Takeoff is one of the most expensive single parts of a construction project, consuming hours and hours of staff time. Like many document-based processes, it is a perfect use of software, and even more so of artificial intelligence. Building on decades of experience at Coastal Construction, Patrick Murphy and his team have applied AI to automate the takeoff process, providing a powerful tool that augments the team and dramatically improves productivity.
Check out Togal here: https://www.togal.ai/
Reach Patrick here: pem@togal.ai
Togal
Hugh Seaton: Welcome to constructed futures. I'm Hugh Seton. Today I'm here with Patrick Murphy, CEO of toggle.ai. Patrick, welcome to the podcast.
Patrick Murphy: Thanks for having me Hugh.
Hugh Seaton: So let's start where we always do with what does Togal.ai do?
Patrick Murphy: Great question. So first of all, let me tell you what, what Togal.ai means, because people always asking me what the heck, I don't even how to pronounce that word.
So my name's Patrick Erin Murphy, I'm Irish, and in construction. And this is a company that automates the takeoff process. So when we were thinking about what we wanted to name it, we turned to the Irish language of Gaelic and discovered that the word for "builder" is tógálaí, T O G A L AI. So we said, holy cow, this is perfect, put a period at the end, "togal.ai".
So it means builder in Gaelic and at the core, what the company does is automate a mundane sort of boring process called takeoffs, right? That square foot analysis, whether that's an area analysis, whether that's a linear takeoff or object counts, we are reducing what the user for their, you know, sort of whole career has taken days or weeks to complete.
And now doing that in a matter of seconds, with oftentimes greater accuracy. So, it's one of those solutions that when most people see it, they say like, "holy cow," it's like sliced bread to them. It's so obvious it's right in front of them. And hopefully a solution that helps add a lot of value to the industry.
Hugh Seaton: Thank you for that as a good intro for what Togal.ai does, but there's a lot of history behind how you got to where you are. Do you want to talk a little bit about how you arrived at this? Cause it wasn't, it wasn't a couple of people sitting, over lunch saying, you know, it'd be really nice is if we could do take off. You guys really kind of put in the hours to get here.
Patrick Murphy: Yeah. Like anything, there's definitely a story Hugh. So I was born and raised in a family business called Coastal Construction. And Coastal's now the largest general contractor in Florida and spend, you know, quite a bit on pre-construction estimating, like most contractors, the pre-construction department is the biggest piece of our overhead in the company.
And so, while I grew up in the family business of construction, I've worked in, in sort of other fields as well. And one of those was, public service and I was in Congress in DC for four years and, being the youngest member of Congress at the time, I was spending a lot of time focusing on the future of work, artificial intelligence, the sharing economy. You know, really trying to think about how we should be training the students of today for tomorrow's jobs and spent a lot of time thinking about legislation and how we should be improving our country. So that was 2012 through 2016. And when I rejoined the family business in 2017, I had been sort of out of the business for about 10 years or so, and was sort of shocked to see how little had changed.
And it wasn't just, our company was the industry as a whole. There was very little innovation despite everything happening, in the world with technology, construction has remained fairly stagnant and we've seen, you know, a study from McKinsey and many other reports showing that construction is actually evolving at a slower pace than, than basically every other industry.
So I've got these two sort of different worlds kind of going past each other. One, I'm trying to advance, machine learning and the other one is stagnant and was really trying to think about how we could bring that future into construction. How we could start to bring some innovation into specifically our company.
And I'm a CPA by training. So I spend a lot of time with our CFO and I was going through our financials with him. And that's when, you know, I was sort of reminded that this pre-construction department was the biggest piece of our overhead. And that's when sort of the light bulb started to go off. That holy cow, in pre-con about 50% of the hours that we spend with our, our team is just doing takeoffs, right? Constantly getting plans, getting revisions, constantly asked by owners and developers for an updated number. Hey, I'm thinking about buying this land. What's this new hotel going to cost. Give me a quick number.
Everything starts with a takeoff. So that's when we said, wow, this is something that's got a set of training data, and it's repeated two perfect things for machine learning. So that's when we had the idea, I was lucky enough to meet some, some people much smarter than me in the technology world. They gave us the confidence that we could actually do this.
And we've now spent the last couple of years developing the training data and algorithms to automate this process.
Hugh Seaton: That's a really cool way of looking at that story. And that is you brought to your, your decade or so outside of the industry, a pretty baked in understanding of it that allows you over time to kind of let ideas percolate and you're bringing together what's going on in other parts of the economy that may be further along in certain areas back to construction.
And I love also the fact that, you know, often products come about because somebody started with the technology and then said, okay, let's see, where can we apply that. You were the other way around where you said, let's look at our business and try to solve what is right now, the most painful, certainly economically painful part of, of what's going on.
I've literally never heard somebody say that, I looked at the whole business and said, where do we start solving problems? And then said, okay, well it turns out that's a good one for machine learning. So let's get after it. Really interesting approach and obviously one that, that I am sure a lot of contractors do the same thing.
Don't get me wrong, but we're talking about Togal, not a contractor. So it's born of felt pain in the field, in practice.
Patrick Murphy: Exactly. And, you know, when I looked back several years ago, at those early days, It was really just doing, doing just now looking at other technologies, I've started to read about, drones and other technologies in the field.
And was kind of looking for where's that low hanging fruit, where's that obvious opportunity to solve a problem. And it just kind of made sense to start with the big cost drivers and what could affect the bottom line. And if you can shave off, heck 10, 20% of someone's cost, of their biggest number, that's a pretty big amount.
And you know, something that applies to businesses of all sizes, you could be a sole proprietor or Turner. It could be the biggest contractor in the world. Everyone's got to do a takeoff, right? Everyone's got to know how big the job is, they're about to embark on.
So that's kind of the next transition, I didn't really mentioned a second ago to you, but you know, in the first several months, we were thinking we'd keep this technology to ourself and heck this will make us more competitive. We'll get bids out quicker. Our team can now focus on the higher value tasks, like value engineering and scoping, and maybe finding some new trade partners to help us bid on the work. And you know, really impressing owners with all the work we're able to do in a two or three week period.
And that as we started thinking about it, we said, gosh, you know, why would we keep this to ourselves? Sure. We might win another couple of jobs a year, but you know, that's often the capacity is often driven by our field teams and, and their ability and how many team members we have in the field to build these jobs.
So. And that's when we said, look, let's spin this thing out. It's applicable to far too many people to keep to ourself and have it be a standalone company. And that's when we actually spun it out of Coastal, raised some money from third parties, industry type people and, developed a board and all that.
And now finally in market.
Hugh Seaton: That's great. And are you finding that responding to pressures that are outside of just one company, you know, makes you better, best of breed. Like it allows you to hear from people that are experiencing things that are maybe different kinds of work or just, they happen to have different issues or have different requests.
But are you finding that, that one of the benefits of opening this outside of Coastal is that, that you're hearing from more voices and it's really kind of enforcing a discipline that internally might have been different.
Patrick Murphy: It's such an important point and something we really try to go beyond our comfort zone.
And whether that means talking to a competitor or talking to a GC or trade partner that does a completely different type of work. Maybe an industrial or power plants or, you know, something, water facilities; and, and really understand how, how they do their takeoff, understanding their flow.
Because the last thing we wanted to do is spin it out and then go build a product that only worked for a handful of companies. Or even worse just ours. So what we've tried to do is from the beginning, get a diverse pool of investors to help in that sort of counseling. But then also in the early days of the beta users get a diverse group of folks that do different types of work in geographically are spread out, to sort of influence the design and mechanics, the flow of the software. And, you know, look, it's never a perfect for everybody, but, you know, hopefully we've built something that the vast majority of contractors and trade partners can work with.
Hugh Seaton: Yeah. And I think, again, the story of all the depth and experience and understanding of complexity that working as a GC and growing up through a GC brings plus the exposure to the outside. And that, that story continues right. Is that as you, as you increasingly onboard outside customers and have those sales meetings, you bring the best of both worlds together. Cause I think one of the areas that construction technology has historically suffered from is people not marrying those two, where you've got the technologists who think in abstract, and don't quite realize that their knock on effects to everything they're proposing.
And the contrary being true, where the contractor doesn't really understand the technology and what's possible. So they're just automating workflows. They already understand which is a great first step, but isn't really where you want to get to.
Patrick Murphy: Right. Yeah. I mean, you nailed it, hugh. Sometimes, I mean, sure you and I both, seeing technologies out there that are cool, but they just don't solve the problem in our industry. It's just not a big use case. And contrarily a problem that we're solving automating takeoffs. I mean, if you polled all of Silicon valley, I don't know how many people would even know this problem exists.
Right? These are some of the smartest people in the world, but you don't know what you don't know. And unless you've worked in the industry, it's hard to probably understand or just fathom some of the ways we do things. And they're not always the most efficient, sometimes it's backwards. People still ask a lot of questions about just the industry and the way we operate with yellow pads and spreadsheets. And email's still a big advance as is Excel, I think for a lot of construction folks. So we've got a long way to go and I think it's that hybrid model you're describing where we have the construction folks that understand the industry, combined with the technologists who know what's possible because I mean that, that is an equally hard challenge.
And now that you can even solve XYZ problems such as automating takeoffs.
Hugh Seaton: And how did you guys organize in the beginning to be able to collect this data? I'm conscious of the fact that I'm talking to the CEO, not the CTO. So stop me if I get too, too deep into it. But one of the big issues often in, in contractors, and honestly in anywhere in software is gathering enough data to be able to train a model that is sophisticated. It's not so hard to teach it, to recognize a dog versus a cat, but to get into dog breeds or to get into more categories. And you know what I mean? Like what you've, what you're doing is, it takes a bit, it takes a bit of sophistication for Togal to do what it does.
Talk to me a little bit about the data organization process that, that you, you oversaw.
Patrick Murphy: Yeah, Hugh, uh, I guess on the heels of what we were just talking about, it, it almost would have been easier now that I've learned more about this to develop an AI company for a self-driving car, because there's actually more labeled content on the internet for stop signs and yield signs and roundabouts and cars. Right. And there's a lot of publicly available information out there for this.
But there is next to nothing for it, labeled information for the construction industry. So that was a challenge for us. When we started bringing on the tech team, they said, oh, let's just go get the label data. We'll build the algorithms and this'll be easy.
It turns out there is none. And in order to have accurate algorithms, You need to have really consistent data that is trained in a consistent manner. So we were thinking, well, gosh, well you know, Coastal's got all this great data and all these labelled plans for however many years of takeoffs we've been doing.
But as we started diving into it over, you know, 15 year period of different. You know, sort of employees, team members doing takeoffs, there's some differences on how it was labeled, right. And how objects were identified. You know, is it a bathroom, is it a toilet, you know, like you can call the same room four different things.
So then we realized we're really starting from scratch. So we had to go to the drawing board and just find raw data, just find plans. Yeah. Some of them were schematics sort of DDs, really sort of early on sets of plans. Um, and then some were very developed. We wanted all different types from industrial buildings, warehouses, hotels, multifamily, and get a sort of wide array and then build a team of architects and engineers to actually do the labeling and do it in a consistent manner where we basically have a very thorough spreadsheet and class on how our labelers work, and have a team of people that check their work. Because if the labeling is wrong, your accuracy is never going to be what's needed for the industry. So we've spent a ton of time and continue to, developing that training team, the labeling team, checking all those plans and then running them through our algorithms, constantly trying to improve those.
So, that in hindsight was one of the biggest challenges and, and perhaps will be one of the biggest sort of attributes or values that that Togal has to the industry is that, that data rich set of information that we have.
Hugh Seaton: And are you guys set up? It sounds like you are, but I want to just make sure that data that as projects get executed and as more and more data goes through the system, it continues to build on and refine the models.
Patrick Murphy: Yes, exactly. And as we start to grow as a company can use it grow and see more diverse plans and more objects, et cetera. We will continue retraining our algorithms and it's a fine line. I won't get too deep into it and bore your listeners here Hugh, but you know, you don't, in the early days, I think you probably want to be careful with your users.
And if they're making corrections... is that. Is that a real correction that we want to be sort of company-wide or is that just because that's the way they want it, right? Or are they even correct in that quote unquote correction or whatever they're doing on the tool? So, it's all about sort of numbers, right?
And one bad plan won't necessarily ruin it for everything, but you start to get certain trends and, and that could affect the accuracy. So, we are continuing to make sure that our own humans, that we've trained our labelers are involved, in that training to make sure our accuracy doesn't ever go below 97, 98% in fact goes closer to 99, a hundred percent.
Hugh Seaton: Actually, there's a couple of really good points you just bring up. The first one being that any AI is, is giving you a probability. It's not giving you a, a definitive, this is the truth, but often that's a very high probability. The reality is people do this too, we just don't call it out, right? Like someone's applying judgment, there's a likelihood they're right. But what's interesting is you, you set a cutoff, a key part of building any AI system, which you, your team has obviously made this call is at what point are we accurate enough to just make the call? Um, and I think that's one of the areas that, that often people don't quite get, is that there's a business decision about how accurate is accurate enough to say that this is true. And I think that's something you learn a little bit and that's, again, one of the benefits of, of coming from a, uh, from a contractor is you're able to make that judgment in a way that maybe a tech person wouldn't quite know how to do, without some deep research.
Patrick Murphy: Yeah. You know, w what we did on that note, in order to make that decision is... and thankfully have coastal there, to go through and we're able to go through some historical plans and really look at the quote unquote, accuracy of those takeoffs. And interesting. Yeah, so we kind of looked at that and, and determined that that 97% mark was a pretty good number.
That if you can be greater than 97%, you're probably as good, if not better than most human takeoffs that were done. And, you know, for various reasons people were distracted or bored or, you know, whatever,
Hugh Seaton: whatever, just transcription error alone is going to get you a couple percent.
Patrick Murphy: Sure couple of points, exactly. So that's when we kind of looked at that, said okay, well that's, seems to be a good number.
And what we had also did is made the determination that we're not going to show or label a polygon, or an object, if it doesn't meet that criteria, we would rather, it just sort of let, let the user then label. Just say it's a bedroom, right in this case, that if the computer isn't, the algorithm is not 99% sure it's a bedroom, just keep it empty. Right?
Draw the walls, make the polygon, let the user go fill that in. Now if it's a stairwell and we're a hundred percent sure, then we label it, right. We, we do that work and a lot of objects, a lot of spaces, bathrooms and kitchens and laundry rooms and hallways and, and items have very distinguishing factors that we can automatically, label and identify with a great degree of accuracy.
So that's kind of the way we've looked at it. I don't want people losing faith. Well, wait a minute. That's not a bedroom, that's a living room. And then they get upset and they start to question everything. So I think, as we're in these early days of a learning curve for the industry, where people, our estimator specifically are used to doing this, labeling, doing these takeoffs manually taking days and days to do it. They're now clicking a button and they're making a minute or two of maybe modifications or tailoring it to the way they want. And getting used to that is quite different.
Hugh Seaton: That's a really great point you just brought up and that is the tolerance of folks in the industry sometimes with something that isn't a complete replacement for a person. And I think it's, it's incumbent on people in the technology side to say, we're augmenting you. And we're allowing you to do in in 20 minutes, what used to take half a day?
But it doesn't mean it's a person. It's just a tool that just like any other tool on the job site, it's doing things under the control of a human and doing and allowing that human to do drastically more than they could on their own. But nevertheless, it isn't replacing a human. And I think that's important that, that it's not like you've delegated this to another human who could come back to you with it completely finished.
It's more like this is saying no it'll automate all the easy stuff, but there's going to be some things that require judgment that there's no quote unquote right answer for some times. It's more like someone's going to have to make a call and it's better that a human do that than an algorithm do that.
Patrick Murphy: And Hugh, we like to say that we're supercharging your, your human estimator, right? We're we're giving them a tool to kind of remove this "coloring." part of their job, you know, the young men and women that we're hiring to be estimators right now at coastal didn't sign up because they like drawing polygons and manually, right clicking and labeling objects for days or weeks. Right. That's not what they dreamt of. So take that away from them and now allow them to be that engineer, that creative thinker, that estimator problem solver, that's looking for a nice creative way to save the owner or the end-user some money. You know, a better way to build a mousetrap, maybe a better material, finding some errors and some scoping, right? Doing all the leveling, all those important things that have to be done by a human. And that as you know, is what often can make or break a decision to win a job, or even that go, no, go. So, that's what we like to say is it's supercharging the humans by no means are we displacing anybody, but we are allowing them to do a lot more with a lot less.
And if you're a small business and you're trying to grow, or you're in a busy environment, like we're all in right now, there's just work everywhere. It's hard to keep up. So, how do you make sure you're going after the right jobs, right. And not making a, a mistake like bidding the wrong job that you're never going to be competitive at.
Hugh Seaton: Yeah, that's great. And it speaks to the quality issue, right? Is that that if people are spending less time on repetitive tasks, they can spend more time thinking a little harder about a value engineering problem, like, you know, particular issue or whatever. You're just going to get more quality and creativity out of people that have more time now to really sit back and think.
Patrick Murphy: Right.
Yup. That's, that's kind of the value proposition that I would say are our biggest, best users right now of the software have identified and said, holy cow, this is allowing me to do so much more. I can get bids out the door quicker. I can get better bids out the door. Uh, this is going to make me look better to my bosses.
You bet on signing up for Togal, this is a no brainer. And that the folks that identify that seem to be the best users, we just got to keep finding more people that see that.
Hugh Seaton: That's right. Well, it makes sense. And I, are you hearing anything about retention? One of the things that products like yours sometimes help with is what you mentioned earlier as you've gone and you've recruited somebody, you've trained them in the way of the individual contractor.
You spent some money and invested in them and then you have a retention problem if what they're doing is drudgery and they hate it. And you know, this is obviously a bit of a generational thing too, but retention is a problem across the economy and giving people better tools that make them feel like they get to grow and solve better problems is a way, not just in construction, but generally speaking, you're hearing companies really start to think about the tool sets they give people, so they feel like their days are more fulfilling and they enjoy their job.
Patrick Murphy: I don't think I can say it better than you just said it, hugh. I think there is a generational component to this, uh, but there is just a, sort of a human value component, a psychological piece to it as well, where individuals want to do higher value tasks and want to really see that their work really matters, and that they're really changing the trajectory of a company and adding that value.
And the traditional way of doing takeoffs is it just that it's traditional, it's not really value add it's you do it cause you have to, but you know, you're certainly... it's not the best use of a human's time in today's age.
And when I think back when I first started working in our construction company, Coastal, my grandfather was the head of our pre-construction estimating department. So I learned with him and I had a roller and a ruler, right. And a yellow pad, and I did all the manual takeoffs and the highlighting and all that.
And then the digitizer came out and that was a big, you know, sort of change. And then the sort of on screens of the world, these manual softwares, that are basically that same process of rollers and ruler, just with a mouse. And we haven't seen much change in 20 years. Right. I mean, since then, then that's really been about it.
And I actually did a couple of panels recently and talked to some estimators from around the country. And one of them was asked what the biggest innovation in pre-con in the last 20 years was, and they said Microsoft Excel, right. Where that was the big innovation. So this sector is certainly poised for change.
Hugh Seaton: That's great. I wanted to shift gears now and talk about how somebody starts with Togal. Like what's the process of, like, let's say someone says, you know, that sounds like a good idea. What do they do?.
Patrick Murphy: So first, go to the website togal.ai. And you can request a demo there, uh, or heck just email me pem@togal.ai. So my initials pem@togal.ai, and we'll get a demo on the books. We will, you know, show that interested party, the software a quick sort of 30 minute demo, and get going. Basically most folks see it, they like it. They say, Hey, let me, upload a couple of plans on my own and kind of mess around with it.
And we get them started with a trial period, get them signed up and we're off to the races. We're currently charging $250 per user per month. And you know, obviously willing to negotiate with those enterprise clients and such. And it's as easy as that, once you you've got it, it's in the cloud, you're not like downloading a software in your computer and like that you log in, you have your username, your password, upload your plans anytime of day and you can have multiple users seeing what's going on across the country or across the world for that matter. They upload it, they click Togal and boom they've got their takeoffs right there. They can export it to, to their Excel spreadsheet, they can map it to their estimating software, whatever they'd like to do. We have numerous APIs to connect with other softwares.
Hugh Seaton: And the way that sits in a workflow though, is that you know, at any time, I mean, forgive me, I have not myself worked in a pre-con team so I, I want to be careful about how I described this, but, you know, you've got the team and when they need to do the first takeoff, and you mentioned before, a lot of the value in this, it sounds like is when things change and you can pretty quickly redo it. So there's this online portal that they can go to and just upload what they need and out comes the takeoff is that right?
Patrick Murphy: Yeah. So they will, they upload their plans and the computer within seconds will automatically draw all the polygons. We'll identify each of the rooms, the shafts, the hallways, elevators, whatever it might be, draws all the lines for the walls, and then does object counts and then it will automatically classify and label the room types.
Again, back to what we talked about before that it's confident in, the wall types it's confident in, in the objects and the user can then if they want to move a wall around, if they want to relabel something, they can do that. They can a right click and we have some, some shortcut tools, uh, for the user.
And they can get it into a sort of a spreadsheet format that they like. And look at it right there on Togal, or they can export it to, to what they're comfortable using. If a month later, six months later, a year later the owner sends them a new set of plans, right. And they've got to get a new budget, then they can just compare, right on Togal and compare the two drawings and see what changed and a new feature that we're working on is really exciting, Hugh.
Not only do we show the user where the, maybe the lines or the polygons are different, but we're going to quantify for them, what exactly changed. So, we're not just telling you the master room, there's a line move. We're telling you it's now 32 square feet bigger. And instead of a tub, it's now a shower and we're going to quantify all those changes.
So as a job is in pre-construction and then as it moves into the construction phase and you're getting updated construction docs, the contractor subcontractor is going to have a very powerful tool to make sure they're tracking all those changes and getting the necessary change orders filed to their owner. And in a process that's fully transparent and everyone can see.
Hugh Seaton: This is great. You just hinted a little bit about where you're taking it in the short range, but long-term, how do you see this continuing to grow and where do you see the pre-construction or the part of pre-construction that you're, you're really leading in, where do you see that going?
Patrick Murphy: So right now we've focused on automating the takeoffs for architectural floor plans. That's been sort of the core of what an estimator does and that's the core of what we do. But as we look forward, we want to take that same thought process and apply it to elevations, to structural plans to MEPs. And continue, using and developing those same algorithms to do a takeoff for an electrician or plumber, right, on a very detailed set of plans.
So again, that just say a HVAC sub is not focused on how much duct work there is. The computer told them that. Now they can figure, is there a better way to be running those ducts? Is there, is there a better system to use? Here's a better way to install this and focus on those kinds of tasks that the human should be in involved in and not just measuring the ducts, right? And we want to keep going deeper into the AI for all of the trades.
Hugh Seaton: Very exciting. Well, Patrick, this has been great. I actually learned something about estimating and pre-con so I appreciate that. Thank you for being on the podcast.
Patrick Murphy: Well, thank you Hugh, I appreciate it..