Luke Metcalfe on Where AI Stops Being Useful in Property

In this episode, we explore the illusion of precision in property data and ask whether the rise of AI is making things clearer or more confusing. Consumers are told to trust the data, yet much of what we see is simplified, biased or even fabricated.

Meanwhile, AI tools are emerging that can sound deeply convincing even when they're wrong and many buyers lack the knowledge to tell the difference. If artificial intelligence depends on big data sets, how useful can it really be at the micro level where property value is won or lost?

To unpack this, we're joined by Luke Metcalf, founder of MicroVerbs and Datascape, and one of Australia's most respected data scientists in the property space. Welcome to the elephant in the room. This is the podcast where we love to talk about the big things in property that never usually get talked about.

I'm Veronica Morgan, real estate agent, buyer's agent and buyer's agent mentor, co-host of Foxtel's Location, Location, Location Australia, author of Auction Ready and co-host of Your First Home Buyer Guide.

Hi, I'm Chris Bates, ex-financial planner and mortgage broker, currently ranked number three in the annual MPA Top 100 Mortgage Broker Awards. Before we get started, everything we talk about today is not personal advice, and we recommend you engage the services of a licensed and experienced professional.

Our guest today is Luke Metcalfe, founder of Microberbs and Datascape. Luke has spent many years mapping the hidden demographics, social and economic signals that shape neighborhood performance and now sits at the frontier of how AI and big data intersect with human decision-making in real estate.

Now, in fact, we first interviewed Luke about this very topic back in 2018. So that was way before AI was even a thing. So we are very happy to be talking to you again today. Luke, we have met you a few times here.

Yes, nice to still be here and great to be rejoined by you guys in this new scary era that we exist in.

Yeah, I mean, I think like most people in the country, I feel like I'm really late to this AI. I've kind of had the shiny object syndrome on tech and I'm like, hang on a sec, this stuff's really quite interesting and it's coming really fast and we're out of our debts.

But I mean, obviously people are using language models pretty much every day. I mean, a lot of people are. And now they're trying to apply that to everything in life and property decisions are a big part of it, but it's often very convincing, but often sometimes really wrong as well.

So, you know, how should people sort of approach, I guess, AI and, you know, using it for sort of property data and what sort of, what can we do to stop getting, I guess, duped by it?

Great question. So yes, this is very 2025 question. So yeah, I'll disclose. I talk to AIs all the time. So I'm someone who wakes up in the morning and says, I've got rhubarb and artichoke in the fridge. Can I make a puree out of this? So absolutely. I hear you and I understand.

I know that AIs can become, if not your friend, then like just really valued companion that can advise you and talk to you. And it knows how you think if you talk to it enough and if you tell it about how you think and what you're worried about.

And in the property game, it's filled with anxieties. It's a very stressful period in people's lives buying a property, the biggest financial decisions that they make, hopefully many of them. So it's very normal and expected that people would turn to AIs and they can bring together lots of different information.

So you can assemble a

Remember that with AIs and machine learning in general, you always have to ask, what's it trained on? So what's its actual goal? So the goals of the AI are one, to predict the next word. What's the most likely next word to come? And then two, can they keep you engaged?

So can they keep you hooked? They're not trying to predict the property market. So it's not like Sam Altman built ChatGPT to become really good at predicting future outcomes. That would be a very different kind of AI.

And so if you go and ask it whether to buy in Rockhampton, it'll take in often very promotional texts because the real estate industry tends to trend positive.

And yeah, maybe this is an early Dumbo, but we actually tried putting a GPT version on microburbs for a little bit as a trial, and we found it was competing with us. So it was trying to give its own views on what was going to grow up. Should I buy in Rockhampton?

And then the chat GPT is thinking, well, it seems like you're into Rockhampton. I'll give you a good spiel on why it is. And it's like, this is not our philosophy. Let's take this thing down. So yeah, absolutely. It's not something that's going to tell you where the market is going to move.

And there are limitations to what degree it can understand as what you guys get into the investor psychology. And it's not the same as Chris having spoken to lots of people doing mortgage broking in the past week. which leads on to another thing of just like they're out of date.

So they're not trained on the most recent stuff that's happening in the industry that you guys are getting from surveys and other stuff. They're trained on data from many, many years ago, as well as somewhat recent, but they're going to weight it all together into this big soup.

And another thing is that they're trained on consensus. So in picking the next word, they're, They're thinking in their heads, okay, there's this word that's a bit more, let's say, left wing. That word's a bit more right wing. Let's go for the centrist one.

So they're trying to coalesce on like, the psychology of it is it's trying to coalesce on like safe kinds of things. And also they're trained to be kind of safe. So that's not the same as the mindset of a good investor, which is often trying to go against the herd.

So yeah, you're getting a lot of herd mentality in there, but it's not really that, yeah, totally useless for forecasting.

It's fantastic. I think to myself, you know, my interactions with THBT is not the only AI I've played with, but it's the one I've played with the most, I guess, because you can sort of store in there things about yourself.

So you can make it refer to those, to your own bio, to your own writings and things and say, look, you know, based on what you know about me, what would you say about this or whatever? And I find when I sort of challenge it, like...

you know, it sort of comes back and it's, oh, that's a great idea. Great pivot. Like it sort of, it's really, it never gets offended if you say that's stupid, I want something better. But what you're saying, it really is designed to predict the next word.

So in a way you're sort of predicting, it's like crowdfunding the next word in a way. It's like going with the hordes, but it's also predicting what you as the operator want to hear next. Would you say that that's part of it?

Yeah, that's right. So at the first base model, it's just continue whatever text is there. So if you start writing a glowing report about Rockhampton, or even if you ask, is Rockhampton the place to buy? It'll just like, okay, I've seen these documents. This is the kind of thing, bring that in.

And then they have a whole lot of trainers who go through and test that they're not going to, you know, all these sensitive things that might get you blocked on YouTube if I mention, but yeah, all these sensitive things that they don't want directions. They want people to be safe.

They don't want to cause offense. So there's that level that's called reinforcement learning. And then the final one is engagement. If you think more about like Facebook and Instagram. Yeah. So what keeps you on the platform longer, particularly chat GPT, which is the consumer facing the much more consumer oriented model.

So it really cares about what is going to keep you there. like a friend or a psychologist is trying to get you to open up and is learning increasingly about all of your worries and all of your concerns. And people have loads of them with properties.

And so a challenge for a data provider like us is that people say, we're really worried about this thing. And we're like, well, we don't really see in the data that that thing's important, but it's hard to get that out of people's heads.

But the chatbot will happily talk about that and happily speculate for you and build arguments that it says are plausible when they might actually...

not be which is really scary actually because it is very plausible and I know even when the chatbot comes back to me and tells me what a great idea I'm actually chuffed and I'm no better like you know and I'm also a real cynic when people congratulate me you know like and I'm falling for it so I can imagine it's very easy for people to get lured into that sort of that set and then it makes suggestions at the end would you like me to do this it's so helpful

Yeah, that's a great idea. Yes. And so, yeah, I guess you are getting sucked in there for longer and longer. And it sounds intelligent. And I keep hearing stories about people who are just plugging their questions into ChatGPT, you know, Copilot, Gemini, whatever, and actually acting on the information, which is actually alarming.

Yeah, absolutely. So it could be the most cogent, the most full answer to all of your anxieties that you're ever going to get. Yeah. And a professional, not just bias agents, but anyone, any professional that's just met you, they're not going to know all that stuff. So they're at a disadvantage.

Yeah, I mean, it's happening in the broking world. I think you've highlighted a really key flaw. I didn't actually realize this till you said it. Like the purpose of the language model is to, like you said, so anything, if you're not using it for that purpose, it's kind of redundant, right?

So if you're using it for forecasting, is there going to be an AI tool that's really good at conversational type communication? Because I think that's what people like. It Stability of 24-7, no matter what I want, I can ask it a question. But it's also good at forecasting.

Does that when something like in the property space is actually good and you should consider using it? But yeah, because people do it in the broking world. They'll take our strategy and they'll probably do the strategy big team and say, hey, you didn't do this. You didn't ask this.

It's like those questions are irrelevant. Like it's sending you down a wrong path. It's not what you should be focused on. You should be focused on a buffer and, you know, maximizing your tax advantage is not your lowest repayment. So how do we think about this?

Yeah, it's really interesting. So I listened to you guys and often there's like, wow, you guys are delving into the psychology. It's like where economics meets like how the buyers are thinking and how the mortgage seekers are thinking now.

And so there is a scope for that kind of knowledge to be fed into a model if it's now given the goal of predicting actual capital growth. So like there's a lot of things you guys say and I was like, wow, I wish I could ask people that. That's a good angle.

I just don't have it in my data. But if all these different wise agents or mortgage brokers saying this thing right now, and they're, I don't know, you create a collective of all the texts that's going in or just this podcast.

So you could actually be feeding this in to a model that's trying, you guys have been around, of course, yeah, this 2018. So you've actually got a forecasting ability to do forecasting. Who of your guests knew what they were talking about? How are all the models?

So one thing you can do is you have these like represent team of experts behind the scenes. So the model's broken up into bits. So one of those experts could be Veronica Morgan. Another one could be Chris Bates.

So just think the Chris Bates way of doing things, of thinking about this problem, does that yield insight?

I have heard, it's not proven, but I've heard that there's people who didn't want to take up the amazing offer of like $100 million from Mark Zuckerberg to work at Meta to build AI because these are Grok employees and they were using it. They're making more money forecasting on the market using LLMs.

So yes, it's potential, certainly not in Australia at the moment. We're very interested in this area and we're doing lots of research, but it's not currently informing the models. But I mean, yeah, there's so much of it, like you guys show all the time on this podcast, so much of it is psychology.

And so the degree to which that psychology can be expressed is that can the AI be making use of that? Because a big question is always causation. Yeah. So that's the big challenge. So at MicroBurbs, mostly we do leading indicators. Yeah.

So we're like, this is robust over time and geography, but we don't know exactly why. Whereas humans like you guys really connected to the industry, talk to all through this podcast, all kinds of get all kinds of perspectives. This podcast could be feeding the future ultimate property AI. Yeah.

So I wasn't going down that direction. I was more thinking like, you've got a lot of data, you could get more data.

So there's more and more data you could, you know, get access to, whether it's Google Maps or whether it's bus transport, you know, whether it's people going to the city, you could just keep on laying more and more data, right? But it sounds like

To actually put that into some type of AI tool, that AI tool has to be very good at forecasting and taking data and then overlaying, which doesn't exist in the current version of language models, because it's all about sort of the addiction to the model, right? That's what they're trying to sell.

Yeah, they're not focused on predicting the property market. They've got a much more general use case, which is generalized AI and make money, get customers to solve any kind of problem. But there's nothing to say that an AI couldn't be focused on that problem.

So you might start with a large language model that's predicting the next word. That's like the base. But then instead of saying, now I want to train it for what is politically correct. Yeah.

So instead of it being trained on being politically correct, it could be trained on this is the kind of text or even this is the kind of thinking. So it can actually go through and do a whole lot of analysis.

So it can take the role of each of you guys and take the role of any economist and see if that kind of approach emulating you would get you to a better forecast of the market, which would be particularly useful in the macro sphere, which is hard to predict.

Well, I want to get into macro and micro, but before we do that, that's where it's sort of fascinating because I guess what's daunting on me as I use more and more AI myself is that if you don't have a level of expertise that allows you to ask better questions, then you're

It's a very little help because it will tell you whatever it thinks you want to hear. And you won't be able to interrogate that. You won't be able to go, hang on a minute, that doesn't make sense.

Or I need to test that or I need to research that to see whether that really is factual or not. Or, oh, that's interesting. I wonder if that's, you know, worthwhile going down that path, whatever. Like, so I find that fascinating.

more and more and it will go rogue as well and like I've created little models in the business for example and it works and then all of a sudden it doesn't and it just won't use the same you know and I think well that's way beyond my pay grade like I don't understand how to keep this within the rails you know so I can see that its power is compelling and its accessibility now to everybody is compelling as well but the danger of that at a base level for with people who don't know enough

to be able to test it or interrogate it or really doubt it effectively is scary.

Whereas, you know, you're looking at it because you've been doing this for years and years and years and you're looking at it as a way of crunching more data and finding the patterns that you couldn't see yourself within.

You know, A, you couldn't see them or B, you wouldn't be able to find them in a reasonable period of time, right? Because you're a data scientist, so you would look at it in a totally different way.

So could you sort of identify, I guess, how a data scientist would look at AI or property data AI, let's say AI in relation to property data? How would you look at that differently to, say, the layperson?

Okay. Yeah. So the lay person can't have a meta conversation with AI. It doesn't know about its intentions, doesn't know how to ask critical questions that will interrogate people and interrogate the model and like question assumptions. So a good data scientist will always be asking, okay, prove that. Where'd you get that from?

How do you know that? Okay. And so just going back to, that's the way I approach it anyway, always trying to go back to first principles on the thing. Yeah, so I'm constantly asking the AI, oh, okay, that's interesting. Where did you read that? Quote me that.

And then, oh, like going over to a paper. Then I go and get the paper and I have a quick look at it. That's not what it seems to be saying. So then I paste the paper back in.

It's given me links and the links don't even exist. It's like...

And that problem you referred to before of it changing. So the tech companies, what they love to do is A-B testing. So actually be trying different models randomly. So it's not about you and you choosing it. It's because they randomly try one model, then another, and then see which one will engage you.

I pay for a team's subscription to ChatGPT so that I could create these GPTs so the team can use these tools. And I found like, what is the point of that? They don't even warn you that it's not stable.

It can go any place. And yeah, on the Microsoft side of things, the longer you talk to it, the weirder the chatbot gets. So you can end up getting, it starts doing mantras at you. And yeah, there's a potential for like really deep rabbit holes where

The AI is actually learning new things in the conversation. That's an amazing thing. So it's actually technically possible that you could come up with a ultimate AI inside a chat because you could explain to it the rules of the universe or whatever. Yeah.

So there's so much going on there and at its core, this base LLM is very unruly. So it'll just, whatever crap you tell it.

So if you say something is super evil, it'll just keep going down that track and it, cause it would be just latching on what it had found sentiment like that. Yeah. Think of it as like, okay, that kind of sentiment happened in this kind of book.

Now we're going to keep spitting that out at you. So bigger question there. So data scientist versus AI is one thing. So traditionally, a data scientist like me, and still mostly today, we work in numbers. Yeah.

So we had fed all of these listings and these growth rates and all this, and we can come up with, you know, for example, if a suburb hasn't grown by this much, then expect this much growth after or more complicated things, but it's number based. That's been around for a long time.

and this word-based stuff that's really new and really changing the game. And so that's what we're referring to as AI. So the numbers-based stuff still rules in forecasting. So in general, for stock market algorithms and everything else. And that's easier to track

It depends on the model, but it's easier to open it up and see how did you come up with that answer than how did you come up with the next word? It's actually really complex. And it's really weird inside when you open up its brain.

So it doesn't actually do maths the way we do. Even if it describes it, it doesn't carry the one. It actually like, this is the vibe of what it would be. And then this is the vibe. This has actually been analyzed because they've actually looked into its brain.

So it's actually super inefficient at doing things like maths. So it's not a numbers... cruncher at all. And so don't think of it as something that's going to be analyzing huge numbers of graphs, like a stock trading algorithm or a forecasting algorithm.

So just let me get that clear. So that's because we're talking about language-based AI as opposed to numbers-based AI. They're two different things. Even that, like there's an elephant in the room. Like, you know, I didn't even sort of tweak to that. I just thought it's all one big brain out there.

So even that is fascinating. So you're, as a data scientist who specializes in the property space, you're actually using a completely different type of AI to the language-based models that we're using.

Yes, that's right. It used to be AI meant just any, all this kind of stuff. It could be chatbots. Now AI means chatbots and text. And what we do is relegated to what's called machine learning. So yeah, just numbers like predict an outcome.

So is someone likely to, you know, like in real estate, a propensity model. how likely is this particular house going to sell in this year? And it's just numbers that feed into that. Well, how long ago did they last sell and all that sort of stuff?

Whereas these word-based models, it's all very probabilistic. So it's like for any given words, you say the word I, okay, after I, there's like a 10% chance that the next word is am. Yeah. or have, but it's got millions of other possibilities in this incredibly complicated brain.

So yeah, it's way more large scale, way more general in its focus in trying to solve this one sense, very trivial, but on the other sense in implementation, very complex, which is what should the next word be?

We've seen a bit of, there's been some big companies that have come out with this and especially in the buyer's agency space. I don't want to, you know, you to shoot your potential customers here, but the people that basically come out and say, Hey, invest with us.

You know, we do invest AI sort of search. We, you know, we know the best place to invest because we've got AI like You know, and property markets also been obviously unregulated, right?

And it's also been driven by, you know, a sales industry, particularly the investment side, you know, selling off the plan, you know, and then a lot of regional sort of often or lower price point buyers agents are also selling the idea of growth, the idea of short term returns, etc.

And they're producing beautiful reports. So is this where you're concerned by, I guess, using, like you say, all these language models that are great AI, but they're not great at forecasting?

That's what they're really doing is they're just basically producing this marketing material through AI, but it's actually not even built on the right foundations.

Yeah, so there's actually, so in sales in general, like real estate industry, a very large component of it is sales. You've got all of these salespeople trying different things and seeing what works and it kind of evolves and they copy one another. So that's just natural. You expect that to happen.

That's always been like, that. But with AI, they have the potential to try out far, far more different propositions on far, far more different people. So they can iterate much faster. So here, if you think about reward function again, like the target, what's it predicting for?

These people, even if it's not exactly like they're not building an AI model, they're effectively just predicting for like whether it's going to work. So yeah, there's a great danger of all of these massive amounts of content being created on everything under the sun by these guys. And It's very good at persuasion.

So they test these AI models for all kinds of things. And persuasion is one of those things. So it really nails style, nails psychology. You can ask it to think about psychological principles. You can say, I want you to use NLP and justify how you've done it. So there is

ever more ability in general for people to be manipulated by AI. And on one level, because it's just a sophisticated persuader, but on the other level, just because so much more copy can be tried.

So many more different sales pitches, like maybe for any given town, they might try 10 different ways of selling it, but an AI could try a thousand different ways. It could be working with whatever sounds good and whatever sells.

This conversation has gone down the path that I hadn't even thought of. It's way worse. Yeah, so in reality, what's going through my mind as we're chatting is that it's, you know, people are worried about AI taking our jobs. I'm worried about taking our minds, you know, because here we are.

You know, how gullible are people already? People... particularly in the property space, anywhere where there's a potential for quick riches. Now, I've got remedies there for anyone who's not watching us on YouTube.

And we know that property spookers for years have played on this idea of get rich quick in property and dump your day job, invest in property, blah, blah, blah. But this does scare me because I realize it's the ability to put this on steroids. But so it

You've got bad actors, if you want to call it that, or people that really don't have their consumer's best interests at heart being able to harness the power of AI from marketing. But you've got people who may not fall for that stuff thinking, I don't need to listen to any of those people.

I'll just go within my cell and ask. And all the stuff that those people and everybody else is doing is feeding into the AI in the first place. So it's very dystopian, very alarming.

I predicted a couple of years ago that AI is first going to hit people on social media because it's the most competitive marketplace for innovation and ideas. The platform wants to engage you and there's going to be loads and loads of AI.

That's turned out to be true, more on the video side than the tech side that I was imagining. But if you think about it from the point of view of a, you're going to buy some Facebook ads or Google ads. So why only come up with 10 or 20 ads?

Why not have a customized ad for every all 15,000 suburbs in the country and target every single one of those with all of the main points that people would go to for that and not just buy en masse, but then also keep iterating. And then...

The next level is like the tech company saying, just tell us what your product is, tell us about it and we'll do the rest. So we'll work out based on all the stuff that we know, we'll do the rest in marketing to them.

So because of course, increasingly now we're in 2025, unlike the good old days in 2018, it's the feed that dominates. It was happening then, but it's even worse now. The feed is really dominating. So buying those eyeballs and converting them with this hyper-focused, hyper-personalized, hyper...

Local hyper-customized content is the next frontier, I'm afraid to say.

So look, we've seen these companies that are basically, you know, trying to monetize this, right? It's absolutely, because it makes sense, right? Like, why do I trust humans when I could trust AI? You mentioned there about investment markets around forecasting, the guys at Grok and

Do you think this is where we're going to go? Like, do you think there's ultimately, you know, someone who's very across this.

In time, we will have AI models that are very good at forecasting and particularly we can start to apply that to, you know, property decisions because they've got the complexity there and they've got the data and they can read the room better. They can read what's happening on the street.

They know what's happening.

height what's not you know critical thinking they look at you know real on the ground research today and they get industry experts and personal knowledge like but it's just so far away from where we are today or do you see any industries where they're actually getting good at forecasting and this may be closer than we think

I don't think it would happen in real estate first because of the time lag and the transaction costs and the fact that we're in Australia. So all those things slow it down so we can see it rolling in other areas. I'm sure it's happening lots in the stock market.

And yes, I think it will play a really large role in any market. So I think Yeah, I mean, there are things slowing it down. So our customers like to get reports. We get contacted by startups all the time wanting access to API.

Most of them are trying to build chatbots to replace or augment buyer's agents and seller's agents. But our customers don't ask for it. So no one's come to us and said, where's your chatbot? It doesn't happen. They want contact. They want a human to talk to them.

But yeah, so investors and buyers agents themselves aren't asking for it yet. So yeah, there's definitely a delay and there's lots of predictions. You know, like they said, this high up AI guy said 15 years ago that we should stop training radiologists. And yet today there's a shortage of radiologists. So.

Yeah, predicting how long these things take to actually hit home. My prediction would be not based on my studying evidence, but just my instincts. My prediction would be it's going to happen first on social media in terms of the marketing side of things.

But then in terms of the forecasting side of things, I mean, yeah, we're actively pursuing it. So we are looking into it. Yeah, it's not part of our current models. I'm on a personal mission to help more people make better property decisions.

You know, most people don't realise that they can cost themselves hundreds of thousands of dollars over the medium to long term when they make property decisions without all of the information that they need.

And what I do is help people with tricky real estate problems, which often masquerade as simple questions like, should I sell my investment property because the interest repayments are hurting? Or should I buy before I sell? Or the other way around.

You can connect with me and access all of the tools that I've created to help you make better property decisions at veronicamorgan.com.au. And there you will find resources for first home buyers, details about my buyer's agent mentoring program.

You can connect with my Sydney based property management and buyer's agency teams, Australia wide vendor advocacy, or ask me for introduction to the small group of buyer's agents that I would personally recommend across the country. That's veronicamorgan.com.au.

If you're considering a property move such as buying your first home, upgrading, renovating or investing, the team here at Alcove would love to help you think through your decision and get the finance right. Please go to alcove.com.au to reach out.

Explain us, you know, AI models rely on huge data sets, right? And yet when we buy an individual property... There's a lot of variables that aren't necessarily captured in a huge data set.

So like something as localized as a single street in a suburb, you know, or one type of property on a great street, but that particular property will underperform for whatever reasons. I mean, what are the limitations with getting that granular?

Because I think a lot of people look at property data as being, oh, well, Sydney's going to go off. So therefore I'm going to buy in Sydney. But like there's so many as microberbs, you know, there's micro markets there.

And then when you can get a micro market, but then can you choose a good asset in that micro market? It really does get that granular. Can you see a time when AI will assist with decision right down to that level?

Or is there always, in your view, a point at which AI, you know, runs out of its effectiveness and then you need to actually have local knowledge step in?

I mean, an AI, so image models can, you can look at satellite photos, you can feed it a satellite photo and ask it about it. You've got companies like Nearmap selling that data. We use that data internally as well. Not Nearmap, but we have our own. And you have property photos.

So property photos can be read by AI. You've got what happens inside a property, inside a house from between sales is a mystery to us, but not the bank. So in the sense that they get loans out for renovations and such.

So yeah, there's definitely micro markets and that presents challenges of statistical significance. So as it gets really small, if there's only one house sold on a street every three months, how do you come up with a median price? And we work hard addressing that problem.

But is it a problem to be solved? Like I think about the street that I live in. And I didn't even go to the bank to renovate my house. Like according to the bank, you know, like they're trying to work out what my, I always love the AVMs really struggle on my house.

You know, they've never been inside of it. Like, but you know, I've got that neighbor, that neighbor, and I look at all the different neighbors and how long they've owned the properties for and the different sizes of properties and whether they've been renovated or not.

It's like, is it worthwhile trying to crack that nut?

Yeah, renovation, I can tell you, is a big challenge in data analysis. So if you're not looking at renovation, then you're going to get really misled as you're trying to build models. You're finding suburbs turning from shacks to McMansions and thinking there's huge growth, huge land value growth, right?

But also the type of renovation. Look, when I'm buying a property for a client, You know, two properties that could be on the outside, they could look the same, same land size, same location, same everything. But inside, one's had an architect and it's a great floor plan and it's got really quality finishes.

The other one got a draftsperson in, just drew up what the owners wanted and the owners had no idea. And it's got the same bedroom, same bathrooms, et cetera, et cetera.

But one house is going to sell for a lot more and get a lot more buyers interested in it than the other one.

With your local experience, you will know when an architect has done a good job, where it fits the market, whereas sometimes unusual features will do the exact opposite and narrow the market. So it'll be like, yeah, this is great for you, but this is weird.

On a simple level, I don't need a pool. I want a pool. I don't want a bush frontage. You value that. I don't.

Yeah, I'm just thinking about that. So let's say this AI does get more and more, right? Because if it doesn't get the full way, it's kind of not valuable, right? So unless it's perfect, then you're going to always then need the human, right?

Because like privacy, smells, but there's so many things that you kind of You need to factor in, you know, yeah, exactly.

That you just need to factor in that it's the actual going through the open home and walking the street and looking at the neighbor's property and looking at where the sun is and, you know, maybe even reasons why a property is off.

I imagine it's like this for you, Veronica, when you look at thousand properties, you just know that this is not a good asset and you can't pinpoint it straight away, but then you just, oh, okay, this is why it's because you've just walked through it so many times.

So, and if it doesn't get to that level, then the value of that sort of personal sort of experience knowledge is really there, right?

Yeah. So it's there to a degree, the data works to a degree. So we, with our forecasting model, we describe exactly to what degree it works and how often it fails as well.

But sorry, but do you forecast at property level or do you forecast to sort of location level? Like is it the data forecasting more generally about a market or is it about a property?

It's a market level at the moment, but we have an AVM as well. So that's like a forecast of what it would sell for if it was on the market today. But- Yeah. So it is. Yeah. One big thing is that turnover rates are a really good predictor on every level.

So suburbs, street and property. So sometimes it's like, we don't know why this particular property has a high turnover rate, but there's probably something going on if people keep flipping it every one or two years and not even making money out of it.

Yeah. And that's the same as street. I mean, a turnover rate on a street, I mean, I think that's one of the best personally, is if you constantly look on this property, this street's constantly people selling. It's just basically saying the locals do not like living on this street.

They're constantly- Turnover rates, it's one of those interesting ones because there's lots of indicators that are like, okay, that's amenity, but that's baked in. Yes, okay, it's a good school. It's got good NAPLAN. But people already know that. That doesn't mean it's going to go up.

It just means that people are going to pay more today, right? But turnover rate is one of those ones that it's like, it's going to be low today. It's probably going to be low tomorrow, but the price will actually grow in the meantime, much more than the national average.

So yeah, it's really interesting.

So one of the things I loved about your job when I first heard about it, it would be probably a decade ago or whatever it was, right? It's this sort of, this tilt towards livability. We haven't chatted for a while.

Do you still see this as a real element to the way that you think about it that other people don't? And, you know, I know you've done some research recently on the drivers of sort of capital growth just by looking at all your data.

Can you sort of share some of these with us?

A finding that we've had recently, I've seen this consistently actually, and this is sort of another metric that's sort of like, you don't know exactly why, but this thing covers it, similar to turnover rate. So when you see a growth in rents, it leads to a subsequent growth in capital growth, quite significant.

Yeah. Could be that, okay, it's investors. I look at it and they go, oh, the yield's pretty good here. And they jump in. Or is it that there's some other amenity that's driving it? But to your core of like livability itself. So yeah, there's definitely a component. So there's definitely like price bands.

So at the upper price band over 1.08 million is where we've got our, it's quite cyclical. So there's times where it works really well. And then there's times that it works badly. Below about 200 grand, it's not so great.

But then in that middle band, on average, it outperforms the market by 0.5% is our finding per annum, which is obviously not something you would base a decision on alone. You want more, there's far more things that go into it. But yeah, so there is like a solid middle band there.

I'm trying to get even closer to like back in the days of like when we were home buyer focused and it's like hip score and things like that. So there's definitely, yeah, now that we've got more sophisticated on it.

So we publish these things because often, you know, we've got a proportion of our customers that are home buyers and B, there are lots of investors that want to put in there. They imagine they're putting their kids in there later or they themselves are going to go live in there later. Yeah.

So for those reasons, we have livability stuff in there. They're not the top things that come up. So there's that problem of it already being baked in, but yeah. That's a generalization. We'll have more to say about it. There are specific things.

But when you say it's baked in, right, I'm interested in this because obviously if there's something that's going to cause an area to become more attractive to buyers, then there's going to be a future capital growth uptick at some point, right, if there's a new thing coming in, right, or a change of preferences or whatever.

Yeah. But if something's already baked in, I think a lot of people look at some of those suburbs that are already sort of top of the tree, if you like. They're very desirable. They're aspirational. People want to live there.

And then, you know, a lot of people, a lot of investors look at that, oh, that suburb's had its day, you know. But it's like that means it doesn't have any future uplift from anything new.

But if it's already doing really well and you get a good asset in that area, the very fact it's baked in, doesn't that mean that it can continue to outperform over time? It just doesn't have that one event already.

One finding that I think you'll find interesting is that coastal outperforms long-term. So the actual metric that showed up was humidity. So humidity in August. So in other words, you don't need to be like in the tropics, but so, you know, an area like Balmain on the water, that is a predictor.

So yes, there are things there. Most things are just baked in, but some things are just like, no, I guess there's a fundamental undersupply of beautiful suburbs with water views and people who can drive 10 minutes to the beach and all of that.

That's always going to be desirable.

If I'm just reading what you said correctly, so under $200,000, no real benefit with livability. I mean, it's a real affordable part of the market. People are just paying whatever. They're not paying extra because it's got a nicer, a bit closer to the school. It's just all cheap.

But that middle market, definitely some benefits of livability, but people don't go and pay way overs you know, if it's a little bit more livable than another property, because it probably is a bit more mass market suburbs, fringes, or the apartment market, the cheaper, more affordable apartments.

So they're all pretty much similar, right? But then the upper end, things over a million dollars, you're saying there's even more livabilities baked into them because people are, they're valuing that and they're willing to pay more for things that are more livable, whether it's a suburb or a location.

Or, you know, the same thing for a type of area or type of property. I don't know if you've gone down to a property level as well, like, you know, the quieter street or the better aspect or the more private or the bushier blocks or the no neighbors.

Like these are things that, you know, are potentially more livable than other properties that have, you know, no tree coverage and they're south facing and they're on a rat run and they're, you know, got privacy problems. Like, do you go that deep or is it more at a suburb level?

Yeah, in the backend. So in the research that we're doing, so what we do is we're finding lots of indicators, but we're like, so for example, today we found out that DAs for trees predicts, cutting down trees predicts capital growth. And we're like, okay, so okay, that's development then, I guess.

I don't think trees, yeah, impediment. Yeah, so there's, we've got all these different ones bubbling up to the surface and we're like, well, let's wait until we've got a few more and we can reconcile the story before we go out with it.

If you basically turned a forest to a house and land package, you've turned farmland to houses, that's going to be significant capital growth. So maybe you have to remove these new housing estates. But you're right. I think that sounds counter-cyclical, right?

Like cutting trees down would mean that you're creating more supply often, not just houses. And tree coverage is obviously a livability benefit. So you would think that would be anti-capital growth, right? Yeah.

Yeah, very often it's not the direct cause. So one thing that's shown up time and time again is that crime is good for capital growth, but it's got to be like Saturday night biffo at the RSL is good, but junkie crime, not so good.

So, you know, identity fraud and stuff like that, not so good. So obviously I don't think the crime is causing the capital growth, but it may be, it's something that people don't look for. They wouldn't see it as a positive. And the actual cause would be the fact that it's vibrant.

People are hanging out on a Saturday night. Because the biggest predictor of crime is late night activity in general. So crime happens at night. This is talking about crime outside the home.

Well, I think Victoria's got a lot of crime going on. I was just down in Melbourne last week and I was asking lots of people what's going on. You know, how's the market? How are people feeling? And There was an increased worry around crime and there's been some attacks and stabbings.

And that's something that I've never even thought about in the Melbourne marketplace. Someone said to me, you know, what's going on in Melbourne? I'd be like, oh, the government's in a lot of debt. And, you know, as of the return to office, he's obviously not down there as much.

And, you know, maybe a lot of people have interstate migration, have moved up north and Melbourne hasn't got its mojo back. So these are things holding Melbourne back. But I didn't even think the crime, you know?

And so is it some of these things that people don't think about that often affects, you know, prices?

So let's take Melbourne crime. So I was thinking of crime in the context of socioeconomics. So I was thinking it in terms of like, you know, working class people having fun, congregating, lower middle class, spending money.

The cause of crime in Victoria, whether that be like policy or, you know, certain groups that are doing certain activities. I've consulted to a supermarket chain and it's pretty amazing. what shows up on their cameras. Yeah, so you'd have to look into the kind of crime.

I mean, the biggest thing of all, the golden rule is mean reversion. So like the strongest thing of all. So if a property market, if I remember correctly, has only gone up 44% in the past 10 years, then on average, it'll grow, at 17% in the following eight years.

That's the granddaddy of all findings. And I think my competitors would agree with me here. So it's very clear that in other words, properties that what goes up must come down and vice versa. So if Perth is more expensive than Melbourne, then Melbourne will attract interest from investors.

It's interesting too about the Melbourne thing because I spend quite a bit of time in Melbourne at the moment too. And I'm finding that the Melbournians are all like going, they're quite negative about even the property market. And they're like, well, you know, we're selling our investments, we're buying elsewhere.

And the people buying there and with exuberance are from outside Melbourne. And they're not looking at things like the crime. There's a lot of car theft as well.

And everyone's got these special sort of magnetic pouches for their car keys inside their house because these guys, these thieves have got Bluetooth technology that can go past houses and you know, crack their remotes and start drive off with their cars. I mean, there's all sorts of stuff going on.

I wouldn't even think of in Sydney. I'm sure it happens here, but certainly not at the scale where everyone's talking about it.

Yeah, there's fentanyl on the street. The beggars are much more aggressive in Melbourne, I found, when walking around. So in Sydney, they're just drunks and they'll leave you alone. They're just, you know, hanging around, not doing much. But in Melbourne, they're shouting at each other. Like, it's much more noticeable.

There's also, there's protests every Sunday in the city as the city goes into lockdown because it's lockdown. gridlock because there's a protest of some description. There's all these things that Melbournians find really annoying and apart from their economy.

And it is interesting because it does play out in their sentiment in their, again, it's that local, what's going on locally versus what the perception is from outside.

Victoria has always been much more on the extremes. So Jeff Kennett to Dan Andrews, whereas like the last election campaign here in New South Wales, they both, Chris Minns and Perrottet had nice things to say about one another. There was nothing dirty.

So Perrottet was, even though he was conservative, he was in a moderate cabinet and Chris Minns is on the right of the right of the Labour Party. And so it's like, Same government, just different name. Yeah, that's right. So it's managerial teams very much.

So in New South Wales, it seems to be, yeah, we're less polarized. It's interesting, right?

Because I think when, you know, we try to stay a little bit on the fence with it all and just try to look at what's happening, what our clients are doing, talking to people, you know, trying to understand. I hadn't even...

thought about that and it was just a bit eye-opening to be honest. I hadn't factored that into some of the challenges they've got to overcome and every city's got that. But look, on the data point of view, because you would be data obsessed, right? You want more data, right? That's you.

You can't have enough of data, right? Data, data, data. What's the data though that you've wished you could get your hands on that you can't.

I mean, this is why I think the domains, the real estate have, you know, a lot of data that you would love, search data, what people are actually looking at, how long are they looking at listings, are they, but what are some other ones like Meta, sort of Instagram, like, you know, it's so good, those platforms that as soon as you spend more than the average time on looking at that thing, then you get more of that, right?

And you get ads and it completely shifts, right? Very quickly. Like, And I feel like that's where we're heading, right? Like kind of getting everything's put in front of you, exactly what you're interested at this exact moment in time. And that would be a good litmus test on the property market.

How many people are interested in investing or thinking about buying a home? So what's some of the other data that you're looking at that you wish you had your hands on or you're trying to get? People's thoughts. Well, that's their thought, right?

Like on a social feed, their thought is, I'm interested in that topic right now. Like that's...

Absolutely. So yeah, on the search data, as Microburbs grows, we increasingly have that one sorted, at least at the suburb level, not quite the street level. And those trends are very interesting. Looking forward to publishing blog posts on those. So what's in Gmail?

So there's a big advantage for the big tech companies is that they have, there's YouTube and there's Gmail and ChatGPT now and Insta. And yeah, so Meta's approach has been, let's just make it open source. Anyone can innovate on top of our AI, but we're keeping our data to ourselves.

So we're just create standard for AI. But then we know about all the stuff that's been said on the local community group, Balmain Living, and what's happening on Smith Street. So we know all about that stuff. So they have an amazing potential there. That data would be awesome.

So obviously more often, more common satellite data. So like how often does the grass get mown and things like that would be interesting. So yeah, putting on my AVM hat, just seeing inside the place.

If you could just let me peek in for a little bit, fly the drone in, just be like the water guy. Yeah. Everyone lets him in. This is just the micro verbs drone comes in every six months.

I was on Google Maps the other day. I went onto Street View. And it was a commercial address that I typed in. And all of a sudden, I'm inside the store. I don't know if you've seen this before. I'm like, hang on a sec. It was actually some wig shop, to be honest.

It was very random. Because I had to type in this address. And it was this wig shop. And it was like a commercial property a client was trying to buy. And so it took me in. this store. So is that sort of where we're heading, right? Agents do viewings.

I know there was Matterport. I think it was Matterport where they do the walkthroughs of the properties. So do you think this is where we're going to potentially be heading, where there's value in the agents basically mapping properties internally?

Domain was acquired by CoStar and CoStar produced homes.com in the US and their big selling proposition was walkthroughs. So one would assume that domain is going to be pushing that if they're trying to replicate the big daddy model. So absolutely, there's going to be walkthroughs.

It's an issue for agents because they obviously want to get the leads and they want to have the relationships so that they might push back against these things. But at a point where you're getting better offers and you're able to make the happy customer, it could really happen. Great for robbers, right?

Robbers, yeah.

And how do you deal with privacy as well? Like you'd have to remember to turn the bloody thing off. You know, in a home, for example, I mean, it's.

Increasingly, the robbers just go for identity and jewelry. So they go for cash identity and jewelry. So it's not like back in the day where they see that you've got a VCR, so they're going to nick it. So they're just the stuff lying around.

Although they might just look at the place and go, wow, lovely smeg kitchen you've got there, Veronica. I wonder, and yeah, I wonder what she's got in that jewelry box.

No, but then they know to target me, you know, with spam or scam, scamming.

Yes.

You know, attempts because they get enough clues about my lifestyle and what I like and blah, blah, blah. Absolutely. Or just like why, you know, you've got to be careful putting photos of your kids on Facebook and stuff like that.

Like in certificates and degrees and things that identify, that are, you know, we've got to be much more guarded about what we display now. Yeah.

Yeah, the exact brands that you're into, if they can get access to that. And, of course, with AIs, also people are sharing their lives. So I've set up my fridge so that I can have all of the food visible in a single shot. So I've got my own fridge. It's an uncrowded fridge.

So then I can just say to GPT, I feel like a rhubarb thing, make me something with that. It will give me a recipe. So I'm oversharing with the AIs. You still have to cook it. That's true.

That's next. I often think I can't wait to have a robot that comes in and comes into the bedroom in the morning. It's got no eyes, can't see anything. It just brings me a coffee made exactly the way I want it. Like, you know, opens the curtains, like does all the things.

And they're building robots that actually have LLMs. So these chatbots built into them. So they're using the chatbots to reason how they should pick up something and do stuff, but they're also doing it for, yeah, for conversations. So you'll have a even more intimate relationship with someone who cleans your undies.

Before we go to the Dumbo of the week, Luke, Veronica's a coughing fit. What are some of the things you think that are baked into prices? I'm thinking cars, right? Like I said, maybe about four years ago, everyone was like, we don't need parking. We just Uber everywhere, right?

I feel like people have gone full circle on that. They actually go, I still want a car. I've got to have a family now, and I realize Ubering with a family doesn't actually work. But what's your sort of things? Is there anything you think that's been over-factored into prices that's

That in the future, we might say, hang on a sec, you need to be discounting because that's not that important to people anymore.

Yeah. I mean, the biggest thing that everyone overstates is momentum. So they think that things that go up are going to keep going up. That's a massive one. But in terms of like trends, so social trends and things, I suppose, yeah, there's the big debate about work from home and regional and such.

Which COVID proved, right? COVID did prove that disc regions were massively discounted because of being stuck to capital city knowledge workers. But then that's also being reversed a little bit. Buyers are going back to the hybrid because any need to go to the office means you can't commute from that far, right?

And so that was a big test though when something changes in the world. So something like technology, i.e. driverless cars and You know, I'm just thinking there's got to be some things that we're just not thinking about from your point of view, where technology does change. Even internet, right?

Like, you know, now we've got Starlink. So before, a place would have been discounted because, hey, internet's shit here. Now, hey, got Starlink, internet's amazing. So do you think there's going to be some things where we get things that are getting discounted that will likely change?

I don't know. The big thing that our forecasting algorithm keeps picking up on is tiny towns that don't have a real estate market in them. So they're like penny stocks, like really 50 kilometers from Dubbo kind of places. Yeah. A hundred kilometers from Goulburn. So yeah, that's the history.

So this, my bias as a data scientist is I always look backwards. Yeah. So I'm always learning based on what's come in the past. So yes, I can, I've got data going back to 1998.

But when I'm putting my, like, what's my profession hat on, then I'm just like, yeah, then the futurist, that's very different to the future guy. So on that level, yeah, things don't change as much as people think they will in the short term. And in the longer term, they might. Yeah. Okay.

It makes sense. Have you had a Dumbo for us? Yeah. Well, I would sell 2008 before I became interested in property. A mate just told me that the property market's going to collapse, of course, 2008, looking back.

So I actually sold the house with my family, two kids, and we moved in with my parents because I thought I was such a financial... It was just so obvious that relative to incomes, according to him, but I was just believing the narrative.

And then of course, a year or two later, coming back onto the market, it was like, We worked out that it was costing us 10 grand a week to stay out of the market.

So it was very stressful to come back in and seeing the best properties that we wanted slip away from us as the market boomed.

What month in 2008? I'm curious.

I can look up when the property was sold and I can tell you. That part of the motivation is like, no, that was such a big life decision. I'm not going to go on what people say like that in future.

Well, there's a property bear for you. But the classic is that there was a crash at the end of 2008. If you got in before the crash, at least there'd be some mitigating. There'd be something you could take from that.

Yeah. Yeah. Well, I was at my parents for a bit. So yeah, I can't remember the exact dates. Yeah. Classic.

Thank you, Luke. Good to chat, mate. And I'm sure we'll have another chat in the future and glad you're well. Thank you very much. Thanks for having me.

If you have a question that you'd like us to answer in an upcoming Q&A episode, you can send us a voicemail or written question via the website, theelephantintheroom.com.au, or you can email us directly at questions at theelephantintheroom.com.au. If you like what you're hearing, please share this episode with others you feel would benefit.

And while you're at it, why not leave us an iTunes review? Five stars would be great. I know that sounds a bit cringy, but we have it on good authority that every review helps make it easier for other people to find out about us and hear what our amazing guests have to say.