Beyond the Pilot: How AEC Firms Scale AI Adoption

Speaker A: Foreign.

Speaker B: AEC Business the Construction Industries Innovation and Technology show hello and welcome to the

Speaker A: AEC UH Business Podcast. In this episode we discuss how AEC uh, companies should implement AI in their businesses to realize the full benefits of the technology. My name is Arni Hayskanen and my guest is Adeline Chan, the CEO and co founder of Hong Kong based AAL Innovation, an AI business transformation consultancy academy and multimedia creative studio. Thanks for joining us, Adeline.

Speaker B: Arnie, it's a pleasure. Thank you for having me.

Speaker A: Can you tell us a bit about your background and what inspired you to start your company?

Speaker B: Of course. So my background is that I was and still am a licensed architect. So I studied to become an architect and in the process I found out a reality that wasn't very clear to me when I was studying to become an architect. And I thought to myself, I wanted to prove something that want to do all the steps. I want to get my license, I want to start working in a design firm, I want to work uh, in a contractor, I want to even start my own design firm. I did all of that to understand the, that actually there are certain processes that can be transformed. There are just ways that are, um, rules, rules that could be broken, let me put it that way. And I started off as that. And then it came as a moment when AI arrived. Well, it has been here for a while, but when it started to boom, I then thought, this is actually uh, very interesting, there might be some answers in here. So even before AI was here, I always thought of these processes are repetitive, which is contradictory to why I wanted to do architecture and construction in the first place. It was to solve problems, it was to connect. And it came as the right moments, I think. And um, I decided to look into how can I bring impact, how can I bring value? And I was looking into what everyone was asking at that time, how can AI help us? How do we adopt AI? And down this rabbit hole I discovered a market demands and I looked into what could be possible and just really started from there. Really there was no one to shadow then and here we are. So thank you. That's a bit of my background and the journey there.

Speaker A: So, um, you discovered that architecture is not just about creativity. There are also processes that take a lot of energy from, from, from the creative part.

Speaker B: Exactly. And I think now, I was just reflecting about this the other day that I am now putting the creativity in the uh, business strategies. I think I'm now putting in the creativity in. Oh, this is how we did things in the past. What could be a creative way now to implement AI to then make things easier and what would that look like? And so I think now the creativity is really in here. And so I think it's a mix of design thinking with engineer thinking with business thinking together. So it's a very fascinating uh, year so far. Uh, well I've been doing this for two years and a bit more now, but particularly early 2025 to. Up until now it's been quite, quite at speed I would say. And, and quite fascinating. Yes.

Speaker A: Before we uh, talk more about AI, specifically from, from companies, AEC company's point of view, can you tell uh, a bit more about your company's services and who, who your clients are?

Speaker B: Yes, sure. So we have three teams in our company. We have the consultancy team, we have the academy team and then we have the creative studio which is the multimedia that you just mentioned. And for the consultancy we serve three different types of clients. One is the kinds of stakeholders in the construction industry. And it ranges from a real estate developer to an engineering consultancy to uh, a subcontractor. And we help them to go through AI transformation and AI adoption. And every time there's a bit of a diagnosing period and we kind of triage their problems into um, different categories. So hence we have a database now which helps us to find a solution a little bit quicker, at least find a direction a little bit quicker. And the other two groups would be European investor groups who are interested in investing into Chinese construction tech or prop tech startups. So that's also a very interesting demand. And then also anchor buyers like contractor groups who wants to buy specific hardware or tech from China as well and they want to come shopping. And then there's also a roadshow demand there as uh, well, which is very interesting. So along this process, now along this process of pre sale and the sales uh, cycle in general is very long. The reason is because sometimes people don't know what they want and they don't know what they don't know. So here comes the service of education, training, the leadership level and the workforce level. For the leadership level it's about direction, it's about data governance, it's about investments, strategies and uh, the whole concept of change management. And in the workforce level it is about productivity and of course safe to use of AI. Then the creative media is taking AI and plug into design. Basically what my team has been studying their whole lives, um, which our backgrounds as architects or designers, we always have one foot still in there and we deliver the designs now using The AI that we have done recently, research on, evaluate on and we have our own workflows and that is functioning on its own sacred media. We serving real um, estate developers to do feasibility studies for them and designs as well, casino uh, groups that's quite interesting. And villas in Bali and schools, international schools in Hong Kong, um to construction methodology animations that we do for contractors. So anything that's visual, that's even interactive world building, that's a lab on its own. So yes. So those are services that we provide.

Speaker A: Yeah, sounds like a good portfolio of services and very exciting ones. Uh, actually yeah. Um, when we think about AI implementation in companies, I often see that they start with some sort of testing and piloting and everybody's enthusiastic about something. But it's really difficult to actually scale these pilots uh, within the company but also across the whole project portfolio. So. Mhm. Do you see that as a problem? And have you found any kind of methods, uh, or ways for scaling uh, the AI implementations?

Speaker B: I love this question. I have two thoughts that came to mind just now as you were saying. This one is the pilot mindset and what that is, that is the comparison between the finance industry versus the construction industry. In the finance industry, banks, they cannot afford to be 10 times lower than their competitors and they will be 10 times lower if they do not keep piloting and testing and trying to push the edge and therefore from speaking to the financial sector because I want to see how I can cross contextualize the application to construction. I realized that what they're doing is they're at the same time simultaneously having 200 POC teams like different teams testing on proof of concepts and then competing against each other internally to find ways to which one is going to work and who's going to work. And they will get um, awarded. With that they become the AI champions of the team. And for finance the return of investment is quite clear. It is numbers. And what I see from doing that is that they can quickly because everything is at speed and so many at the same time, they can find one that actually works out of 200 and then that one would then be the champion and that one would then be scaled across. And uh, I will go into the second thought that I have is how that could happen, how that could be scaled. But coming back to construction in comparison, the mindset is oh, I actually prefer to be the second one. I do not want to be the first one in the market to investigate something, to find something new. And that mindset comes from not wanting to waste the mindset is they are more afraid to waste than they're afraid to miss out. So this is an observation that I have from speaking to two different sectors. Now coming to the question of how that can be scaled, there's actually a practical answer to that. For example, when there is an AI agent that seems to work for a specific use case, what I've seen in successful cases is that there's immediately a template library where there are templates which then can be utilized in different use cases. And the teams would then cross, uh, discuss their success and their failures. Because I think right now there are certain AI champions in a team that is pioneering their workflow. But that workflow may be quite niche to what that let's say engineer is working on, maybe as a site report generation. It may not be applicable to the HR departments in the contractor for example, but there could be some of the nodes and the workflows that could be uh, applied there as well. So these companies successfully have kept a template library in house and I see that as a way of scaling, um, now to reach the highest level of scalability in this and the highest level of adoption across the company would be everyone having the ability to build their own AI workflow. And that would be the most scalable when everyone has a certain level of AI literacy.

Speaker A: Okay, so you're not afraid of having like uh, shadow it, so to speak, in the company, that is people creating tools for themselves without any kind of governance.

Speaker B: Love that. So regarding governance, there should be a structure in place. Actually it's a traffic light system. Red, high risk, green, low risk, orange, something fishy. So if something is considered green, and when I say considered, it means there should be a framework and everyone should be notified through this A.I. uh, policy which the company should put in place at the same time as upskilling their staff. And this is the leadership's responsibility as much as the it's responsibility because the leadership is responsible to put in the checkpoints and IT department actually a lot of times do not have the power, uh, to grant access to certain folders. Right. And so with that AI governance triangle or pyramid or framework with the red, the uh, orange and the green. If it's red, then it's a no go. If it's orange, raise it up to higher management. There needs to be a decision made by someone with authority. If it's green, good to go, it's safe. And a lot of times this can be achieved when we're using things like Microsoft Copilot Studio to build an AI agent on their own when their server is contained in the same uh, Azure system in the same Microsoft environment. And that is already agreed from the very beginning when they use Microsoft 365. So yes, that's uh, what I think.

Speaker A: Yeah, sounds good. Uh, another issue besides scaling is the question of data. And I have discussed with several founders and I see two opinions. Some say that uh, we need well, um, organized data for AI to be, for it to be successful. And the others say okay, now the modern AI can manage with any kind of data, whatever format, whatever structure. It doesn't matter anymore. How do you see that?

Speaker B: I think the data structuring in the data lake is quite essential at this moment in time. And I think there are certain companies and softwares which can use AI to structure the data. And I've seen things like that in real estate as well. I would say like real estate use cases for real estate developers, especially for the use case of market analysis. I would say it is about time and cost. Now if the data is structured uh, by humans at this moment and then they implement AI on top, I would say that is a less costly uh, experience. From what I see now, it can change later. But if we use AI to structure the data and depending on the size of course, a lot of tokens can be burned there. So now it is the comparison between the cost of the token burns by using AI to structure data versus the human cost man hours in structuring the data. So it often depends.

Speaker A: M, yeah, but I have seen there was a conference in uh, Helsinki recently, uh, AI in AC and there was somebody from a UK construction company and they said that they used two, two years to organize uh, safety related data within the company. So they had a lot of data but it was not organized. So to make it useful they had to spend a lot of time for that and maybe AI can help in that process uh, in the future. But still I, I, I, I kind of think that if you have good data to begin with, you save a lot of time and money even with AI.

Speaker B: Yes, absolutely. Yeah, absolutely. It's, it's good practice as well and someone in house who knows how to do it.

Speaker A: Um, so one thing that is also discussed a lot is the ROI of AI. What does it actually, uh, provide us that is of real value? Uh, can you share any examples from your clientele that you can share with us about uh, that question? Have you seen success stories in that sense?

Speaker B: Yes, actually I have thought about this uh, quite often in regards to the success stories of the return of investments, um, There are some concrete ones and I also want to share my thoughts about the return of investments concept in general. So the concrete examples first. So this concrete example is one of our partners in Shanghai. When they provided a solution to the state's grid in Shanghai, 80% of the man hours were reduced from the generated BIM model method. And that I think is very concrete. And in a bit perhaps I can share my screen to show that. Okay, so this is a concrete example of a use case that achieved return of investments. And this is a recipe from Shanghai Jiao Tong University. And I say recipe because it is custom made for this use case and it can be customized for something else. And this particular example here is a substation of A, E and M room. Uh, so this is an auto generation of the MEP room. And this is particularly useful because the design doesn't need to be beautiful, it just needs to be compliant to the rules, code compliance. So what we're seeing here is Revit is a user interface and in here we have a plug and play situation where we put in the specifications, we put in the standard codes and other requirements and voila, it automatically generates the room and it has achieved also the BOQs. So you also have all the financial numbers. And if it is M modular integrated construction, it can then actually be sent to the factory and be manufactured in the same stream. And the other one I'd like to show here is a product uh, of a KFC chain auto design. So actually I say product, what I mean is also another recipe customized. And what we're seeing here is the a KFC franchise never needs to be special each time. There is always the same set of FF&E furniture, fixtures and equipments. So therefore what we see here in the logic is we have the brand standards and the codes and the design principles and that's the first input which is the pre trained AI algorithms and we produce the CAD first which is the spatial layout iteration and we can make the human decisions here. And then, then there is the process of inputting the site information of a different franchise location and the specific um, uh, client requirements if there are any. But basically it is the same rule. So you can imagine how much cost and time is saved through this process because the demo here is sped up five times, but it's only sped up five times. So that is actually a very quick workflow to produce a design inside Revit where you have all the information of the BOQs and you can also see the design and walkthrough and renderings. And so this is a massive success. So just like any other franchises that uh, for example McDonald's would, would also benefit from this. But I would like to mention that for example something like Starbucks probably wouldn't want this because in each of the location they will want some character. So there you go. I also want to mention um, this concept of ROI and why sometimes I feel like I should clarify this um, a little bit is that when a conversation goes into ROI and a conversation goes into um, reduction of man hours, I do hear a lot of uh, people start to feel a bit uncomfortable about that and I can imagine why. Um, but I would like to say that if a company has good culture and if a company has good direction and good vision, they would know to repurpose the people. And the return of investment then needs to be calculated when we repurpose those staff to something else. And what is the new business value generated? Now I have to admit I don't have clear metrics of that. Um, but I do know a good case recently that Ikea uh, have shown that they have repurposed uh, their staff instead of letting them go when they implement AI. And last but not least about this, just uh, a string of thoughts from me. The pre AI metrics, pre AI performance metrics on human staff should not be used in post AI performance metrics. We should not use the same actually to say uh, how you perform pre AI should be used, um, the same evaluation should be used after. So I think there are lots of little uh, parameters there as well like how much are people burning their tokens should not be a metric of how well they're using AI. It's about the quality of the output etc.

Speaker A: So yes, yeah, yeah that's a good, very good point you're making. Um, uh, by the way you, you mentioned Chinese startups and M. We know that uh, the US and China are leading M A AI ah development and, and in LMS and robotics. I would say that China is even ahead of Western countries in some areas. Um, so do you see any differences in, in how they, the Chinese startups and companies approach AI in aec? Mhm.

Speaker B: Yes, uh, I can say a bit about that. I would say they are very aggressive and they have been starting out a few years now and I think what needs to be mentioned as well is the economy, the economics of the construction tech markets, um, for example certain features that we think is worth becoming a product feature or selling it, uh, pricing, it is actually given out for free in China a because the tech is Quite advanced. And they have been doing R and D on this uh, for a few years now. And the government is really supportive of technology. They're giving funding, um, lots and lots of uh, government funding and uh, initiatives to back up the technology. Um, the thing is the tech is very advanced but the market is saturated. Which means for example something like uh, drawing compliance or uh, writing tenders for you or reviewing tenders, this kind of AI feature, they come as a add on feature to something else. Maybe that something else is for um, example uh, an AI that helps with the quantity surveyors or an AI the house with uh, program planning. So those are the more concrete uh, AI services. But then the add on would be like oh yes, this AI can also help you write your tenders by the way. But then I feel like in the west, um, or in America perhaps or even in Europe, this type of feature is a product on its own. But in China first of all, everything is unified in one big app or a few big apps and then you have the big players that kind of takes over. There's also a lot of open source out there as well. And I think another thing to mention is the data, uh, how there's massive uh, amounts of structured data inside China also because of the unified uh, way of uh, how. What is the best way to put, put it. So everyone has trust in the system and so the data is actually uh, able to be taken up very quickly. And this is also I feel another reason why the speed is so advanced, um, the tech is so advanced and also at speed. It's just the market is quite fragmented at the moment I would say. And I think that's why a lot of Chinese startups are actually hoping to go into um, other markets in the West. Um, and I know there are some regions where uh, currently the attitudes may be a little bit awkward but in general I could feel the, the construction and prop tech industries, they really want to go abroad with their very, very competitive technologies.

Speaker A: Yeah, that sounds fascinating. Uh, I'd love to talk, talk with these people as well in the future. Uh, speaking of, of the future, how do you see the future of AI in ac?

Speaker B: Wow, that's really good question. I think right now in terms of AI adoption and the rate of how AI is being adopted, it is much slower than the technological advancements. I think uh, both you and me would probably see the AI adoption in the strategy level, operations level and the technological level. And for me I see the tech is way ahead of the other two. The strategy level is what the leaders need to get uh, it clear in their minds the operations level is really the bottleneck. Uh, what I mean is I think it is the change in the way people work that is going to take quite a bit of time. So I would say change is definitely happening. Um, and I would say in, I can see in five years time, but I may not be able to see, uh, further. But I would say in five years time, uh, there will be a lot of agentic workflows. It would be agents working with another agent working with another agent. And then I think what humans need to do is to avoid the intent drift where when one agent is only 99% accurate and then they pass to the next agent who is also 99% accurate, you get like a compound drift away from what you actually want the agents to do. Um, but coming back to the future of AI, I think the humans are, in our profession, we are really AI and aec. The AEC professionals will really be there to curate, will really be there to guide the agents to do work. So I see that, but I feel it will not come as quickly as uh, we envision it to be like we think it would be. I probably think it will come a little slower. But the tech is way advanced.

Speaker A: Yeah, yeah, yeah, yeah, you're absolutely right about that. Yeah. So, um, but yeah, nobody can predict the future regarding AI it seems. But yeah, but uh, it's, it's all about the people how, how they uh, want what they want to do eventually and use the technology.

Speaker B: Um, exactly.

Speaker A: So if our audience wants to learn more and get in touch with you, what's the best way to do that?

Speaker B: They, they can add me on LinkedIn and they can search my full name and which is Adeline Chan Hockman. But if you search Adeline Chan and then you type in Aal, it will probably come up and they can also find me through your podcast and uh, they would be able to find my website.

Speaker A: Very good. Thanks a lot for this opportunity. It was really fascinating. And let's talk again soon.

Speaker B: Thank you so much, Arnie. Um, I'm. This was very fun. Yeah, let's talk again soon. Thanks for listening. Subscribe to this podcast and visit aec-business.com the award winning blog for more news and stories.