How To Help AI Reach Its Full Potential In The Construction Industry
AI and data driven decisions are the hot topics and buzzwords of the day. But what does that all really mean in practical terms and what are the prerequisites and best practices as it relates to what machine learning and AI can achieve to drive data driven decisions.
Artificial Intelligence (AI) is only as powerful as the data that feeds it, making the quality of data critically important for AI systems to deliver accurate and valuable results. On this episode of the ProjectReady podcast, we delve into the intersection of AI and the Architecture, Engineering, and Construction (AEC) industry, exploring the challenges and opportunities of leveraging AI in decision-making processes with industry Digital Transformation expert Jeff Walter, and Salla Eckhardt, Senior Vice President at OAC Services, Inc
Ultimately, the goal of this episode is to challenge the notion that AI, alone, is the key to data-driven decisions and assert that the real challenge lies in providing accurate, comprehensive, and scalable data. After all, as the old adage goes “Garbage In becomes Garbage Out.”
With the sheer volume of data spanning across a project’s phases and actors the AEC has its own unique challenges. And what is key is to understand the human interactions and collaboration driven by the workflows and content that comprise the history of a project’s transactions.
With the focus on the importance of high-quality data and its integration across the AEC industry to feed AI’s potential, listeners will learn more about:
- The need to bring together disparate data sources and contextualize them effectively.
- How scalable taxonomies driven by metadata provide the foundation for AI to make a significant impact.
- Use of AI today and the potential of AI in the AEC industry.
Sign Up For Updates
Check out these additional resources:
How To Overcome The AEC’s Siloed Data Problem
Contact Us Today For A Free Demonstration!
Transcript
Joe Giegerich:
All right, everybody. Want to thank you for coming out to today’s podcast. Very excited on this particular topic, and to have digital transformation expert Jeff Walter and Salla Eckhardt from OAC joining our call today. We’re very big on taxonomy. The whole way our product works is to scale on taxonomy, and you just see all this stuff about AI in the press and data-driven decisions, but the old adage, I thought at least, holds that garbage in, garbage out. So how do you prepare your data? What data are you analyzing? What are the use cases? And there is actually a piece in January in Construction Dive about how AI is here and it’s going to transform the industry, but they call out the fact that data is siloed, there isn’t connection of data to the other vendors and partners on a project. That’s a problem. And so that’s my contention, is that if you don’t have your house in order, AI is only as good as the data it has to chew on.
So with that, I’m Joe Giegerich. I’m the founder and CEO of ProjectReady. On the call today is also Shaili Modi-Oza, who is our head of development. Jeff and Salla, if you would introduce yourselves.
Salla Eckhardt:
Thank you. Good afternoon or good morning, everyone. I’m Salla Eckhardt from OAC Services, and I’m currently the senior vice president of commercial market at OAC. It’s our largest market and it is the largest opportunity for having discussions like this, so thank you so much for inviting me onboard.
Joe:
Yep, thank you.
Jeff Walter:
And hello, everyone. My name is Jeff Walter. I’m sitting in Toronto, Canada, here. I’m currently the digital consulting lead for the Canada region, as Joe mentioned, helping clients go through a lot of the similar challenges that have been prefaced here, but definitely also come from a background of big infrastructure delivery, having been involved with a number of large infrastructure projects from big complex bridges to subways, et cetera, and like to keep my feet on the ground in terms of what’s happening there, as well as kind of guiding different clients to those challenges, as well. So that’s my background.
Joe:
Okay. And trying to figure out the best way to kick this off, can do something semi-cute. What does AI mean to you? Where do you see the application of artificial intelligence being applied in the industry, both based upon your experience and where you think it’s going?
Jeff:
Do you want to go first, Salla, or…
Salla:
Sure. Where I see a large potential is in the generational transformation that the industry is going through as it’s going through its digital transformation, and both journeys or tracks are never ending. But what I see that AI can do is to put more human-centric approach into the projects, and AI can do a lot of the tedious or labor-heavy, burdenous jobs that people these days are not necessarily interested in doing, but they need to be done and they need to be precise and verified, et cetera. But AI can help us elevate our skillsets and make more room for more in-person or teamwork online using digital platforms and give people more the opportunity to exchange thoughts and brainstorm together, innovate more as they don’t have to be working heads down on the traditional processes and deliverables.
Joe:
And what about you, Jeff? What’s your take?
Jeff:
Oh, I really like the fact that, Salla, that you led with the human kind of side of AI. I think as we’re all seeing in the news and the industry, I think everyone’s trying to scramble to find the balance between the human component of these emerging technologies and not losing those elements and empowering them really with this type of technology, but also realizing the technical and efficiency potential of AI. And AI to me doesn’t cover many different kind of elements. One is dealing with this type of structured information that we’re kind of talking about today, but there’s a lot of different other areas for AI application. But I think generally where I’m seeing from my lens, the greatest opportunities emerge is really where the most data is. The real application of these types of application of these types of technologies into our industry.
I guess some of the early challenges was having enough data to help support some of those outcomes, and I think in certain areas of our industry, we’re starting to see the rapid exponential curation of data in different areas of our industry. And I think that’s representing opportunities for AI. And I guess where the earliest signs of it applying to our industry and again, related to how much scale of information is starting to look at big picture scenarios and whether it’s from a sustainability picture or operational or analytics, being able to see big picture and get insights and start to understand where those futures could potentially go, that’s where I’m seeing a lot of opportunity sitting present. So the big picture where data is flowing up to high levels of potential insight.
Joe:
So the human element obviously counts. Everything is to serve man, not to reference The Twilight Zone, if you know the episode. That’s a great episode. But the question is, where are you seeing AI applied now and what are some practical applications that you see immediately, and sort of desired outcome? And Salla, if you would like to start.
Salla:
Yeah. I think a majority of AI usage now that is still very early on in its development phases is in verifying for confirming people as decision makers, that they are making the right choice. People like to have some kind of a nudge that yes, you are on the right track, keep going, and AI is providing that because it’s so quick at analyzing the data into information and refining it into information that then becomes knowledge for people that we can almost have responses or replies instantaneously rather than spending a lot of time doing research and verifying what we already think is the right choice.
And then, people tend to be visual thinkers and consume information differently, and also comprehend and interpret information differently. So having AI support creating unanimous understanding of what the implications of a decision are, we see people use AI for that and using especially visual tools that run on AI to seek the understanding and alignment with the rest of the project teams so that they are not in misstep or having blind spots or gaps in their own knowledge. But AI is really giving us that extra confirmation that we need as human decision makers, as our projects are becoming more and more complicated and complex, and hence the risk related to finances or project budgets are becoming greater as well. But AI overall, it’s in early stages of usability for the built environment industry, but there’s a lot of promise in it.
Joe:
Yeah, and it strikes me that just picking up what you said, yes, it’s the human element, but at some point for democracy to move forward, you have to stop voting. And so if I understand correctly, one of your contentions is that AI can settle debate. Good debate is a healthy thing, but too much debate leads to uncertainty and then becomes a guessing game. Is that fair?
Salla:
Exactly. Thank you.
Joe:
Sure. And Jeff?
Jeff:
No, that’s a really interesting perspective there for sure. And definitely I can see the capabilities of supporting decision making and traceability and kind of auditability are key elements around AI for sure. I guess just a couple of thoughts on practical examples that I’m seeing in the built environment, and I know this is not about product marketing here, but definitely has been involved with the development of tools and toolkits that are supporting AI for clients. And for example, pipe inspections, which is just as I guess hardcore infrastructure as you can get. Many times those inspections involve spending countless days and weeks going through video and then trying to identify different outcomes from that.
And so we’re leveraging AI technology from a video and image recognition perspective to essentially allow that individual to free up more time to do other things instead of doing a lot of that processing. So there’s a lot of practical advancements where you’re seeing large volumes of information again, whether it’s structured data, unstructured data, video that kind. So there’s real kind of practical application there. But I think maybe as it relates to where we’re going in terms of integrated data environments and silos of information, and from my experience working on large infrastructure projects in silo, you can definitely attest to this. Many times you are dealing with, again, on a very scaled level, different software sources that are producing and curating data. You could have schedule information, cost information, asset information, sustainability information, all these different sources of information related to the delivery of that project or program.
And I can definitely see a maturity in the market in terms of being able to integrate those sources together and normalize and report, et cetera. But really I think there’s a great opportunity for AI that I’m seeing to add layers of context for information in different silos. And I think that’s a real practical application as we’re starting to invest a lot in infrastructure growth in North America in general, is kind of getting a grip on contextualizing information, getting more insight from all of these typically siloed sources of data.
Joe:
So, two things I would add to that, and I do want to bring Shaili into the conversation as well, is for one, I was reading a very interesting article. I believe Hungary is the country that’s leading the charge on the application of AI to mammograms. Because what brought it to mind was the pipe thing. There’s so much data to sift through and they found that now with AI, they’re getting better predictive outcomes of finding things earlier and of a particular type. And so this is sort of a wide applicability, I guess, of one of those applications of AI.
And the other thing, just as a slight pivot, we just did a couple of podcasts around the art of the possible, which is really about the emergence in the industry of more and more and more standards to exchange data. And so Shaili I just discussed this at length, and Jeff, you mentioned un-siloing data. What’s view on how to get the industry together and to get that data in a way that people can start to share it and make sense of it? Or is that too open-ended a question?
Salla:
It’s a good question. Overall, the discussion about how people might start to adopt the new digital platforms and digital tools, we definitely need standards so that everyone has the same platform to start building on top of and there are rules that can then be applied into the processes that should need to be pre-engineered now that we’re introducing AI as a new coworker into the mix. So overall, taking the approach of being very inclusive and taking the approach of open data as much as we can, but not forgetting about cybersecurity and data security. So it’s a different discussion and rabbit hole, but creating opportunities for citizen developers as well as professional developers and engaging people into that collaboration, that’s the key that we’ve all been part of all kinds of standardization groups and small R&D groups, et cetera, and we’ve all seen that we kind of boiled the ocean as a small team, then others couldn’t keep up.
And I’m hoping that with AI and with greater amounts of usable data and more verified data, higher quality data, we can actually crowdsource more of the development of solutions rather than keep it siloed into small groups that will make leaps, but then it’s hard for others to follow.
Jeff:
Yeah. I can definitely add onto that just in terms of… I totally agree with that and it comes back to this question of getting data in the best foundation for value and growth creation there, but just in different practical kind of application of standardization and compliance, and I know Salla, I mean, we’re both interfacing with different kind of groups that are looking at the life cycle of information and infrastructure and how to manage that work, the data flow through that lifecycle, as well. We’re definitely being influenced right now by things like ISO 19650 information management requirements, which is a kind of lifecycle compliance requirements, auditable, traceable environment for delivering infrastructure. And that’s really changed a lot of perspectives and focus for data management in the built environment is having those compliance requirements coming through on our requirements for projects. And as well, certainly on the environment and sustainability side of things, regulation around data management capture, validating engineering is also becoming important as well.
So I can see a few different emerging regulatory and compliance elements impacting our industry as a whole. And I think that you were saying, Salla, is also an important way of getting that data into a foundation for creating those opportunities.
Joe:
But practically speaking there, so the legislation will drive the final, how can I put it? Will drive the final drive. It will finally force the industry in large measure to start to come up with more and more consistent standards across organizations and platforms. Is that fair?
Jeff:
It’s definitely been a big push, for sure. I mean, I think inherently would’ve happened, maybe naturally, so I’m not sure, but this definitely accelerated maybe kind of the industry.
Joe:
I have a lot of clients. I don’t think it would’ve happened naturally, personally.
Jeff:
Maybe.
Joe:
And I’m a born witness to Salla’s comment about boiling the ocean, I’ve been in panels and the like, and I’m like, “Can we just agree on five terms in a submittal that we can pass data on?” Not this full overlapping ecosystem, but just to start to chip away at it. And that’s my personal view on it is that the industry is so used to engineering big things they’re not addressing the small things that are really hamstring that effort. Now from a technical perspective, Shaili, what do you see some of the challenges around getting consistent data around and how AI could apply and where you see the limitations for AI as it relates to data purity, if that makes sense?
Shaili Modi-Oza:
Yeah, definitely. And I think just definitely to start with, like all of you mentioned, having some sort of a standard in place is something that is very difficult to achieve, but starting small and having some sort of a starting point where we have standards and how we can start aggregating the data in a way that it makes that data meaningful. The key ingredient where then AI and machine learning can start to even help would be that the data needs to be accurate, the data needs to be in a particular format where the AI can essentially make sense of it.
So I think from a technical side, having all of these various data systems where the data terms are standardized, bringing that together in a way that it has that kind of a fixed, reliable, high quality data, which can then be used to analyze and it would essentially have that layer of machine learning on top of this data, which would then essentially help these systems to learn. But I think that the biggest technical difficulty here is it needs a lot of advanced hardware and software, which is very expensive to even get that in place to begin.
Joe:
But that’s the cloud, right?
Shaili Modi-Oza:
Yeah.
Joe:
I mean, they’ve already done that.
Shaili Modi-Oza:
Mm-hmm. And but essentially having that training, what kind of a workflow are you looking at? We can take an example of let’s say change orders happening in a system. They’re kind of tracked differently in different systems, but having that kind of a layer on top of your data, how long does it take to get information in, what are the costs associated with a particular phase of a project? And once all of that data is hosted in a system, having that AI layer that makes sense of it and gives that accurate analysis essentially of data, which we’ve been talking about, it’s kind of going to help the managers and the people who are working in these systems to essentially basically get that kind of analysis.
Joe:
Yeah. It’s like I used to be a researcher in college with the Russell Sage Foundation, and I would go to the bowels of the New York Public Library, lifelong New Yorker, and I would go into the basement, but at least I had a card catalog system. So I was talking to a friend at Microsoft and he goes, “Well, AI, you can look at and do semantic matches.” Yes. And I could also walk through the stacks of books in the New York Public Library and eventually find that book. But clearly the Dewey decimal system was invented for a reason. I think it just had its 130th birthday, in fact. My was a librarian. Near and dear to me.
But that’s where I really think the garbage in, garbage outpour comes from. Yes, you can have a data set, it can chew on it, but all these things, all this FUD, fear, uncertainty and doubt that you see about ChatGPT, oh, the answers are wrong or they come back hostile, it’s because the quality of the data is awful. How-
Shaili Modi-Oza:
Yeah, and that can cause a lot of issues because users are still trying to get used to even relying on AI and incomplete and incorrect data, which can just mislead and cause a lot of problems while we are in these early stages. So I think it’s very important to get the first part right where the data that goes into these systems is organized properly.
Jeff:
Yeah. I mean, I’m kind of, again, just to bring that pipe example, I’ve been seeing this kind of firsthand, the effort in terms of maturing the quality of an AI solution. And when you think that you have it ready and learned and ready to be something, there’s something else that always comes. So I think definitely starting from that foundation is important, but it is definitely an evolving thing. And I guess I was going to maybe ask your guys’ perspective on this as well, is within the context of organizations and projects, how do you put… as this becomes an emerging area, how do you put the quality control in place to make sure that that data and those results are meeting those needs?
Salla:
That’s a good question, Jeff, and I’ll hop into the discussion at this point because we thought about the same question a couple of years ago. That’s why we delivered a project for data master plan, to respond to what Shaili said. It’s starting small so that it’s something that is manageable for people, then allowing change to happen, too. Because if human beings, we want AI to always execute the same process and deliver the same outputs, we can never expect a different outcome.
And that’s why I’m curious about creating a data master plan that will allow creating different nodes as milestones for the data that is actually being delivered and creating different kind of paths to achieve the outcomes that we want to achieve, and that way enable the continuous evolution and development by using AI. And that way we are not continuously repeating the same process and same outcomes and outputs, but really enabling the AI to support us in improving. And that’s why we need to have the data master plan so that we’re not overwhelming people with big data and expect them to be able to comprehend it all, but really make it more targeted and meaningful and enable people to grow the number of different nodes and the different milestones, and that way have that continuous evolution and development as a whole.
Joe:
And that would be applicable to an individual company and very much needed for the entire industry. So basically approaching data like you would a building. We’re going to do the foundation and then we’ll do the scaffolding or whatever it may be, little bits at a time. I’m all about steady, incremental progress. It’s been the lesson I’ve learned in life.
Jeff:
I mean, again, this is just… I know there’s an industry level of conversation there, but even I’m seeing in something like the ISO 19650 requirements on delivering infrastructure, there are elements and documentation as part of that that create a scenario where you do have to present that data master plan in a sense that you’re talking about, Salla, where you’re giving those milestones exactly when every component of that asset is being delivered to the construction, to the trades, to the operator. It’s a full kind of data map of a potential project and that’s contained within that. Now, I haven’t fully seen that implemented yet, but there is a structure there for a data master plan in some of that compliance. But yeah, I like that idea, data master plan, for sure.
Joe:
Yeah, that’s great. And this also makes me think of the other thing, too, is what do we mean by data? It’s like AI itself, it’s such a big term at the end of the day. There’s a lot of data out there that has to be massaged. And I think you can run these things in parallel. The industry obviously, because you build amazing structures, will look at data for, well, you could apply it to a digital twin, et cetera, and so forth. Obviously our bailiwick is the transactional components, the collaborative components that exist in a project both within your team and externally. So is it accurate to say then that your data master plan, are they run in parallel? Is it one big plan chipping away at all these different corners? I don’t know if the question makes sense. You have a lot of data to wrestle and they have different applications that you’re going to want to apply AI to.
Salla:
Well, thinking about how to apply AI for delivering the data master plan or using it as a platform for a project delivery, once you have the master plan, then AI can start doing the pattern recognition and recognize if there’s anything that is an outlier in the data that is being used, and that way it’s easier to course correct less in time rather than wait for issues to become problems later on and then they become more difficult to fix because then you don’t exactly know where the pivot point happened.
So the data master plan, it’s kind of a live document so to speak, but it has the core structure of what needs to be delivered at what point in time and what nodes can fill in the gaps of data and what can AI continue to learn by itself based on the data that it has access to. And that way there might be multiple different journeys or paths that it can take in order to achieve the final outcomes that we want to achieve. And that’s what I see as the potential usage of AI, that it’s not a single direction journey. There can be multiple paths and they can crisscross and merge-
Joe:
Yeah, exactly.
Jeff:
Yeah.
Salla:
… and then divide again.
Jeff:
And how those multiple paths are enabled, I think. And just maybe this is a segue into the data master plan also kind of becoming, I guess part of it is a metadata master plan, and certainly as you’re saying, Salla, there’s the great opportunity once you’re starting to aggregate all these silos information and getting to a level where you’re able to identify those gaps and potential outcomes. Sorry guys, I just lost my train of thought there.
Joe:
No worries.
Salla:
I think I get what you’re saying, Jeff, and where I’m going with your thought track is that once we start having AI recognize some of the paths and start to understand where data is actually created and produced and where it’s consumed and where it needs to be refined into information, then we can start seeing which paths need to be amplified and which paths can already be retired from the overall process, and that way we can start simplifying the construction processes that we are managing today because not all of them need to exist as we go through the digital transformation.
Jeff:
Yeah. Thanks for the reminder, there. So definitely exactly what you said in terms of how metadata potentially fits into that is in that the common data master plan where we’re collecting, I think there is an opportunity to add new layers of context to that information that again, like I was just mentioning, enable those pathways in a sense and maybe even create new pathways that might not emerge from that structured, I guess regular data. So I think metadata becomes an important part of seeing those opportunities.
Joe:
And metadata becomes, I mean, it’s all metadata on some level, but metadata, I think in the more traditional context, it’s the lingua franca, frankly, if you’re going to be exchanging data, and it’s the layer of language humans can most heartily sort of belly up to the bar on as opposed to underscore this DBO object, what’s the metadata? The metadata is how humans I think interact to describe things. And then you can have a common vocabulary.
Jeff:
And I think that’s maybe where a great opportunity for AI is to be able to rationalize all that, right?
Joe:
Oh, and analyze it, too. What’s the most common terms that people are using.
Jeff:
Right?
Shaili Modi-Oza:
Yeah, definitely adding that layer of advanced analytics on top of this, that’s what makes it so powerful. It can kind of recognize looking at the metadata, the trends and how it changes across the project and essentially that’s what makes it that powerful, for sure.
Jeff:
So I guess in any kind of decision making scenario on a construction project or whatever, the more of that contextual information and the intelligence that comes with that creates those opportunities for different scenarios in terms of how a decision is made. Again, getting outside the box, Salla and everyone knows here, sometimes the linear thinking that we take when we’re approaching decisions, whether it’s only cost, only schedule, sometimes that’s the only focus. But having that wider lens, it kind of empowers those different kind of solutions.
Joe:
Yeah, it tends to be myopic. Just the accounting system, just my one system, just my one silo, just my one company. Is that where you’re going?
Jeff:
Yeah, yeah.
Joe:
Well, I guess this is… been on this for a while. Maybe it’s time to start to move toward a wrap up. So the challenges we know are getting standards opened up, a continued maturity of APIs to even allow these systems connect, the trust in the governance of the exchange of data. Personally, I think anonymized data would go a long way in the industry because then you could share more effortlessly and having more standard terms. But with all of that, have everybody take one more whack at where do you see, just one more time with feeling, where do you see it really having immediate value and long-term where you think AI will really be an effective tool to drive data-driven decisions? Too many drives, but…
Jeff:
I think when it comes down to decision making support, I think it also comes down to the speed of that feedback loop. I know again, I think we’ve all been around the kind of digital twin concept where you are using a digital asset model to drive feedback loops in information, and I think AI is going to help support that digital twin feedback loop process. And whether it’s an architect who’s using Revit and chatting with a bot that’s supporting decision making in that way or it’s rolling up managing big portfolios of infrastructure and getting faster feedback loops as part of that overall digital twin approach. So I think there’s a unique link between the emerging concepts around digital twin and AI, but it’s all really about a feedback loop.
Joe:
Right. And the quicker you can make these informed decisions, the less risk, the less rework, the less disasters, better bottom line.
Jeff:
Yeah, for sure. And again, I’m not saying necessarily that efficiency is the only outcome of that opportunity there, because certainly as we just talked about opening up those channels for emerging ideas and solutions, new thoughts in that feedback loop is also really important.
Joe:
Gotcha. Salla?
Salla:
Yeah. I see potential in multi-generational organizations that we can actually upskill the new into the industry people and re-skill those that already think that we know it all, and that way create completely new level organizational wisdom of how to carry on as the projects become more complicated and there are more demands from the developer side of what they want to achieve with their project, et cetera. So AI is a very healthy way of challenging what we think is best and creating that healthy conflict without crushing anyone that is collaborating with us or working on the same projects. So I welcome AI as a new teammate and a new partner.
Joe:
And an arbitrator.
Salla:
Yeah.
Joe:
And Shaili?
Shaili Modi-Oza:
Yeah, I think all of the above, and I would only add that we are in the early stages of adapting to AI right now, and as it becomes more in use and over time, all these algorithms that are in place, once they get more enhanced and they kind of learn more over time, and it’s only going to make these predictions even more precise and effectively just get better with time, I think. So we are going in the right direction, but it’s going to learn itself, so I think the more data goes in, the more algorithms it’s fed in, the more effective its output is going to… it’s going to keep getting better.
Joe:
And I guess my final comment would be sort of what the topic was, that as data improves, as its relationships improve, as standards improve, AI, because of how much it can do for your bottom line, much like ISO 19650, they drive an outcome. So you want to use AI, you need to have better communication of data, better standards of data, better taxonomy that scales. If you want to take care of 19650, you need to be able to… same thing. And so those are the other two trends that I think not just the tech itself that’s driving this, but that there’s this understanding that for this to go well, the quality of the data, the interchange of the data, you got to clean up the room so there’s no garbage going in so you can get quality coming back out.
All right. That’s my last piece. Look, I want to thank everybody very much for taking the time out today. We’re going to continue to do these things. Jeff, Salla, again, thank you very much. And Shaili, always.
Jeff:
Thank you guys. That was nice. Yeah.
Joe:
Okay.
Salla:
Thank you so much.
Shaili Modi-Oza:
Thanks.
Joe:
Excellent.
Salla:
It’s my pleasure.