150. Predicting Buyer Behaviour With AI With Will Zhang

This episode of Professional Builder Secrets is brought to you by Apparatus Contractor Services. Apparatus provides full scope, end to end QuickBooks, Precision Construction accounting and CFO services for residential builders. If you feel like you can't completely trust your weekly and monthly financial reports or you feel like you're flying blind, Apparatus Route will rebuild and operate your QuickBooks accounting system and give you precision construction accounting capability you can trust. Learn more@apparatusteam.com or on the APB Rewards Partner Directory.

AI in my estimation still really struggle to make any judgment or and that's where like why human guidance and control is super important.

Welcome to Professional Builder Secrets, the podcast for building company owners wanting to grow safely and securely. I'm your host, Will Blunt and today I'm joined by Will Zhang from Openhouse AI. Open House uses AI and machine learning to predict the behavior of home buyers and qualify leads for builders. Welcome to the show, Will.

Thank you for having me. It's great to be here.

I'm not sure I got that spiel anywhere near accurate, but I'm really excited to hear about everything Open House AI is doing and your knowledge about AI as well.

Cool.

Well, let's start with your background, Will. What's brought you to where you are today and everything you're doing in the. In the AI space in construction.

Oh yeah, for sure. So my background has been working in machine learning. Of course the current buzzword and more technical term for AI. I personally think machine learning is better describe what it's actually doing as using machine to learn pattern. But. But it's more technical term. And my background has been deploying AI technology for big corporation tradition in the past because if you think about in the past technology and compute is very expensive. So I cover like experience from utility. I work in Europe for six years and help ship the first iPhone across six countries in Europe, our hedge fund and across different industries in oil and gas. And eventually in roughly in 2018 and 19, the compute and infrastructure become increasingly accessible. And also my experience in my career thinking it's time to bring technology to a more meaningful industry. That's housing. Because actually I think housing providing shelter is something that technology can make a meaningful impact to a lot of people's lives. Yeah. So that the reason why we decided to start and bring everything we know to the housing industry and we decided to start with home building first. So yeah, that's why we're here.

Cool. For people out there that don't understand, including myself, what's the difference between artificial intelligence or AI and machine learning? How Would you explain that?

So artificial intelligence certainly the easier way to understand it at a broad term. Yeah, you have intelligence that's artificial that built by human. But machine learning is actually really the reason success. If you think about AI in a broader term, it cover a lot broader than in cognitive science behavior and the robotic. But machine learning is specific about using compute to actually understand and extract pattern and then help people make decision and automate those decision. If you think about even the self dry car. Right. So that Tesla is working on is basically can I use the same sensor and information to simulate what how human drive car? And so that's basically the technical term. There's a broader term in AI, but I think it's just too technical. AI is a good shortcut for people to get understanding what it is too.

Okay, okay. So I guess my understanding and maybe I've misinterpreted it a little bit myself, but the machine learning is pulling together all of this data and learning from that data to help people make decisions. The output of that data is the artificial intelligence. Is that kind of right?

Kind of. Now it appear to be intelligence. To be more precise. It's even today the most advanced model. The thing that people gaining a lot of recent interest is that in the large language model, when you ask the algorithm something, it will output the text that really sound very intelligent. But still the industry working towards reasoning capability, planning capability and optimization. Those are the level that actually as a practitioner like myself we're working towards. But the cool thing about the breakthrough is that now we actually seeing a glimpse of potential emergent behavior that can take us next step forward. So that's actually really exciting for the field as well. Yeah. So first of all it's easy to get started think about intelligence similar to artificial intelligence, similar to human intelligence. But the most common misconception actually stems from there is using human intelligence as an analogy to think about it. Now if I just quickly unpack one of the reason why people fear about it. There's a lot of mixed reason why they're fearful about it. For one is that well, what if we lose control about the AI and what's going to happen? So the common fear is like the Skynet in the movie, right? Yeah, interestingly. Well, if you think about that the intelligence in biology evolution for human is that intelligent means dominance almost. They are synonymous a lot of time. But artificial intelligence is how you build it. Intelligence does not necessarily mean dominance. So that's one of the bigger fear about. You know, the doomsday of AI will bring about to losing control. And now there are a lot of unintended misuse. That is actually real. That's actually what AI risk that people talk about. That's another side of fear is that what if it's actually put in the hand of a hand of bad intended people or unintended consequence. So that's actually a lot of people working on that. Now the other side of fear which you also mentioned is that well now people fearful about if I don't get onto developing AI then I will get out compete. I think the reality is more about how do you get into using AI Now I think that's actually most realistic. My personal most realistic forecast into the future is more about it is real that we should all become user of AI because actually for you for reason one you actually shaping how technology is empowering people. Second is over time when people know how to use those system they the one who use it has substantial competitive advantage. And also we can learn from history. If you think about in let's say 1960s, how many people know how to use a computer? Only probably Department of Defense have a handful of programmer know how to use this mainframe computer. If you don't know how to use a computer in 1960, you probably still can find a job no problem. But if you think about today, if you say someone who have no idea how to use computer. It's a lot more difficult to find a high paying job that require all those modern process. In a typical B2B context. Now we kind of with a new generation like they're kind of born with. I don't need to learn how to use computer. But I actually kind of like I've seen in from 90 to 2000 like the transition of learning how to use computer to learning how to use Internet to the change of behavior how using Yahoo to think about how to search keyword in Google. All these learning curve. I think it's actually meaningful for people to get using it. That's pretty much what I see is in the future everyone will be using some kind of AI tool to empower them.

That makes sense. It just seems to be moving a much faster speed than those other examples that you gave.

Absolutely. Yeah. So the speed is actually what causing a lot of people's fear as well. I think it's real. But the problem with think about is more about AI as a technology is very unique compared to Internet and and computer. The two recent technology transformation is that AI is a technology adapt to human unlike computer. If you think about people need to learn how to use computer and I still remember using dial up modem. I dialed up and then I load a picture and then I turn off to save money. A lot of things need to learn to how to optimize. Using technology versus AI is interestingly is a one technology transformation that it's actually trying to adapt to our life. So I think it's actually more about using it now. The key is that the outcomes most likely is related to something we call Pareto Tin's principle. Many people might have heard of 20 people actually make 80% of money or 20 people do to 80% of work. What could happen with AI unlocking those constraints become 10% people doing 90% of work. So the competition most likely people are using and leveraging AI as their competitive advantage will actually gain a substantial competitive advantage over those who doesn't use it. It's not really the technology itself, it's the people who equip the technology become much more competitive and that's a real impact to the economy.

Interesting. What are we calling that one? The zhang principle, the 1090?

I would love to be able to do that, but I'm not a mathematician and don't believe I can.

Well, it makes a lot of sense though. It does make a lot of sense. You're talking about how using AI is the best way to keep up with the trends and make sure you're staying competitive. So open my mind about how builders in the residential construction industry are using AI or can use AI. I mean for the most part everyone's using ChatGPT to run an email, write a blog article or something similar. But expand my mind about what else there is out there.

Absolutely. So we had started our journey working with production home builder because they have the more volume and data and then open our mind about what AI can do on the construction industry. So a lot of our experience now is talking more immediately impacting the production home builder. But I can also offer a glimpse of what the future will like for even a smaller home builder in the future. So immediate in immediate term already it's actually happening is that if you are a builder have substantial amount volume of web traffic to your website and and have a larger volume of home that you're building, the AI can already extract all these behavior pattern to understand in much more nuanced way the behavior pattern of those high intent buyer versus the tire kicker. Now website actually offer a great tool. It's basically a 24, 7 model home or show home on the website that anyone can come visit it now then you can Have a lot of tie kicker who are having having aspiration. Okay, what if I build a new home? How would that look like? Now those are great because you're getting your branding your presence on the Internet. But when it comes to actually selling, their behavior pattern is very different compared to those who are serious buyer. And in the past, if you work with a marketing agency, they will give you more generalized metrics about how many web traffic, how many conversion, how many click. What are the bounce radars classic metrics. In the new era, using AI, you are now able to decode them to separate them from. Well, these are the tie kicker traffic. But those are meaningful traffic that most likely will convert into sales. And with that we are able to say out of these market condition, what are the short term prediction in 90 day into the future, how much sales you're going to have? What is the impact of interest rate? How does it reflect it to specific you as a business? How how many volume of sale you will have? Why this is important because many builder will say, I listen to economists tell me the interest rate changes. How would that affect my market demand? Great, but that's more at the national or provincial level or state level. But when it comes to specific my market, how does it affect me? Because say an increase in interest rate? Well, now housing is not as affordable, but it affect different market differently. Because sometimes we push some other market by buyer into a previously colder market. So the dynamic of that is really what impacting the in real term the business operation of a home builder. Now those pattern is basically best to extract that by deploying AI to understand exactly inside your sales funnel. From the beginning of the web traffic, the first click of ads to the sales. Who are the high intent buyer, who are the low intent buyer? How do they behave differently? Therefore we can analyze within your funnel how much how many of those will be selling in the next 90 days. Now one layer deeper. Once you know how much, how much home you were looking to sell, what are actually people looking for? Because a lot of time like what will be selling, what are reason why they're looking into that. That's where AI can extract a lot of behavior pattern in terms of understanding why they're looking to move and what kind of solution of the new home that can solve their problem. And knowing that really give the builder who knows that a competitive advantage to help them to relieve the concern of moving or the fear cost to basically get them over to help them accelerate the sales cycle. So that's one use that actually have been being deployed and used by many production builder in the US already. So that's basically one case.

So you're utilizing the data which is generated by the company as in the traffic to their website. Are you pairing that with industry wide data at all as well or is it singular to the company?

So we use both. So high level how it works is that we have the fortune of working for already a few years to collect a behavior pattern how people make decision. And that give us a massive database of 4 to 5 million consumer behavior profile. Now that database is what we use to teach the AI to become intelligent. Now then we deploy for each home builder to learn about. Okay. For this home builder how the pattern looks like. I'll give an analogy to help understanding is that you might have a buyer who every night at 10pm start browsing pictures on your online. Right. That's probably my wife getting some inspiration to look at. Well, what would be a nice kitchen looks like, right?

Yeah.

And then. But one day actually this kind of couple years ago when we moved like one day we say well actually for a reason for in my case I need a garage. I want a garage to store my kayak. And then now we talking, now we're talking buying. Now all of a sudden the behavior pattern will change from just randomly browsing picture to okay, what about pricing? What are the location? Now you're starting to look at different things. And those behavior patterns shift is actually signal to knowing that buyer move from one pattern to the other. So we have roughly 4 million of these we call decision fingerprint to know the change of behavior pattern indicating they're moving deeper into the funnel and becoming increasingly high intent buyer. So that's what we use to teach models. So that's our database. But once you teach that we also know every single market for every builder the way that they attract buyer the way that the target market that they actually try to attract cause them to have different traffic go to their website. And that is what the algorithm need to learn from that builder about. Okay. Out of this market for this builder with this specific strategy. Let's say a builder who try to attract first time home buyer the buyer patterns look very different from a 55 + community who who are buying a retiring home. And that's actually where they will learn specific about that builder to understand for that group of target market how it does it look like. So that's actually how we are using AI to inform them how many home you will sell and what are they looking for specifically.

Interesting. So how much data does an individual Building company need to make it valid at this stage.

Now we are still in the early stage of AI technology evolution. @ this stage we still need quite a substantial amount of data to be able to do that. And we're talking about at least like probably 6,000, 10,000 unique session on a monthly.

Traffic.

Okay. We're talking about probably 200 home, 200 or 300 home per year in one market. Then we can actually have meaningful pattern extracted to it. But I believe as we making the AI system more and more intelligent, you think about a sales agent, the sales agent doesn't need to do 300 home sales to be a good effective sales.

Agent. So yeah, that's true. Do you think that the more data and more knowledge that the AI generates that you'll be able to apply that to smaller custom home.

Builders? Absolutely. It will take some time, but I absolutely see. Well, actually by the way, when we think about taking sometimes 18 months cycle, we think it's actually very likely that in 18 months or 24 months these kind of technology can be propagated to a smaller builder. We actually run a pilot project now internally is in one of the major cities in North America. We already have 80% of web traffic thanks to the customer base we have within a specific market. We are able to now make the prediction for the entire market because like if you get builder for who focus on first time home buyer, you have builder who focus on cross segment, you have builder who focus on 55 plus then then we can actually understanding the common pattern because these buyers tend to go to different website and shop around and that actually allow us to make that kind of prediction for smaller home builder in that market. But it basically needs more learning before we can generalize it. At this point it's a bit early, but hopefully in 18 to 24 months we're able to trial that with a smaller builder as.

Well. It'll be interesting to see how it plays out because I think like a person building a custom home, they probably have a much longer sales cycle. They're doing a lot more research around the type of of products they want in their home, the fixtures, the layout or that kind of stuff. Whereas someone building with a larger, you know, a spec builder, it's a bit more out of the box and they don't have to go through that same.

Journey. Yeah, it is now what's common though is actually in what we're now going into is starting to continue increase our high intent funnel to become to offer the communication capability. So initially we basically deploy an app on the website to give everybody an idea. Then you click through a few quiz. Now through those quiz we actually learning from the behavior pattern present the home that most fitted for this quiz using generate AI content to test it out. Actually did I hit or miss my recommendation to that buyer? And then we draw them down further down in the conversation. That's where we already at 12 months ago. Now most recently we recently deploy an app. Now once you say I'm interested, we actually have an app for the sales agent to start immediate texting, email communication or phone call. Now as we increasingly building this capability to understand the conversation, having meaningful conversation between two human, the AI are able to extract the decision pattern. And once we can totally see this kind of tool can be deployed for custom home builder further down in those conversations understand what's the most effective way to ask meaningful questions to identify the true need of a.

Buyer. Okay, so you're essentially asking or communicating the questions based on where they are in that.

Journey. Yeah. So and then also we can learn from the top sales agent like how do I ask question to drive the conversation and understand the buyer's need. Because I believe we have some knowledge on these custom homes selling cycle. A lot of time is like the home buyer might not necessarily saying this is the home. Exactly. I have perfect pixel perfect picture of what my future home looks like. A lot of time is a conversation between the architect and the buyer and the construction manager to know what's feasible, what's best used. Right. So all those things a lot of work is actually just going to understand the true need of the buyer. And if we actually go getting more and more meaningful conversation, I believe AI can basically learn about it and be a effective assistance to extract all those meaningful talking points to help the sales agent and architect to think about. Well, if I the express problem is A, but the deeper why is B. If I actually have a solution, rather than go offer a solution for the surface expressed problem, we can actually have a better solution, maybe even lower cost in terms of building it to actually solve the deeper need. That's where we see AI will go cool.

Cool. So in terms of that technology, what's the output the builder is getting and what are they seeing? Is it more so from an experience communication point of view or are they getting an output which is saying this Many customers need this amount of information before they're ready to purchase a.

Home. Specific for Open House, I was actually more generally talking about the capability specifically open House. We solved three problems for production home builders and I do see Two of them potentially applicable for smaller customers builder as well. Problem number one is the market is cyclical. And then we predict the market based on all these behavior pattern so that the builder don't have to be reactionary to the market. So when the market is actually changing it becoming more hot. For example, if a market become hot market from a cold market, then you can consider charging for example changing pricing or charging a higher price. Then you can protect your cost erosion. When the market is hot and everyone looking for same trade or same.

Frame. When you say a market, are you talking about a tight geographic region or how big are we talking here? Like a.

City. A city. A city or. Yes, typically it's a city level. A metropolitan city. Yeah, because that's.

It. And how far into the future are you predicting this buying.

Behavior? We're looking into 90 days. Right now we confident predicting 90 day. We're working towards extending the time window. Okay, but now if you see what's come ahead in the next 90 days then you can say well when the market changing the 90 day foresight is actually very powerful. Because now if the market has been steady, you see uptake, then you can hold our inventory a bit longer. Right. You have a carrying cost. We understand a lot of builder will say well I want to unload them as fast as possible. But if the market is uptake, having an uptake, you might want to hold out a bit more. Or if you're doing pre sell then you might face cause erosion later on if everyone competing on same trade. Now why don't you charge higher to protect that? Now when the markets cool down, sometimes you look at a funnel. Well, my funnel is solid, but I might want to secure a few deal ahead of time so that I can keep my volume going. So these are the problem we're solving using AI to help them. That's one area. And this one area is actually more related sales. Like I mentioned a lot of technology out there already to basically creating personalized experience. But specifically the output is that when a lead is converted the AI actually already learn about why are you all the basic question about your budget, your timeline, what home are you most interested in? All those shallow level of the question have been answered and then delivered right inside the app to show this is what will is looking for. Then allow the sales agent to ask deeper question. For example like instead of I'm not a good salesperson, but like I was if I will as a average salesperson you walk in oh, what's your budget? That's probably a very Bad question. But I no longer need to ask that because have been answered by the AI because actually when the customer fill the quiz we actually know pretty good about your budget. But now I would ask deeper question. What brought you here? What make you interested in buying a new home? Now now you can more focus on deeper why question. Because all those shallow level qualification question and answers already in front of your face right in your app, you have it.

Accessible. Yeah. That's really powerful because there are so many builders that waste so much time on what you call tire kickers. They've got to answer all those really basic questions and then they get five questions in and you realize that that's not even the right. You know, not a prospect for you, for your company. So you've got to then move them on. But you've already spent 10 minutes with them and then you feel rude if you don't spend another 10 minutes with them as well. So it's a whole bunch of wasted time. You know what I.

Mean? That's what we see is basically supercharging the current CRM. But basically we put a skin on top of the existing CRM to basically analyze and extract those patterns and engage the buyer using generative.

AI. So predicting buying behavior, qualifying leads. You said there was a third thing that Openhouse AI is.

Doing. The third thing is more a production builder case is basically called evenflow. There's a holy grail of evenflow construction for production home builder. Now a lot of builders take pride in achieving even flow but it's very difficult to maintain it because you have the uncertainty around construction and market demand volatility. So we use algorithm not by knowing the prediction and the consumer preferences. What would be for example like best combination with structural option and upgrade to allow you to simplify your construction process while hitting the most relevant market segment that you are trying to target. But now offering certainty and predictability in your construction velocity. And that help the cash flow because like you have certainty in possession, you basically have certainty and stable cash flow. So that's more a production.

Builder. Yeah, yeah. I mean the first two solutions they sound extremely valuable for any building company. I think especially predicting buying behavior. Everyone wants to understand that. From what I understand that's a customized solution you offer for your clients for the larger scale.

Builders. Yeah. Do.

You. Do you plan on having something more accessible for a larger audience where they could maybe buy something out of the box? For example a 90 day. That 90 day behavior predictions and things like.

That? Absolutely. It is in our long term mission that we actually want to do that. This actually ties into deeply to our purpose. Right. So the reason why we started out this technology in the beginning is about how do I take technology from all these sophisticated industries like hedge fund financial and telecom to housing industry. But our long term ambition is how do we keep learning and make IT technology increasingly accessible so that all the other people can also use it. It is absolutely on a roadmap. Now the question you might ask is okay, why do we do it? Now the interesting thing is about technology, the simpler it is to use means it's actually encapsulating more complexity inside. So we are still working through that. If you think about an iPhone today, it's just a one screen. You can touch it, you can do whatever with it. It's super easy to use. But think about the complexity that Apple and Steve Jobs needed to build to encapsulate all those complexity inside a little computer. It's certainly within our mission to do it. We're working very hard to make it happen. We're very excited about our future, but it is certainly in our ambition to do.

That. Yeah, I mean that first computer that was ever built was probably the size of my apartment. Right. And to think that there's such a small device with more power than that. Even that first computer, that original computer even had, that's the journey, that's the journey it goes.

On. Right. So yeah, hopefully I'm sure someone will do it and hopefully we will be able meaningfully contribute to part of.

It. Yeah, it's crazy. Well, it's very exciting and you're extremely knowledgeable about everything. I found it super interesting just having a chat to you. So I'm sure all the builders out there are excited about getting their hands on some of that data and certainly the buying behavior and the forecast sounds super interesting. But where do you see any limitations with AI at the moment and what's happening, what's evolving over the next five to ten.

Years? So the current limitation of AI is the reasoning capability. So if I there's kind of like people talk about five level intelligence but I will just simplify just three for now. One is the layer, one is what currently we already achieve is generative, generative AI and the predictive capability that using that tool essentially reading pattern from human activity and simulate and imitate what have been done. So if you think about all these ChatGPT amazing capability, it's trying to create the text that fitting your expectation. So that's level one. But the cool thing about what we learned is that when in the journey to create that force the language model to establish some real world understanding of the phenomenon through text. Now we are incorporating more they call multimodality model meaning not just text but vision sound. Now you actually start to incorporate more and more into those model that allow us to hopefully unlock the second layers reasoning. So you now can ask a question. For example, you can do a simple test. For example, if I say hey, I put five cube of ice into a fire stove and then I put two more cubes inside a human, you know, then I ask how many, how many cube of ice is actually left for as we all know, well, you just put in a fire stove or fire pit, there's no ice left. But the language model will say well you put three and then two. You still have five block of ice cube because it doesn't understand fire. Right. So there's no reasoning capability. But there's actually when you actually prom it to say well wait a minute, I put ice cube into a five pit. How come the ice is not melt? And then language model pick up say oh actually all the ice will have melted. So that's actually step by step reasoning, we call it tree of thought is what the researcher working on to increase and really achieving reasoning capability. And then that's kind of layer two. Once we are able to using that generative AI capability to achieve reasoning and multi agent then now you can basically ask, hey, you specialize in this task for Agent 1 agent who is special in that task. Then we can go into organization optimization, meaning multiple agent can optimize for a coordinated effort to achieve certain tasks for a business environment. So that's actually where we're most excited about it becoming very realistic for us to achieve that. But AI in my estimation still really struggle to make any judgment. And that's where human guidance and control is super important. And then that's why we actually see the most successful combination is human in AI Empower human decision.

Making. Yeah, yeah. That's crazy to think that you're talking about essentially AI working with AI to essentially businesses will be running primarily just with AI agents doing everything it looks like wild.

Today. But if you actually, if you it's good. A lot of lesson can be learned from history. If you go back in 20 years or 30 years ago, you tell people that well actually it's not only you using computer, but computer working with computer and talking to a computer. The people in the 90 will say well that's crazy. Actually every computer on the Internet is talking to each other to achieve A lot of tasks that we don't need to worry about today. It's already happening.

Yeah. So yeah, it's crazy. It's blowing my mind. Well, I appreciate you sharing your knowledge today. Will, what advice would you give to builders if they're hesitant to test out new technologies like this and learn about where everything's.

Going? I think it's twofold. One is the AI is a tool. It's best deployed when you as a business owner has clarity in a business priority. So just because AI is a very cool tool like ChatGPT, you can use it, basically, does it impact your bottom line business? Does it actually in alignment with what you want to achieve in terms of business objective and strategy, does it how you're going to grow your business? I think there's two sense of it. One is if you're interested, I would encourage people to keep using it so you can see the change of technology as it come, as it unfold in the next 18 to 24 months. So keep yourself in the loop as a user. On the other hand, when you actually come to actually deploying for your business objective, business context is make sure it's always aligned with your business priority. Now, what AI can potentially unlock, is that a lot of previous constraint that you might think, well, I might need a lot of manpower, a lot of resources to achieve that is where AI could potentially challenge that. Those constraints might no longer be the case thanks to AI. So that's, I think is the best way to think about.

It. Awesome. Well, Will, thank you so much for coming on Professional Builder Secrets today. I've had a great time having a chat to you and learning about everything that open house AI is doing and all your knowledge around just machine learning and AI as well. I'm excited to see where you take things in the future. Hopefully it's going to have a big impact on the custom home building industry. Where's the best place for people to go next to find out more? Open House AI, is that.

Right? Yes, that's the place to go.

Yeah. Yeah, perfect. Well, I highly recommend everyone checking it out and following Open House's journey. Will, awesome to have you on the show today and thanks once.

Again. It's a great pleasure. Thanks a.

Lot. Also, a big thank you to our listeners wherever you are in the world. If you like the show today, please subscribe to Professional Builder Secrets on your platform of choice. And if you're feeling generous, leave us a review. But until next time, have a great day. This episode of Professional Builder Secrets is brought to you by Apparatus Contractor Services. Apparatus provides full scope end to end QuickBooks, Precision Construction accounting and CFO services for residential builders. If you feel like you can't completely trust your weekly and monthly financial reports or you feel like you're flying blind, Apparatus Route will rebuild and operate your QuickBooks accounting system and give you precision construction accounting capability you can trust. Learn more@apparatus team.com or on the APB Rewards Partner Directory.