Reinventing Construction with Data & AI with James Garner, Head of AI & Data at Gleeds

0:00It's really hard to take this technology and try and put a business case against things because by the time you've written the business case, sent it, got it approved, put an implementation plan together, it's out of date because the technology has then moved on to the next thing. So a better way is just to experiment in a safe place with it and come up with proof concepts.

0:23Welcome to Data & AI Mastery, the podcast where we bring you cutting-edge insights, practical advice, and inspiring stories from the leaders shaping the future of data and AI across the globe. I'm your host, Raoul Gabriel-Urmer, founder of Cambridge Spark, the leader in transformational data and AI upskilling, career development, and progression. In each episode, I will be diving into real-world case studies of companies harnessing the power of AI to drive innovation, reduce costs, and create new business opportunities. So whether you are an aspiring data scientist, AI engineer, or seasoned executive, this show is designed to give you the tools and knowledge to stay ahead in a world where data is transforming every aspect of business.

1:08Stay ahead, stay inspired, stay masterful. Welcome to Data and AI Mastery.

1:18Hey, James, how's it going? Good to see you. Yeah, great, Raoul. Thank you so much for having me on. I'm really excited for this chat today. Yeah, likewise. I can't wait to deep dive in your story. Also geek it out a little bit about AI in the construction industry. That should be quite interesting.

1:39So look, I think your story is fascinating because you basically joined Glides in 1998 as a chartered quantity surveyor and now you're the head of data and AI at Glides so we've got to talk about that so can you walk us through a little bit your journey? The initial plan was to become an academic of some kind because I really enjoyed that side of things and you know I'd quite like to go back into that at some point in in my later career as well but that didn't quite go according to plan because 20 years later I was still doing quantity surveying and then what happened was was in around 2020 I was always still interested in the tech side of things but also in the kind of economic side of things I got given the opportunity to take over a vacant position that was then called the head of R &D at Gleads but it was less about data it was more about market conditions and stuff so I did that for a while and then that department effectively got subsumed into a wider digital department which included all the stuff we were doing market research but was also starting to look at our newfound digital strategy and then that's evolved into now huge keen interest in ai and then the link between ai and data to where i am now so it's been it's been quite a journey actually yeah it sounds like quite the the roller coaster journey so since 2020 right in the action

3:09we've got to talk about data and ai right so yeah like what sparked your interest in in ai in the first place and then you know what are you seeing in the construction industry i'd love to hear kind of like maybe let's have an example of low hanging fruits that you think that you know the industry benefits from. I've always been interested in IT and computers and just watched with interest through the AI winters as the AI winters were the time when it didn't look like there was a lot of progress being made. But then the next big moment came, actually. It was in my Quantity Surviving career when I did a project for Imperial College in London who were a big client of ours.

3:52And I was lucky enough to meet an incredible man, and Professor Peter Chung from Imperial College, he was the head of the electric and electrical engineering department. I was doing a project for him in terms of refurbishment of one of their facilities. And he said to me, we got on quite well, and he said, why are you doing it like that? And I was like, well, that's the way we do it. And he said, well, look, I've got all these amazing students, undergraduate students, all looking into artificial machine learning, neural networks. This is the early days, really, of neural networks. And he sat me down, we went out for lunch, and he explained to me the possibilities.

4:31So obviously, the sci-fi geek in me kicked in, and I was like, wow, okay, this is coming true. So he said, well, I'll tell you what to do, James. Take a couple of my students, they need to do something called an industrial placement, similar to the sandwich year that I did, but for three months. So you take them into your company, and let's see what we can do. So we did. So this was 2016 maybe 2017 and these these amazingly bright kids who i'm still in touch with they're not kids anymore by the way and they're working in data in various other industries that they built our first machine learning model based on some of the data we had and and it was crude it was primitive but it it showed that this this stuff was becoming real and that's when i was like okay i've got to wake up to this what was it predicting or what problem was it solving it was helping to predict costs actually it was looking at a lot of cost data that we had and it you know it was a prototype it was a you know it wasn't and actually what actually showed us was that some of the problems with the data not necessarily our company's data but the industry's data and the way that we collect data we'd spent the last hundred years collecting data for surveyors to read not for machines to read and that's led to a huge mission for me working with the RICS now to make sure that We're collecting data in a standardized way that actually is machine readable as well.

5:51And what were some of the challenges? Like, actually, it's quite an interesting point you're making, right? Because, like, data quality is, of course, a huge issue and it can impact model performance. So, like, what sort of issues of conversions, like, mechanism did you have to put in place to make it work? So some of the main issues we came out with was just data quality in its widest sense in terms of who was verifying the data. because we were collecting data which was more than most people were doing but wasn't necessarily being verified at the right spots before it went into the database so making sure we had clean extracted data that was put into some kind of a table that that the machine could read and then the second part of it was actually about the industry and how we classify different costs so for instance you you might have two buildings and then somebody might put for instance let's say the carpet that goes with that building in one category, where someone else puts it in another category.

6:48So how do you categorize that? And then more challenging issues, such as what happens when you've got a building which is serving multiple uses? So it's got a gym on the ground floor plus an office, which is quite common now, these mixed-use developments. How do you then categorize those costs into the different types so that you can use that for machine modeling? So it became quite complex. It's very granular, and you need domain expertise, by the sound of it to like be able to go into that level of detail domain expertise so what we actually found from this is that you needed this person on the team called a translator who could understand both the domain because it's such a peculiar industry but also understood data science and that's hard that's like a real kind of unicorn territory in terms of finding those people i can see the venn diagram already yeah yeah there's not many of them and which is what kind of what forced me to do i ended up doing a data analyst qualification over lockdown just to keep myself busy, but also to try and understand what was going on with some of the...

7:47Because I became fascinated with what these guys had produced. So they produced this machine learning model and they were talking to me about the first time I'd heard about, you know, things like MATLAB and Python coding and R and all these different tools out there. And I was like, okay, well, I need to understand what these mean. So I ended up doing a data analyst qualification to help me understand it. And that's what sparked the interest for me because these guys had created this machine learning model back in 2017 and i was i was astounded by what they were able to produce and more astounded i was also very clear that there was a lot of work that we had to do to get our data into into good shape that was amazing and then the next big jump came so that that got us on the journey if you like and we started creating some models and you know i'm pleased to say that the that proof of concept that we talked about in 2017 is now live and we rolled out within glides and you know hundreds and hundreds of our staff using that version i was going to ask you about it because usually you know you hear a lot industry that uh you know you make it to pocs right but to get to production and realize value out of it that's a whole different ball game so like how did it happen like how how did you enjoy it is successful actually really incremental so what happened was is that those guys did their industrial placement with us they left us with this idea which opened up the thinking around it and then the next step was actually when I started doing the data analyst qualification and what I actually chose for one of the hack challenges that I was working on was to resurrect this this machine learning model that the guys done and work on it take it to the next level and in that hack we actually won that hackathon which was quite good because we we won some cash but But more importantly, we actually got this thing into a working prototype now, actually something that we could use.

9:41And then basically what happened from that, we then productionized it. We brought it back internally and been working on it. And then we were able to hire somebody who now, you know, that is their job. And it's obviously very different to how the original started. But so it should be as well. You want it to progress and progress and progress. But the germ of what we now have called benchmark was actually this thing that the two industrial placement students from Imperial College did back in 2017. So it's quite a nice story to see how that happened. Yeah, that's great. It resonates a lot because I did my undergrad at Imperial College and I also did an industrial placement.

10:20So it's kind of like... Oh, cool. You know what I'm talking about. Well, if I play it back, you know, let's be honest, right? like the construction industry is is not known for being particularly innovative when it comes to like digital technology right like if we talk pure of ai you wouldn't fear about construction industry so it sounds like there's a really cool story here whereby you know you brought along some students that are like playing around with machine learning model and pad them up or kind of domain expert in the industry and that kind of sparked ideas the art of the possible that unlocked the next stage which is like oh actually we can align that to a business problem kind of like forecasting costs and so on, which triggered, it sounds like, you know, hey, we need to figure out our data lineage or data pipeline or data quality if we want to really get the benefit and then get it to production.

11:06So you went through this stage. What about the next stage in terms of how do you scale, I guess, AI use cases? Like today in your role now, what are you looking at? What gets you excited? So some of the things that we started with were around some of our own information. so around particularly the team that i'm in the digital team had a lot of information so people would ask us lots of questions so i think one of the first things we created was a gpt powered chatbot that just enabled people to ask questions around what we did as a bit of a proof concept but then we actually look we we did some work with the rics and we created some lms fine-tuned on the industry data such as something called nrm2 which is the new rules of measurement too and then that basically became the tool that we now call assist which is our fine-tuned large language model which is you know now far more powerful than it was then but it was a you you talked about low hanging fruit well so a lot of it was low it was experimenting because people still didn't understand it in those days they didn't even understand what chat gpt was so it was experimenting and failing fast like trying it with certain use cases trying it with others sometimes we would just do it within a custom gpt within chat gpt sometimes we would use some tools that are on the market and then obviously when microsoft released the ai foundry were able to create models and then play around with local large language models for more data security use cases but the main point was we just expect we had a team of people who were just fascinated and just wanted to experiment and try things out and i still think that is the best way in this because because it's moving so fast it's really hard to take this technology and try and put a business case against things because by the time you've written the business case sent it got it approved put an implementation plan together it's out of date because the technology has then moved on to the next thing so a better way is just to experiment in a safe place with it and and come up with proof concepts and which which we're now doing now with vibe coding and things like that it's like okay well what does what does that mean and it feels like every six months or so there's another paradigm shift like you know we're mucking around with agents and now we're mucking around with vibe coding like rather than asking permission like you still got to ask permission but you don't have to ask permission if you're experimenting with publicly available data or you're experimenting with a data set from Kaggle or something like that you can just do it just do it and play around and work out what works and what doesn't work yeah i resonate with a lot where you said right that the pace of change is fascinating there was this talk from andre carpatti he was the director at tesla where he talks about software 1.0 where you write code manually software 2.0 your models actually kind of making the decision if you train them with data right so that's a power shift and now it's like software 3.0 where basically LLMs are handling most of it so we're going through some shifts and it's all happening quite quite fast but I guess one takeaway I get from from what you've said is definitely that the hacked on sort of a idea right like this decided to bring domain expert bring technical people give them like creative agenda cool ideas will come up and those people can exchange agents that's something I hear a lot in the industry kind of like a an initiative to put in place and sounds like you guys are like making a lot of use of it you're right it's the experimentation that is so important because what i find in the project delivery profession and i talk about that in the wider not just construction but project delivery could be infrastructure could be energy it could be it or whatever kind of project delivery a lot of them based on very kind of traditional workflows that some of them go back years and years and years and it's that's the difficult thing to break so you have the danger of people just digitizing a really old process without thinking about the actually why are we doing it like that in the first place and that's what the hackathon really promotes is to actually look at the problem statement in the first place and say don't just digitize what we've done because probably what we've done for the last 50 years is not fit for the new world and even if it is is there a better way to do it got it you make me think of the innovation dilemma the whole for like you know you can do a faster horse so you can like look at the car and like experiment with that it's exactly that it's exactly that but it's surprising how many people you know think that the answer is just to you know give a faster horse it is a problem in industry the innovative dilemma right like to disrupt yourself if it kind of works and you can do increment improvement versus reimagine your workflow i guess it boils down to i guess funding priorities culture i guess it's tough to like make that happen so hey look it's quite interesting to hear about you know those use cases actually so we've talked about cost forecasting knowledge management and if i take you to maybe a bit more like visionary now what excites you around what ai can do for the industry in terms of maybe you know not sci-fi like star trek 20 years time but what what do you see in the next few years that could be kind of disrupting and beneficial for for the industry it's a really uncertain time but it is exciting and i think i was talking to somebody the other day and it depends what side of bed i wake up on some days i'm feeling incredibly optimistic and other days i feel quite worried about uh this technology and where it's going to take us what is this today i think i'm feeling relatively optimistic today you'll be pleased to know but i will go through some of my concerns as well but on the optimistic side i think it's the it's the when we talk about ai everyone now thinks we talk about chat gpt but it's obviously a lot more than that and it's the congruence of all these different technologies together that is going to rise over the course of the next five to ten years so of course when we talk about ai it also includes ai enabled robotics and it also is around things like blockchain technology which can feed the agents because if you've got agent right if we talk about agents and like a bunch of llms working together and performing tasks the bottleneck at the moment is the way that it transacts because you know all the problems with kind of currency around different regions territory regions becomes a real blocker but if you bring in blockchain and i'm not talking about bitcoin because i know that people get quite quite emotive about it but just let's say blockchain and things like stable coins where they're tied to us currency all of a sudden that can break that bottleneck entirely because now you can transact within the digital environment without having to take money out into fiat currency so you've got blockchain then of course there's the huge advancements in quantum computing that's going on at the moment now what that will do is it will boost the amount because the another kind of bottleneck on AI advancement is the amount of compute and then the sustainability concerns that comes with that but of course it's a chicken leg situation because as AI gets better and better well hopefully that will sort of answer some of the questions around quantum computing and once quantum computing becomes kind of mainstream then that's going to answer a lot of the compute questions and that will be self kind of fulfilling in terms of making better and better models that helps with the sustainability challenge in the end but all these things rise together right so what's what i'm excited about is how agents are going to change our workflow to enable us to get rid of the grunt work so that we can all add more value i'm really excited about the use of robotics lots of an AI robotics enabled workforce in construction sites because again talking about bottlenecks the reason that construction sites are always late and always over budget is because it's dependent on human beings and rightly so things like say health and safety come first right and then also things like human beings need to take breaks and they need to sleep and they need to you know they need help when they've got to lift something heavy robotics will change all of that right so it's not to say there won't be humans on site but there'll be there'll be these humanoid robots and different types of robotics to assist them so i'm quite optimistic that within the next five years we're going to see huge improvements to our productivity within construction energy infrastructure because of these advancements what i'm concerned about the days that I'm feeling not so optimistic when I get out of bed on the wrong side is the willingness of the industry to embrace this technology and what happens if we don't so putting aside the kind of societal impacts of AI my worry as an industry is that if we don't grasp it and actually make this work for us then other people will and we'll lose control of the industry what i mean by that is software companies taking control now software companies are a big part of it but i think it would be really really bad if software companies effectively take control of the industry like they have done in other industries let's take netflix as an example of taking control of the entertainment industry there's a lot of negative consequences that come with that there's some good ones and netflix maybe is not the best example because they actually has actually given a creative kind of a renaissance with with tv programs but certainly in the music industry if you look at spotify and things like that's had a serious negative impact i think on what happens in in the music industry in the way that our artists are paid so we want to be careful as an industry that we don't fall into that which is why i'm so passionate about this having spent 25 years most of my life in this industry and i care deeply about it is i don't want to see it go down that route because you've got one one shot at this if we get it wrong we get it wrong and there's no going back there's no you can't turn the music industry back now it's it's gone sure that's that's really interesting so it's quite a cool nugget so if i play this back you know next few years is a general like opportune challenge for the construction industry the project delivery industry to i guess be more productive improve health and safety through the use of ai but you know the industry needs to adopt ai in the first place so you know there's a bit of a evolution adoption and the business model might have to to evolve with it otherwise those big tech companies you know could disrupt you so you know i think that's fair and then kind of like your kind of visionary you know if we leap into the future and who knows how long it is you've got a few interesting ingredients blockchain quantum computing ai so we can see all the seeds and sometimes the values integration right like we've seen this with ai with you know i guess the models today are particularly good because we've got lots of data we've got a hardware and the algorithms but you know those were kind of like available and they grew separately and then boom the combination make it a big deal so that future could be fascinating at james

22:32so let me take you to the quickfire round basically short questions and just kind of want to get your your insights sometimes thought provoking so we'll kick off with any contrarian view you have in the industry something i can't disagree with people yeah a little bit that data quality is everything i think data quality is important but i think we've kind of moved beyond that now these large language models are incredibly good at dealing with un-messy that were messy data now that's not saying i'm not that you forget about your data quality but i'm saying that's not an excuse not to do stuff interesting okay that's that's a contender view i like that one and if you had a magic wand then you could solve like one problem in in your industry what would that be attitude and and culture putting this to the top of the agenda in terms of investing in people's skills awesome so yeah education change management yeah it's kind of like that's always hard and takes a long time so i understand yeah but i think it's some i think a lot of people have got good motive you know they want to do this but it always gets pushed down to you know nice to have whereas i think this is this is now we're in the territory of this is a must-have to be relevant in the future that you've got to upskill your workforce oh 100 being a an educator myself like learning is is no longer nice to have right like it's changing so fast if you're done so learning is kind of like a vital part of your day-to-day at this point you've got to i'm 50 years old nearly 50 and i'm going to be you know i'm starting the masters which i never thought would have been anything that i would have done but i feel like now we're all going to hopefully please god live for a lot longer and we're going to be working for a lot longer and we're not in a world anymore where you go to university in your 20s and and get a bunch of skills that are going to be relevant for the rest of your life those days we've got to get used to the fact that probably every five years we're going to have to retrain into something 100 yeah just like yourself big believe in the apprenticeships model to do that next question a bit more personal so what's your favorite programming language oh well i'm gonna say python simply because i'm not really a programmer and that's the only one i know or apart from back when i was a kid i used basic on the zx spectrum actually make that that's my favorite because that was easy i could do that oh my god that's like way way back you know yeah Okay.

24:53But no, Python's the one that I'm using, and I think it's almost becoming universal language now, certainly in terms of machine learning and AI. So I would go with that. Hey, what was your favorite subject at school? Music. So, yeah, that's my other passion, I guess. So I still do a lot of music. I'm in a number of bands. I can see the guitars in the background. Yeah, yeah, got a few guitars there. So I love music. And it's interesting because music and maths go quite well together. So it's not, people think, oh, two totally unrelated things in terms of, you know, AI and music. But I think they go quite well together, actually.

25:37And one releases another side, you know, if I've got a headache, the best thing I can do is pick up my guitar and the headache's gone. And vice versa. Well, you beat me to my last question, which is, what's your favorite music genre? Oh, that's interesting because I like a lot of music. I play in like a covers band which does all the kind of party stuff and then I play in a little jazz band. I guess my music though, like coming from the 90s, was that whole kind of 90s rock era, the grunge stuff. That's, you know, going back to my youth, that was my real kind of passion, the Nirvanos and Pearl Jams of this world.

26:20hey james really enjoyed talking to you today you too thank you so much for having me on

26:30super cool talking to james what a really interesting guy full of energy we definitely had a great conversation there's a few interesting insights i think the first one was how do you bring you know ai in your organization especially in an industry that maybe is a bit more traditional right like the construction industry the project delivery you wouldn't think about ai first so what they did is bring a few like experts i know about machine learning you know lots of energy and pad them up with domain experts and kind of create ideas and focus on the use case in this case was cost forecasting so that's a great way to get started build a poc prove value and then expand that out into hacktons, create more ideas, and then bring those into production.

27:19So that was an interesting journey. Second insight was that clearly generative AI through ChatGPT has created a bit of a wow moment for leaders industry, including executives. So now everyone's asking like, hey, how can we utilize AI? How can we channel that to business problem? What ROI can we get? So there's definitely a lot of interest in industry and everyone's trying to figure out how to get there. So that was also a great insight. Thank you for tuning into this episode of Data and AI Mastery. If you found value in today's discussion, make sure to subscribe so you never miss an insight from the leaders driving the future of data and AI.

28:00And if you're a data and AI leader looking to upskill your workforce with the fundamental data and AI skills to transform your business, Cambridge Spark is here to guide you every step of the way. Be sure to reach out to us on LinkedIn or on our website, cambridgespark.com. Until then, be sure to keep pushing the boundaries of what's possible with data. And remember, mastery comes with continued learning and action. Until next time, stay ahead, stay inspired and stay masterful.