership. So can you share a little bit more about what that entails and what that means for both of you moving forward? 

Mack Rush: 

Yeah, happy to take this one. So there’s a couple of different ways that we’ve decided to partner. Obviously the Boost program being the first. We also participated in reinvesting in Neuron Factory beyond the Boost program, alongside Zakua and Ahmad Ventures and a handful of others — really a rock star group of investors that we decided to reinvest with. And then beyond that, really what this design partnership is about is exploring the ways that AI is going to change the pre-construction function in an organization. There is so much promise in what the Neuron Factory team is developing, and both of our firms need to continue to learn alongside each other. And so the format this is taking is — we have run a number of pilots, I think at this point it’s six to seven distinct projects that we have used Neuron Factory on, and we are on the cusp of this next phase, where the volume of information we are going to be putting into a knowledge graph is going to be orders of magnitude larger. And this is a super important phase for us, because it’s a little bit about the complexity of the jobs, it’s a little bit about the volume of the information — but a huge piece of this for us, and Zaid, I’ll let you speak to your perspective on this too, but a huge piece of this for us is understanding how AI can really change the way that these teams are built, and the way that these roles need to look in the future. I think it’s very easy for people in pre-construction to associate the tasks that they do with their identity — as, “I’m an estimator, and an estimator must do takeoff. I’m an estimator, and an estimator must write our Exhibit B’s, the scope of work. I’m an estimator, and I have to use BuildingConnected this way, or I have to call these subcontractors that I know.” And none of those things are necessarily true. Just because you work in pre-construction doesn’t mean you have to do takeoff. Takeoff is a result of the outcome that you’re trying to achieve. Your role in pre-construction is to figure out what we’re building and what the best way to build it is. And that is very different than a set of tasks that you have associated with your identity and your role, right? So for us, there’s a lot of different theses that we have that we need to test with really complex, really challenging projects with rock star teams. And we need to question everything from first principles. What is the right way to staff a pre-construction department? How should this information flow from BD to legal to pre-construction, to project management, and how should information flow back and forth across all of those different groups? I didn’t even mention design — as a lot of these ENR Top 25 builders become more vertically integrated, the fluidity with which information and context pass between each of these departments is so critical. And I wouldn’t say anybody really knows the right structure for how to manage that right now. We have plenty of ideas, but we need to really test this, and I think that’s really what this next design partnership is about. 

Zaid Kahn: 

Yeah, and I think what we discovered in this first phase of the pilot project that Mack was describing — the other thing Mack mentioned, like how other people are doing takeoffs — what I think is the challenge in AI today is: “Hey, I’m building this neat piece of AI software, I just need to map it to your workflow and let’s go.” Well, that’s actually a fallacy. I think that’s not going to work, because — to Mack’s point — you’re not changing the way you’re working, you’re not really maximizing what the AI can do for you. So we are focused more on outcome. In the last six pilots that we did, we focused entirely on: what is it that you want at the end of this, or this portion of the build process? What do you want? And we constantly ask that. So when we do our feature releases, they are not based on, “Hey, here’s your workflow that I just came up with.” No — it’s actually an outcome. And if we fail at that outcome, we redo it and make sure the outcome matches what that persona in pre-construction needs. And that’s another thing we found out as we were building out our platform with Suffolk, but also with our other customers — that there are so many other personas inside pre-construction that are sitting and waiting on, or are part of, a process that waits around for other people. For example, when we were first building the platform, we thought we were building for an estimator, because that’s what we knew. But it turns out the first person who logged into the platform when we released it was actually a lawyer, not an estimator. So we said, “Well, why is the lawyer logging into the system?” And the reason was because the lawyer said, “Well, normally I have to wait around and ask the estimator after ten days — what have you produced that I can go and understand the risks of this project?” Well, guess what — now they can get that in about an hour, because the documents are uploaded, risks identified very quickly, and it saves the estimator’s time from not even having to think about what they need to send to the lawyer — okay, now how do I get to the next part of my job? So those are some of the examples. We question everything, because it’s like, okay, I shouldn’t think about how I got to this takeoff or whatever — I just want to know what my outcome should be. If I want the outcome to be a measurement of walls, windows, doors — that’s more important than how you get there, whether you’re counting or measuring in a different way. Those things don’t matter. Let’s focus on the outcome. So that’s really the approach we’re taking. And I think what we’re going to do next in the partnership, what we’re really excited about, is we’re going to go and actually look at a site that’s being built and actually — 

Steven Rossi – Zalmons: 

Awesome. So maybe we can dive into that a little bit deeper. For people who are maybe not as familiar with the pre-construction process — Mack, from a contract point of view, what does that usually entail, typically before implementing something like Neuron Factory? And then, Zaid, maybe you can tell us how you go about changing that. You touched on a couple of personas and things, but specifically for those pilot projects, what were you taking on, and what data are you bringing together to change that? 

Mack Rush: 

Yeah, happy to kick off from this perspective. The role in pre-construction might look a little bit different depending on the type of work that builder is taking on. Suffolk is very fortunate to have some fantastic long-term clients, and we’re very lucky to have a lot of repeat clients, negotiated work — it’s what a lot of our culture and our client service has been able to afford us over time. And a lot of those trusting partnerships will provide us with information very early on a job, when the information is super sparse. And a quick shout-out to another of our portfolio companies, Edify — I was actually one of the earlier members working on that software platform, but that was originally developed in-house at Suffolk as a way to manage our estimating data and a lot of our pricing across all of our different regions. Generally, the role of pre-construction at our firm is to try to help these owners get a best estimate for what this work is going to cost and what the right way to build it is. And depending on the level of information you get, that could be super conceptual — all you have is some vague geometries, a back-of-the-envelope sketch, as we like to say. And other times it’s a shotgun start — these are the bid documents, hard bid — you have to get a dialed-in number, buy out the subs, and get the owner your best bid as fast as possible. So you have this real range of different workflows that appear throughout a pre-construction department — sometimes it’s a long-running consultative experience between the builder and the client, and other times it’s more of a competitive bid situation where the documents might be a bit more mature. But this concept of document maturity is a moving target — I think there’s no objective way to understand how mature drawings are, and that creates a lot of issues in pre-construction, because it makes our job super subjective. So the primary role, from Suffolk’s perspective, of the pre-construction function is to divide up the scopes of work and allocate responsibilities for executing scope to various different subcontractors. That includes helping them understand what falls within their boundaries as a specialty trade partner, and what the transitions need to look like between them and another specialty trade partner. And when we think about what makes a really good contract — because a lot of people forget this, they think of a lot of these big builders as builders, and they forget that we are contractors, and a lot of our core responsibilities are writing these contracts. We really want our pre-construction teams to focus on writing great contracts, really clear contracts. In our subcontracts, we have a section called the Exhibit B, where we describe the different scopes of work that will be allocated to particular specialty trade partners. And when we think about what goes into a great Exhibit B, or how we write the best scope of work possible, we always need to rely on the contract documents, because those will always be getting some feedback on that. So what goes into a great Exhibit B? We have to keep in mind that the documents have a stated precedence, and that we need to rely on the documents that were produced by the professionals who carry the professional responsibility to produce these documents. So we shouldn’t be writing anything into a contract unless it’s not clear in the contract documents, or it conflicts with the way something is currently designed. But what often happens with a lot of estimating teams, because they’re pressed for time, is they end up regurgitating information that’s already in the contract documents and already represented. And what that does is it opens us up to new risk. And in addition, because people have opinions on the right and wrong way to do work, sometimes they need to get creative about how they write the scope of work, but not everybody is well-versed in all the different ways you can interpret written word and language. And oftentimes that results in clauses being written that open us up to unnecessary risk. Now, along comes large language models and artificial intelligence broadly — these tools are exceptional at understanding the way that language and communication comes across between different parties and how it’s interpreted. So our role in pre-construction teams is primarily to write great contracts by understanding the drawings, understanding all the materials that go into it, and understanding how the different specialty trades work together to produce a building. And a lot of that effort is reflected in how we actually write these contracts and distribute them and allocate the responsibilities across the specialty trades. 

Zaid Kahn: 

I think what Mack is hitting on is a really good example of an outcome that the contract, that the team needs to work on. But it’s not one piece of a document — it’s thousands of pieces of documents in different stages that come together to that answer. So here’s where I’m going to talk a lot about the technology, and why we are building what we’re building. Because that’s a great example of — an LLM cannot retain all this context. The challenge with LLMs today is they have a very limited amount of bandwidth to work with. There’s a GPU, and next to the GPU sits this piece of memory called HBM, which has a very tight window — it’s called a context window, created with something called a KV cache. And you can normally succeed with ingesting emails, pieces of documents here and there, maybe like a hundred — but once you get to thousands of documents and drawings especially, you basically blow up that context window. So this is where you just can’t simply take a contract, or a set of documents, and shove it in there and say, “Give me the best answer” — because you’ve got this whole corpus of documents, as Mack talked about, drawings that happen at different times. Where do you store that context? So that is kind of where Neuron Factory distinguishes itself from every other AI software provider, because we take all that context and build it into our data, into our knowledge graph. So when we go out and ask agents — “Hey, go look at this piece of the contract that specializes in this section of the Exhibit B,” or the risk agent, or we have fifty-plus agents — they then go and retrieve that context from the knowledge graph, as opposed to relying on the LLM’s context window, or the agent’s context window, which is very limited. And that’s kind of how we go and develop that Exhibit B outcome that is as accurate as possible, that Mack’s team would want, and then the team can focus entirely on making that contract as bulletproof as possible, because — to Mack’s point — that’s key in deciding what the risk would be, who does what, and where we should lean on less or lean on more. So that’s a really good example of us learning about an outcome and applying it to figuring out how do we create all this context across documents, drawings, specs, RFIs, and then put that into a contract — which cannot be done by a single LLM. That’s what we’ve determined, and we’re seeing results in that. 

Steven Rossi – Zalmons: 

Great. 

Mack Rush: 

Maybe, Steven, for some of the construction audience that maybe isn’t quite familiar with the KV cache, I think it’s useful to talk about the moment that we really had our lightbulb turn on with Neuron Factory, because I think that will hit close to home for a lot of the really core construction audience, and hopefully for the people with direct pre-construction responsibilities. I could provide examples of big-ticket items and things that you’re going to say, “wow, that’s a really big risk,” but I’d prefer to talk about something that’s super nuanced, that really requires great context and excellent memory across all of these different documents. So the moment for us was — we fed Neuron Factory a historical project that never went to groundbreak. It got very close — we had awarded over 80% of the scopes of work, but we just never got notice to proceed, and unfortunately the project didn’t go forward. So we had a very complete set of pre-construction outcomes. Our first test with Neuron Factory was to feed all this information in explicitly, just the contract documents. They built the knowledge base, they crunched on those, they understood those documents, and they built this project-specific knowledge base. We then provided them with all of the subcontracts that we had written for that project, and we said, “Analyze these subcontracts and help us understand the risk embedded in each of these subcontracts, and if there’s anything you would change.” Now, as part of this pilot, we intentionally fed it a bunch of Easter eggs that didn’t fit — some of them sounded more plausible than others, but the program had no problem catching those. And again, I don’t want to bring those up as the example, because it had no problem — it found every single one of them. The thing that really made us realize the power of this tool was when it identified a conflict within the concrete subcontract that noticed language around crane jumps was not very specific. And it particularly understood that the logic and the sequence of construction was going to require the contractor to assume a certain number of crane jumps, and there was a number that could work with the schedule, and then there was the language that we had written, which was very vague. And that sequence, the way that we described it, opened us up to a ton of risk for that subcontractor to claim in the future that they made the wrong assumptions, or that they priced it wrong, and that they didn’t assume the right jumps. But we actually had the capability to tell them the types of jumps they needed to make and the sequence they needed to hand over the crane. There’s the responsibility of who owns the crane, and there was no allocation of the responsibility of when that crane was transferred to another subcontractor. Generally, the way our work needs to be bought out is: the concrete contractor, if they have the most people on site, owns the tower crane. If the next-largest trade — when the concrete contractor rolls off — is going to be electrical, then the electrical contractor needs to use the crane to get some heavy equipment up into the roof, then the electrical contractor needs to take that on. Some of these handoffs and nuances between the different subcontracts — when something changes in one subcontract and it isn’t reflected in another — those are the types of situations we need to be alerted of immediately. I think when it got to that level of nuance and understood the construction logic, that was the moment for us that we were like: this is beyond any capability of a generalist LLM. This is beyond any capability we’ve seen — forget about takeoff tools — this was something that understood the project at a level of depth that even many estimators couldn’t reach. With the amount of work we’re winning, and with the amount of people retiring in the construction industry, you have so much institutional knowledge that’s leaving the industry. And we need these systems to be able to think through, logically, the way that work needs to be handed off between one subcontractor and another, and the way that job site conditions need to inform the sequence in which work gets built. And it’s not necessarily that Neuron Factory is producing the schedule — maybe you guys will get there at some point — but within the set of context that’s provided, being able to pursue context and ask the right questions to understand when something is missing and something isn’t — that is a leg up that you get in the contracting space that I think we haven’t seen from anything else. And that was the moment the lightbulb turned on for us. 

Steven Rossi – Zalmons: 

Nice, that’s great, that’s a great example. So last question specifically on Neuron Factory, and then we’ll move on to a couple more general questions about the industry. Zaid — like Mack’s example there, obviously you’re more than just contractual analysis software. So what are all the data types you’re taking in? And also, every project is so different — the documents they have, every contractor is different, the way that they name things. How do you go about actually teaching your models for each project, but also just generally, how do you go about teaching and evolving it when every project is so different? 

Zaid Kahn: 

Yeah, so we look at it as every project has its own graph, its own knowledge, and we embed it in there. Then we also have the ability to look across projects. Our vision, ultimately, is to be the project historian. We look at a project, understand everything about it, build the knowledge around it with the graph, and retain that as much as possible. And then as we grow with a GC, we get to build understanding across other projects, and build a graph across that. Essentially, that’s how we do it. We don’t share information across organizations, obviously — we keep it all within, we’re all single-tenant. We try to retain as much as we can to understand, and become really the project historian that a GC would look to at the end of the day — that’s what they’re looking for. As Mack was saying, if people are leaving the company, retiring, all these things — that knowledge has to be stored somewhere. And that’s really where we’re focused more on what data can we capture. That’s not really our goal in itself — it’s the means to the end. 

Steven Rossi – Zalmons: 

Right, good. 

Mack Rush: 

Maybe one thing to add to this — as investors, I think one of the things that we, and other investors, should really be focusing on, and honestly construction companies too, is: if you’re going to invest your time and money in technology companies, there are companies that will help you differentiate. For many construction companies, when you ask them what their competitive advantage is, there are varying degrees to which you can make that argument, around capabilities and business functions that they have. Some are more vertically integrated than others, some have particular capabilities and scale that give them different advantages, be it on direct procurement or self-perform. But there’s always been this element of being a builder where your people are your competitive advantage, and the relationships that those people have and the institutional knowledge that they have — the local ways of getting work done and working with particular trades. And what a tool like Neuron Factory really allows you to do is turn your company’s knowledge into your competitive advantage, in a scalable way. I think that for particularly the last era of technology companies and software in the construction industry, so much of the value — there’s been a ton of progress, don’t get me wrong, a lot of progress, and technology has helped us in so many ways. But when you look at the actual value and where it’s accrued, the vast majority of that value has accrued to the tech companies themselves, and we haven’t seen a material impact on productivity or minimization of risk in a really substantial, tangible way. I think this is the moment, with tools like Neuron Factory, that we can see really strong differentiation between different builders, by building these internal knowledge graphs and leveraging the technology that Neuron Factory has to up-level all of their people and scale their knowledge. And I think that’s a really important point for people to focus on as they evaluate any new technology within their business. 

Steven Rossi – Zalmons: 

Right, that’s great. So last question before we wrap up, and you’ve kind of led me right to this, so that’s great — the construction industry has historically had a reputation for maybe being resistant to technology, slow to adopt it. But obviously AI and software technology, to your point, is coming — well, it’s here, and it’s definitely starting in construction more and more. I’m curious to get both of your perspectives — do you think that reputation is valid, and how do you see that changing? And what excites you the most about the future of the construction industry using AI and all these newer technologies? 

Mack Rush: 

I think the reputation of being slow to adopt technology is just as much a reflection on the technology industry as it is a reflection on the construction industry. I don’t think the technology has been ready for just how complex and unpredictable and multivariable the physical industry of construction truly is. And I don’t think a lot of people understand just how complex a lot of these stakeholder networks are. A lot of people assume that general contractors have this ability to mandate things and push things down to the supply chain, and they don’t realize that there’s a group of 30 to 50 companies that comprise any given construction project. They don’t even realize that distinct construction projects operate like a distinct business, with a distinct tech stack, with its own P&L — all of these things are set up as LLCs, most of them, so these little small businesses, some of them a lot bigger than others. But I think the moment in time that we’re at, in construction and technology, is this perfect convergence of the ability to structure highly unstructured and unfiltered data and turn that into organized knowledge. And that takes everything from reality capture, to meeting recorders, to abstract 2D drawings, to thousands of pages of contracts and other documents. If you didn’t have all of this progress coming from AI, and physical tools, and edge devices, and compute — the capabilities to handle just how multivariable and complex a physical industry like construction is — it’s never been possible. So I think this is the most exciting time there has ever been in construction technology, aside from the advent of heavy machinery. 

Zaid Kahn: 

Yeah, Mack, I think your perspective is really good — I grappled with this when Salil and I were thinking about whether we should go into this industry or not. Because on one hand we debated: is this industry really backward? Like, my god, they’re still on spreadsheets, maybe they don’t put enough money into it, maybe they just don’t like software — tons of different scenarios we went through, and also when we talked to different people. On one hand you have the Suffolks of the world, who over time have been open to accepting technology, even in the SaaS movement. And then you have some people you talk to who just completely say, “I don’t want to do that, I’m just going to use humans and spreadsheets.” And to agree with Mack here — that’s not because they weren’t interested, it’s because it’s so complex that they didn’t want to break that mold, or didn’t understand the value that it would bring in. Even when you go back and sometimes ask, “Hey, has this software that you bought five years ago helped you?” — sometimes you’d get the answer, “I don’t know, but I’ve been doing it,” because it was a task you wanted to automate. Automation, in my opinion, is sometimes just an incremental lift — it’s great, maybe it improves your productivity by five percent or something, and maybe at the time you felt that was the best you could do. But I think AI changes the game — it’s such a big technological shift that it’s ripe for an industry with hard problems. And that’s where construction is a really hard problem — that’s what’s got Salil and me excited about it, because both of us as founders gravitate towards hard problems. He gravitated, years ago, towards building self-driving cars. I gravitated towards building supercomputers, and never built a chip in my life. So I like that kind of stuff, and this was a really hard problem — construction is a really hard problem, it’s very complex. And AI is just ripe at the moment to come in and catapult the industry to newer heights, because it can actually rethink it — it can look at outcomes differently, it can look at, okay, how do we do things differently, because now we have this piece of software that can look at things from a different perspective. All these things we were talking about earlier — I think AI makes the possibilities even more endless. And that’s where I think every construction company should be excited about it. It’s not about, yes, there’s this, do I trust it, and all that — but open up a little bit and say, how do I change my business to take AI and catapult it. 

Steven Rossi – Zalmons: 

Yeah, that was great — you’re both great answers, thank you. To wrap us up then, when you look into the future of Neuron Factory and your partnership together, how do you see it evolving over — let’s say the next year, maybe even three to five years, I know that’s looking way out at this point — but how do you see it evolving, and what do you think other tech companies and GCs can learn from your story? 

Mack Rush: 

For one thing, I think there’s a ton of learning that we have ahead of ourselves, and the thing I can look forward to the most is just twelve months from now, looking back and reminiscing on all of the things that we’ve been able to learn about in terms of the way that information should be flowing through our organization. I think one of Zaid’s last points, from the previous question, was around how technology actually changes your business. And I think part of the reason why a lot of technologies never really influenced the way construction gets built is because they were more or less digitization of paper processes and the copying of outputs that humans have to produce. They didn’t change the core structure of the business or the way information flows in an organization. And I think we have the potential for that right now. So I alluded to this earlier, around how we are suddenly questioning the way these functions need to be staffed, who has access to what and when, how quickly something should be produced, and whether someone should follow through on a project from pre-construction all the way through the end — who in our company gets access to this information, and why. There are all sorts of reasons why our internal hierarchy is organized the way that it is, all sorts of reasons why our internal permissioning of access to certain documents is structured the way it is — because there’s material risk that has developed over time in the way we communicate these things within our own company, and we’ve needed to create processes around that. But this has really afforded us this opportunity to look at what this company should look like to be the most efficient builder possible. Something I’ll credit John Fish and a lot of the Suffolk team with really embodying is this idea of a seamless platform — rather than having different siloed functions, we are really trying to embody an organization that can provide the right information to the right person throughout this entire construction lifecycle, and every bit of that be moving forward towards completing a building faster and cheaper for the client. So I think we’ve got a lot of learning to do, but I am very much looking forward to seeing what our hypotheses will look like for how we can change our organization, and what our expectations are for how quickly we can get out of pre-construction and get to building. I think this next volume of information that we’re going to be passing into Neuron Factory is going to be really, really exciting. 

Zaid Kahn: 

I think for us, we want to build great products that bring impact — that’s continuously what we want to do in this next phase of the partnership. We continuously want to see how we can build our products to make sure that Suffolk’s evolving business lines up — “Hey, we built this building with the best mindset and the best outcome that we wanted” — and continuously map our product roadmap to that, because at the end of the day, that feedback loop helps us build great products. As founders, we want to build great products that bring impact to the business. It’s not about incrementally making people more productive — I used to live in a world of “I want to make people’s lives more productive,” but productivity also means good outcomes, not just, “Thanks to AI, I can read this thing in five seconds instead of however long it used to take.” That’s not actually impactful — the outcome is more impactful. I think that’s kind of this new project we’re excited about, which I know we can’t talk much about in detail — but when it happens, I want to make sure we map all of that back to what we’re building, because that’s going to help us get to the next thing that we’re going to partner with Suffolk on. So that’s what I’m really excited about. 

Mack Rush: 

Yeah, it’s going beyond the incremental. The last thing I’d add to that — a lot of people ask us about all the tools we’ve tried. When I first joined this group, going on four years ago, we had a year where we ran about 60 pilots with different point solutions, and every single one of those had discrete ROI calculations, and it was all kind of incremental improvements. Some of them did fantastic — it wasn’t all incremental — but there came a point where we were looking at a lot of these studies and trying to think about our perspective on ROI. And if you have to really try hard to understand if there’s ROI in a solution, or what the incremental improvement of adopting these things is, it’s not as innovative and not as influential as you might think it is. And I think the thing that, for everybody internally at Suffolk who has been close to the work we’ve been doing with Neuron Factory — there is just this collective intuition that this is a needle-mover. And I’m sure we could pencil out the right reduction of risk and figure out how to quantify this, etc. But I think, for a lot of those builders out there trying to understand how to really quantify the value of these solutions, I think back to Andy Grove and the idea of a strategic inflection point. When Intel decided to pivot their business, it was a gut feel. And I’m sure we could quantify this — don’t get me wrong, for anybody listening, we’re moving fast, and there’s wood to chop, and it’s exciting. Huge credit to you guys for helping us see what this pre-construction department of the future could look like, and really what the builder of the future looks like. So yeah, one last cherry on top from my side of things. 

Zaid Kahn: 

Awesome, no, agreed. 

Steven Rossi – Zalmons: 

Yeah, that’s great. Thank you so much for your time and your perspectives. It’s a great story, and it seems like you’re building something great here — excited to see where it goes from here. And I’m sure we can talk about this all day, it seems like. So looking forward to hearing more in the future. 

Zaid Kahn: 

Yeah, thank you. We’d love to at some point come back and talk about how that project was successful, and here’s what the outcome is. So more to come. 

Mack Rush: 

Thank you, Steven. Great talking with you. 

Steven Rossi – Zalmons: 

Awesome. 

Mack Rush: 

Great, great talking with you. 

Zaid Kahn: 

Thank you.