AI Assistants In Construction: Enhancing Efficiency And Creativity with Marty Cornish I Episode 279

How can you reduce the amount of time it takes to do a takeoff from two days to 20 minutes? In today’s discussion, let’s open our windows towards the future where AI Assistants can make most impossible things to do become possible to do. In this episode, Marty Cornish, the founder and CEO of Workpack, shares the value of AI Assistants in the construction industry in enhancing efficiency and creativity. He also discusses where AI is heading to the future, where it is today, and the misconceptions most people have about Artificial Intelligence. Marty’s expertise in Artificial Intelligence proves he could provide you with insights into successfully implementing AI in your business. Join in!

Get the One Book Every Construction Leader Must Read:

Construction Genius: Effective, Hands-On, Practical, Simple, No-BS Leadership, Strategy, Sales, and Marketing Advice for Construction Companies:

Need executive coaching? Book a short call with Eric to discuss how he can help:

Watch the episode here

Listen to the podcast here

AI Assistants In Construction: Enhancing Efficiency And Creativity

How can you reduce the amount of time it takes to do a takeoff from 2 days to 20 minutes? That sounds interesting. My guest Marty Cornish and I discussed that on Construction Genius. Marty is the CEO of Workpack, which is a startup that brings automated quantity software to construction cost estimators.

Our discussion is on using AI as a copilot to enhance workflows and productivity. Their main focus is on takeoffs, but we have a broad discussion about the impact of AI on the construction industry, where it is now, where it’s going in the future, some of the misconceptions about AI, as well as some predictions for how AI will be used in the future.

Another thing that we talked about that I think you’ll find very useful is how to get AI implemented successfully in your business by combining the commitment of leadership with the enthusiasm of the younger generation to get the middle of your company involved in using this technology to improve efficiencies. Enjoy my conversation with Marty. I appreciate you tuning in to Construction Genius.

Marty, welcome to Construction Genius.

It’s great to be here.

I know there are lots of college kids and high school kids at the moment using ChatGPT to create very boring papers. We’ve been inundated over the last 18 months or so with the wave of AI products and AI chatter. As an expert in AI, I wanted to bring you on the show to talk about how you view AI impacting the construction industry in the next 12 to 36 months.

Impact Of AI In The Construction Industry

No question, it’s already here. It’s not even 12 months out. AI is changing workflows across industries. I think that not in construction, across industries, we’re going to have a variety of different co-pilots or AI assistants or whatever you want to call it that we’re going to delegate work to. We’re going to have to provide a setup or a prompt and it returns a bunch of work.

As you said, it does turn into a boring term paper. It’s your job as a college student to make it colorful, to add some personality, and to organize it in a way that reflects you, then you can turn it in. As an estimator, you have to turn the initial data that gets returned by your AI co-pilot and turn it into something that makes sense as an estimate to fit your company’s cost structure and specific needs and coordinate with the rest of the project team and anybody whom you’re working on the job with. The workflow changes from the tedious clicking and tracing on takeoffs with our product to let’s coordinate, communicate, and understand projects more effectively and smartly organize them.

Let’s dive into that. In construction, broadly speaking, you got to bid the work, you got to plan the work, you got to build the work. and you got to get paid. Right off the bat, you’re talking about AI as it relates to estimating. Give us a little more detail because an estimator or head of estimating might be listening to this and thinking, “AI, do I need to fire all my guys and produce estimates without any physical human beings?” Give us a little bit more on this idea of co-pilot when it comes to AI and how someone like an estimator will able to use it effectively.

AI Co-Pilot

Totally, with our product, Workpack, we build up an AI quantity takeoff software. When I have been talking to people about the product over the last four and a half years as we’ve been building it, very early people instantly wanted something that would help them do takeoffs faster. That’s a very attractive value proposition but then when you get into the conversation of what are my fears around this? It’s like, “I don’t want to automate myself out of a job.”

What we’ve found, and this to some extent surprised me, is that more than an autopilot, it’s a co-pilot. You have to give it context. It doesn’t have any context. It returns whatever it’s been trained to do. It returns the data based on whatever it’s been trained to do. It produces and that’s it. It’s got no incentives. It’s got no context. It’s got no experience.

You have to provide that initial, and then you’ve got to layer your experience on top of it in order to make it effective. It’s like this AI plus human estimator, harmony, synergy, or whatever. It’s a tool for us to produce a lot more work and we’re going to be a lot more effective by leveraging AI in our workflows in pre-construction, certainly starting now, but across industries by leveraging AI and the best AI tools that we can get our hands on.

Let’s go to the main issue there of the prompt, because I think one of the key challenges that people have in using AI effectively is the quality of their prompts. As you work with your construction clients, how do you help them in terms of creating the prompts that are going to give them the information that they need?

Most of our customers are leading commercial construction companies across the US. The best most effective users tend to be enthusiastic about new technology and that’s obvious, but they’re also the younger generation. Interns and junior estimators don’t have a lot of muscle memory around some of the legacy tools. Also, they have grown up in the world of Google search and have used some of these AI tools and have played around with them.

The prompt workflow is a delegation workflow. It’s almost like an intern doing work for you, but it doesn’t have feelings. It’s software returning work. You have to say, “I want you to do a takeoff on this.” or “I want you to review this part of the contract,” or “Review these images for this.” Whatever context you can provide and the more context that you can give the AI to produce meaningful specific data for your specific needs, the better.

I can see the application with estimating very much. How do you see AI affecting other aspects of the construction workflow process from planning the work to building it to even getting paid as it begins to get developed into the industry?

There are certainly applications on sites for reviewing site images, and some of those exist today. There are certainly good applications across contract review, creating scopes, writing initial drafts of contracts, and reviewing specifications. What AI is best at is the tedious and monotonous like data review and data creation tasks.

It doesn’t understand the relationships that you have with specific trade partners or the relationships that you have with clients and how they want to see their breakouts. It doesn’t have any of that context. We were playing around on our team, and our development, with Mid-Journey producing some sample architectural plans.

It doesn’t even have the context that you don’t put three toilets right next to each other in a residential house. It’ll produce some ridiculous results. No problem, you delete it or you edit it. If you’re the architect working on that project, you delete two of the toilets in the bathroom, but you have the general structure. It’s going to work and be helpful across a variety of applications. It’s a matter of who’s taking it on, and which software companies are building or have built solutions that provide value for that specific need.

It’s interesting, because as you’re saying that, I can hear some of my audience thinking, “I’m going to wait. The technology is new and it’s exciting. I see some of the applications, but it’s not that sophisticated yet. I’m going to wait for 1, 2, or 3 years before I start using it.” What’s your counter-argument to that? It’s not time to wait. It’s time to at least put one foot in the pond and test out the water.

We have one customer who did a door takeoff that was going to take 2 days in 20 minutes using our product. It’s 30X time savings. That’s the most efficiency gain you can get using a tool like Workpack but AI is here. Some of the solutions are mature. Chat GPT is almost mature. There are some chat solutions for customer service needs. There are some contract classifications or reviews of mature AI products.

There are a lot of demo-available AI products. Ours is somewhere in between that where it’s providing a lot of value for our customers right now, but it’s not mature yet. It’s still fairly new and there are a variety of tools but not as many in our category. Not as many are useful as our demo-robot. As you start to feel this wave, everybody’s feeling it. Everybody’s talking to me about it.

I talk to senior pre-construction leaders every single day. I had two conversations where they were talking about how the workflow of an estimator is changing and how everybody is going to be spending a lot less time doing takeoff and a lot more time communicating with the team, organizing data in the way that they want.

The importance of this data for construction. Another senior leader told me the other day that the number one killer in construction is bad data. Those types of things are known now. AI is the way to produce better data to streamline workflows to make sure that we’re being efficient with our day-to-day work in the pre-construction and overall construction day.

I know you can put in a prompt into chat GPT and you’ve got to check your information. Would you say that AI is more effective in dealing with pure data in terms of the accuracy of the output as opposed to something more subjective like writing an essay or even creating an image?

There are different AI tools. Our product, Workpack has seven different AI models inside it. We have different AIs that are trained to do different things for very specific purposes. There’s certainly the chat workflow, where you’re providing that context with a chat interface. In ours, everything is from the 2D PDF plans.

You’re providing context within the drawings of what you want to detect. You have to provide areas that you want to take off inside of and then you have to define these templates for symbols or labels. You’ve got to define the areas where tables exist and the tables that you want to detect those things. It does the work of classifying the objects. They are matching them up and providing additional richer data on that.

Can I ask you a question about that? You said you have seven different models within your product. Can you describe those models a little bit to give people insight into what the AI is doing behind the scenes and what it’s looking at?

There’s a page classification model that automatically indexes like, this is the page number and the sheet type and whatever. These are the walls, these are the doors, these are the windows in general. The wall engine is different than the door engine, and then there’s a label detection engine, and then there’s a label matching, and then there’s a table classification.

There’s a symbol detection that is different from the label. These things are trained to do a specific task and return a specific data result. Some of the classification happens automatically, but you can reclassify or edit or whatever. It’s like Cloud intelligence. We can call it to do a bunch of work for us. There’s nothing artificial about the work that gets created, but it’s not going to be perfect.

It’s not going to be done in the way that you’ve trained your estimating team to do it over the years. It doesn’t have years of experience at your company working for you. It doesn’t have all of that context and experience. It has whatever it’s been trained to do, which is return some data from plans based on some context given to it by the user.

As you’re using your product, or any AI product over time, are you participating in the training of the model in some way with the inputs that you’re giving it?

That’s an awesome question because you hear it a lot and it’s a very good sales pitch. It’s learning on its own kind of thing. It’s also one of the fear-mongering techniques that people throw out, “AI is learning all the time. It’s going to learn everything and then we’re going to be slaves to the robots.” The way we train the models is we train them in batches. We’re doing the work to label the data. It’s the workflow for training AI models. You have to do the takeoff in software programs. AI can learn and then you produce a data set that is labeled training data that models can learn from. You feed it through the model and then it returns takeoff data because it’s been trained to do takeoffs. It returns that.

There’s not a self-reinforcing mechanism. It’s more like Workpack trains these AI models in batches. We’ve got great models that you can leverage. Most AI tools work that way. They’re not relying on users to give thumbs up or thumbs down every single image because you have to rely on having good clean data for AI to be correct.

If people are saying, “This measurement line over here is a wall.” Somebody is drawing a bunch of walls that are railings or whatever they are. The AI model, if it were learning from what the users are saying, “Every time I see railings, I’m going to say it’s a wall.” That wouldn’t be effective. We feed it with clean and properly labeled data, and it returns good results. What I see across the best AI products that I see industry-wide is that they’re training it or they’re leveraging something that’s been trained very well. We train our own models for the most part, but they might be leveraging some Microsoft or similar Google tools.

What have you learned from your clients that surprised you and has helped you to improve the models that you use?

Probably the biggest thing that’s been surprising is it’s not automated. It’s not autopilot, it’s co-pilot. It’s the workflow shift and how seamless the manual plus AI needed to be. That’s how we built it because of that. It needs to be like AI plus human estimator, harmony, and synergy workflow. It can create anything that I want as a rough draft but if I don’t want to use it, I can do whatever I want manually.

It’s limited. If I have an early-stage napkin sketch or something and I want some quick areas, I can just draw. If I have something a little bit more detailed like I want to classify all my wall types or all my door types, or I want to count a bunch of lighting, or I want to have this table to match up the room finishes, I can do those things by doing a quick setup and then it’ll return all that work.

You spoke earlier about how one of the best ways to implement AI is to use the enthusiasm of the younger generation. As you put yourself in the shoes of your typical construction CEO, and perhaps they’re interested in this and they buy into the logic and they can see a value proposition, with your clients, how have you seen them roll it out to the whole company in terms of marrying the enthusiasm of the younger generation with the leadership of the executive team to overcome the skepticism of all those folks in the middle?

We’ve seen some of that and candidly, we haven’t seen anybody roll it out across the entire organization. I don’t know the complete answer to that. I do know that what’s been most effective is having junior estimators in the product as their primary tool, and they’re never going back. The junior estimators are never going back to click and trace.

You mentioned the enthusiasm. We’ve seen two groups that are successful. Enthusiastic early adopters, regardless of age, it doesn’t matter. They could be people towards the tail end of their careers. We’ve had interns who are successful as well. It doesn’t matter the stage of the career and everywhere in between. What we’re starting to see is pulling from others in the team regardless of whether it’s the people on-site to be able to share or take off data in a far more collaborative way, or having multiple estimators working together, those sorts of things.

Our product enables a lot more collaboration than has been historically possible as well. It’s automation and collaboration, but the most effective teams and this is a leadership thing more broadly, have a blend of the rookies and the mid-career and the veterans. That is a sports analogy, but you have that on effective construction teams as well. We have one of the veterans mentoring one of the rookies through Workpack as a tool for the rookie to instantly become more productive than average.

Tweet: Our product enables more collaboration than has been historically possible as well. It’s automation and collaboration, but the most effective teams have a blend of the rookies, the mid-career, and the veterans.

They go from being less productive than average to being more productive than average, and then working together to show the veteran, “Here’s how I use this new AI tool to do my takeoff four times faster than anybody on the team has been able to do these things.” That’s what we’ve seen most effective for teams. That’s a change-anything mentality. You want the rookies driving a lot of the technology change with executive sponsorship.

I think it’s interesting how you’re specifically focused on the takeoff in terms of almost a non-threatening approach with the AI because I don’t know anyone in the world who relishes doing a takeoff.

Certainly, that’s why I started working on it. It’s so tedious.

Let me ask you this. It’s interesting because in talking to contractors, some folks will say, “I know takeoffs are boring, but I want my junior folks to do them because that will help them to get a real feel for the project and build some experience going forward.” What’s your answer to those people as you’re taking away the need to do a takeoff and get it done in twenty minutes instead of two days?


That’s how I learned the job and there’s no question that that is one way to learn the job. I have experience in all the legacy takeoff software, but as I use Workpack day to day, the most surprising thing is that I stay engaged the whole time I’m using it, versus going into this mowing the lawn, turning the brain off, and slug through it.

It’s just click, click, click. I’m engaged. I’m thinking about, “Did something get missed here? Did something get missed here?” The AI is identifying a bunch of stuff. It’s like, “Now I’m in edit or classification mode.” I’m always engaged. It’s not like I get tired from that. If anything, it’s the opposite. I get more excited and enthusiastic about my work.

That’s what junior estimators coming in want to do. That’s what the younger generation coming out of college, the people being hired into companies, that’s what they want to do. They want to be engaged. They want to be enthusiastic about their work. They want to use the newest most exciting tools. As senior leaders, you give them tools to be more productive and more engaged in their work.

You’re keeping talent around for a long time rather than having to train some new estimator and then having them leave in a year because they’re not excited about their work. Historically, we’ve been able to push off the tasks on the new guy. It’s not that way anymore. People want to be engaged in their work in a change. This is across industries. They’re getting their meaning out of their work and we’ve got to provide that.

Zooming out a little bit, as we think about the future of AI 2 or 3 years down the road, what is accurate in the current narrative that we’re going to see AI impacting 2 or 3 years from now?

I think we’re going to do a lot less tedious work. We’re going to be way more productive. You asked earlier about why I need to do this now. I think the companies that are adopting AI today are going to have huge competitive advantages over others over the next few years because the technology is moving so fast.

It’s not even the software speed. It’s software plus, so you get automation efficiencies but then you also get these work production efficiencies. People don’t even have words for this. AI is artificial intelligence. There’s nothing artificial. This is producing real work that people are using in their day-to-day lives. You can leverage tools to do your work a lot better. Companies throughout history that have leveraged the best technology have won, and countries and everybody loved the best technologies. That’s how you win. I think it’s going to develop more. We’re going to have more good AI tools. Much like software has taken over and eaten the world, we’re going to have AI not exactly eat the world, but we’re going to have AI workflows across the board.

What then is inaccurate? There’s always hype and then reality. There’s fear-mongering and reality. What is inaccurate that is not going to be the case 2 or 3 years from now as it comes to these things that we’re talking about?

We talked about the three toilets in a residential bathroom scenario. AI doesn’t have any incentive. It doesn’t have any context. It doesn’t have any experience. Anything beyond tedious monotonous classification of things and creation of rough draft-level things is not going to complete work for us. It’s going to do big heavy lifts for a long time. I don’t see that changing for 10 or 20 years at least if it ever does, but I do see people being able to complete a huge amount more work by leveraging the AI tools because they’ve got initial starting points on everything that they do.

What is going to surprise us about AI 2 or 3 years from now? What is it that we are going to say, “I didn’t think that was going to happen,” then here it is?

I’ve already been surprised. I think I’m on the other side of the surprise. Maybe I’ll share that. It is this co-pilot thing. We have to provide a context. It’s not just press a button. You have to delegate effectively. It’s not just a Google search. It’s a lot more than that. It provides adequate context on what you’re looking for, then the reclassification of data and the organization of data. It’s a change in workflow.

I thought, before building this product, that it’s going to be like we’re going to click a button and it’s going to do everything and then we’re done. We start to trust the AI system. It’s more of an assistant that we can leverage. I think we’re going to have these across a variety of different industries and use cases.

Tweet: We start to trust the AI system. It’s more of an assistant that we can leverage.

We’re going to chat with them by voice, by text, and interface with these systems in a way that is different than point-and-click and typing software. It’s not like we can digitize our workflows. We can have software return work for us and then we can edit or classify. It’s a lot more advanced than software has been where we can store things. We can use the computer to store our data and maybe share our data. We can use the computer now to create data for us. That’s new. It’s a completely new paradigm.

Do you see it’s a time when a construction management school within a university will be offering a specialization in let’s say prompt engineering for estimating or will they be doing classes on how to be a prompt engineer?

Everyone needs to be able to do it across the tools. I think most people will because people will be using so many different AI tools. I don’t think it’s limited to pre-construction. I think people will be using so many AI tools in their day-to-day lives. I think people will start to get used to using AI in the way that we’ve gotten used to using software generally.

There will be generally accepted principles around AI products and then AI use cases. I think the magic will be that some people will be better at creating good prompts. People with experience and skills in construction workflows, there’s a lot of travel knowledge that people have generated over decades. A lot of it is company-dependent, geography-dependent, client-dependent type of work. They know how to do this because they’ve done it.

I think we’re going to see a lot more of that than people focused on how to run the AI systems. I think many people are going to develop the muscle memory to do a good enough job leveraging the tools. Everybody’s going to be using them more than hiring specific VDC-level people to use the available specific tools. I think it’s going to be far more accessible. It already is. Our product, people need a couple of training sessions, and then they’re off and running. It’s not like they have to spend a bunch of time learning how to use it. It’s creating the takeoff or the starting point for you, then everybody knows how to do takeoff in construction.

It’s pretty interesting though. It’s cool as you interact with the various tools that if you hang out and play with it and put some inputs in, get some outputs, and you have at least a modicum of intelligence or experience to be able to evaluate it, you can get so much value from that.

You get a lot of value out of a lot of software tools. We’ve seen zero of our customers be successful doing a trial, and 100% of our customers be successful with three training sessions, which is what we do. The paradigm shift and the conceptual shift is a change in workflow. You’re not starting with a blank page. You’ve got to understand how to delegate, set up, and prompt effectively.

You probably don’t need a whole college-level course on it, but you do need a little bit of training on how to do that effectively. You need to learn how to classify or adjust what gets created. You need to know how to filter and export the data in the way you want. Some of that is our tool. Some of that is AI in general. Some of that is going to happen across different tools, and some of it is specific to ours.

It’s been interesting how It’s not something that people can play around with. It is something that you need some training on. Maybe that need will go away as more AI tools exist but the paradigm of prompting or delegating, creating this setup workflow is usually you’re starting with a blank page and creating right from the beginning.

Tell us a little bit more about your company. Specifically, what I’d like you to do is tell us who is your ideal client who should be contacting you to learn more about the work that you do.

Our product Workpack is a collaborative AI quantity takeoff software. Our customers are some of the largest companies in construction. It’s most useful for finishes. With Workpack, you can detect all the walls, doors, windows, and rooms. You can also detect labels, tables, symbols, and anywhere some patterns are different. Anybody doing a lot of commercial interiors, building hospitals, building senior living facilities, multifamily, and commercial office buildings, we’re going to see a lot of those going up.

In the private markets, we’re seeing their financing change, but that’s probably a whole another episode for somebody more experienced than me in that whole thing. Those are the best reasons to switch today, but you still can create anything you want manually in Workpack so that you can keep all the data together and you can collaborate with the rest of the team effectively. That helps keep everything in one centralized place.

How do people get in touch with you?

Email is good, [email protected] directly to me, or you can go to to learn more about the company.

I appreciate you joining me here. I look forward to getting you back in another eighteen months or so, seeing how the AI has progressed, and continuing the conversation about how it’s going to impact the industry.

We’ll see whose predictions were right at that point. It is changing so fast.

I tend to stay away from predictions when it comes to technology because of my own ignorance, but I appreciate you coming on. Thank you.

It was awesome to be here. Thanks so much.

Thank you for listening to my interview with Marty. If you like it, share it with other people and also go out to wherever you get your podcasts and give us a rating or a review. All the links in the show notes to help you to contact Marty are present. If you’d like to contact him, check out those links and make sure that you connect with him and let him know that you heard about his company on Construction Genius. Thanks again for listening. We’ll catch you on the next episode.


Important Links

About Marty Cornish

Marty Cornish is an entrepreneur and the founder and CEO of Workpack, a startup that brings automated quantity software to construction cost estimators. With well over 400M in project cost estimations successfully executed, Marty is reducing sunk costs at the same time as he is bringing reliability and ease to the complicated and time-consuming process of construction cost estimating.

In 2011, Marty founded Easy Paint, an estimation software for the paint industry. While this software blazed the trail for standardized painting contractor estimation, Marty realized the risk of inaccurate cost estimation was unresolved and developed a new, broader, and more comprehensive solution, yielding consistent accurate data industry-wide.

With his comfortable conversation style, Marty is a frequent public speaker and guest on podcasts, often discussing the intersection of technology, AI, and construction. He is on a perpetual quest to make the business of construction trusted, transparent, and more seamless for the many contractors who are interdependent.