@Mujo, Ben Clark from TMS Private Wealth kind of addressed this issue on an episode of The Call on Ausbiz.

From memory he said that AI could source all the information required by property buyers directly from the real estate portals owned by real estate agents such that the online classified businesses could be bypassed. 

However, Ben still had a buy recommendation on REA as he thinks that even though AI is a threat in the future he believes it is still too early to really know. Oh, and he wasn't the least bit concerned about CoStar (Domain) as a competitor.

Coincidentally this article from The Australian just popped up on my radar.

REA Group partners with OpenAI to launch AI tool for property valuations

REA Group has locked in a partnership with ChatGPT maker OpenAI, leveraging its trove of real estate data to create new artificial-intelligence tools to demystify property valuations and help people find their dream homes.

The new product – RealAssist – is designed to fend off competition from global tech giants and their AI platforms, which are encroaching on the turf of more traditional classified sites and search engines, including Google.

REA’s AI strategy aims to arm millions of homeowners with personalised, data-backed insights to make property valuations more transparent.

RealAssist allows realestate.com.au members to ask questions in conversational-style lan­guage about their property’s valuation, generated from REA’s realEstimate feature.

REA Group chief product and audience officer Jonathan Swift said RealAssist had been developed using tools from OpenAI, which had opened an office in Sydney. He said the product was one of eight “AI-prime features” on realestate.com.au that allowed people to search for properties in the same way that they spoke.

He said people could even ask for suggestions about what would enhance their realEstimate valuation.

Homeowners may also be prompted to update their property attributes – an action taken on more than 140,000 homes in the past 12 months.

“Australians are obsessed with property but it’s often an emotional and complex journey,” Mr Swift said.

“RealAssist aims to demystify property valuation, helping homeowners build their knowledge and empowering them to take the next step.”

Coinciding with the launch of RealAssist, REA Group is expanding its natural language AI search feature to all property seekers following a successful 12-month pilot. The expansion allows users to type complex, spoken-word queries such as, “show me homes with a pool and a double garage in Kew”.

Mr Swift said while the initial expectation was for common search templates, the reality was starkly different. “Just about every query is unique,” Mr Swift said. “[A query] would be like ‘I want a three-bedroom house. It’s federation style that’s five minutes walk from this cafe’.

“Natural Language search is great at answering some of those, but as we continue that development, this is where we think, can we do a better job of understanding what drives your lifestyle preferences, where you’re at financially?”

This finding suggests that a purely filter-based search system is insufficient for meeting the complex, lifestyle-driven criteria of modern property seekers, ­cementing the need for advanced natural language processing.

When discussing the emerging threat of large generative AI platforms such as ChatGPT, Mr Swift said REA’s most distinct competitive advantage was its unique, proprietary data.

The company leverages a massive data set, including details from the one in three Australian homes tracked by their owners on the site (equating to 4.7 million properties), high-volume consumer feedback on valuations, and data from recent acquisitions such as 3D mapping company iGUIDE.

This proprietary information is used to train and ground the AI models and stay ahead of the tech titans – even if it means using their own tools.

“We’ve got all those unique data sets that are not available to any third-party AIs. They sit within our walls,” Mr Swift said, adding that the challenge was creating “the best property experience”. He said: “We’ve got to have the best property experience and best representation and best data sets, bringing to life the best way of discovering and helping you move through the journey.”

Mr Swift said the tool, by providing data-backed answers, not only empowered consumers but also enhanced the relationship between homeowners and real estate agents.

He said that a more informed consumer would engage in a “two-way informed connection” with agents, ultimately leading to better and more realistic expectations for a sales campaign.

Despite the rapid adoption of AI technology, Mr Swift said REA Group was taking a measured approach, particularly with the more complex “agentic AI” capabilities, which would result in the platform acting semi-autonomously on behalf of consumers.

This cautious stance stems from general consumer sentiment. “Aussies have quite a low trust of AI. So we need to take them on the journey,” he said.

He said REA – majority-owned by News Corporation, publisher of The Australian – had introduced “strong guardrails” to ensure “accuracy” and would continue to work with leading technology businesses, including OpenAI, to build on existing capability and power game-changing features.

“Strong early signs of repeat engagement and the uptake of ­actions from consumers using AI-powered features indicate consumers are ready for these experiences that help them find their next home,” he said.

It seems to me like every man and his dog has some sort of tie up with OpenAI these days (or rather, the other way around - they have to stay relevant as Alphabet asserts their dominance!).

@Mujo the AI fear is we dont know how well these applications will work. possibly the agents could list properties with a view that AI crawlers used by buyers find them fast so a listing on REA is not necessary. if they fail to sell they list it on the network. so that depends on buyers/sellers making possibly the largest purchase/sale of their lives relying only on an AI crawler. the search could be a "am i feeling lucky" so avoid the REA fees, then move onto REA. how much leakage could occur? has REA options, can it prevent a free option?? would sellers be happy with only an agent listing, i dont think so? just thinking aloud

i get the feeling we will see many AI fear trades, my view is wait for evidence

For forever, the behaviour of all the actors in the property market has bamboozled me. I’ve almost resigned to the fact that I can’t predict consumer behaviour.

Firstly why do we pay tens of thousands of dollars to what seems from a layperson’s perspective like a professional door opener who spends most their time marketing their own mug. On the flip side though, agents themselves commonly often use another agent to sell their own house, so there must be value in them.

Secondly, how did REA group come out of nowhere decades ago to displace Domain who was once dominant with what seemed an identical product. Still to this day I maintain the experience on the sites are near identical. I asked an old time agent who worked through both periods how this happened and he said that it was purely because realestate dot com was an easier name to remember. Amazing if it’s as simple as that.

What I’ve landed on, is when it’s a decision as large as selling or buying a house, marginal things (really really marginal) matter. Like many would use rea’s site over AI for such small benefits. For example the images showing better on their phone. The point made above that there’s a fear in not finding the one qualified buyer is valid. It could cost a deal.

It’s also a peculiar industry to me based on something else. The amount of business that’s done between agents on handshake agreements. An agent told me occasionally out of nowhere he’ll get a huge cheque in the mail from someone that he referred a buyer or seller to based on a verbal agreement made years ago. He didn’t even know the cheque was coming. It seems like a slightly dodgy industry but in some ways it’s amazingly ethical at the same time.

@PhilO if I recall correctly, REA grow out of the news corp national paper chain while domain grew out of fairfax, Sydney and Melbourne based. For a long while domain held share in the wealthy suburbs of the big cities. There comes a tipping point when volumes attract volumes, the network effect, which REA got to first then it’s a natural domination. That’s my interpretation anyway

Yes. I vaguely remember something like that so it makes sense. Through it all (as far as I could remember atleast) the actual website offerings felt nearly identical. The difference perhaps was Domain trying to package offerings to its business customers with its legacy newspaper offering.

@PhilO I love the flipside aspect of your comment!

You're dead right about the margins that a good Agent can add value to a sale for the seller. Anyone can sell a property, but having a strategy which maximises competition and as a result the final sale price does take years of experience and nuance.

Choosing the right sales strategy is imperative as well. For example, I just sold a property where the successful offer was $200,000 greater than the next best offer, along with the winning offer being better terms and conditions. If we went to Auction, we would have sold for $1 more than the 2nd highest price and would have cost the seller $199,999 in forgone, after tax money.

The cost of REA is trivial in the quantum of value which can be created for an individual seller, but there's definitely circumstances in which Agents try not to use REA.

On your last point, one of the best pieces of advice I've been given as a RE Agent when I was starting was "Buyers and Sellers come and go, but the Agents will be there next week". Even though they're competitors, there are friendly competitors and hostile competitors. The friendly ones understand there's enough to go around for everyone to have their fill and are welcoming to getting the best outcomes for their clients, even if it means sharing a fee or rewarding effort.

The RE sales industry doesn't have any salaries, no annual leave, no sick leave and it is 100% eat what you kill. If you don't sell, you don't put food on the table and thus the hustling treadmill of the agents.

@OxyBBear one of the very frustrating parts of the REA AI is they are not using a hedonic algorithm. Their AI has no idea whether a house has been renovated, extended, upgraded, re-roofed, landscaped, etc etc etc. They look at things like square metre rates, accommodation and suburb casting no differentiation on value depending if it's at the end of the road with a highway or the cul-de-sac end of the road with a park. The valuations they come up with still need some serious amounts of manual updating to get any accuracy, but then that also assumes that all the data is perfect as well. Sure, for a homogenous apartment block where all the apartments are the same size, quality, age but on different levels then these computerised valuations can be fairly accurate, but there's a margin of error that is absolutely wild.

great inside scoop @Jarrahman. Cheers.

I think the idea that AI models will source directly from agents is quite unrealistic given the current economics of AI. We already have a technology that allows us to retrieve directly from real estate agents websites: Google search. Why did the real estate classifieds businesses still see significant compounding for two decades, even after people got quite sophisticated with search engine optimisation? Because the network effects are real and agents probably don't want to maintain their own technology stack.

I think AI will push the marketplace businesses to transform but will not make them redundant.

Costs & Incentives

AI costs significantly more than Google search, and AI models are not as much of a deterministic outcome as traditional search engines. Search engines are simple: they use your search terms and provide the best match to the indexed websites and they do this for a tiny cost. LLMs are a linear regression over the entirety of human knowledge, but each of the foundation models are specified and trained slightly differently so can return different results.

In the scenario where an RE agent simply posts to their website, they have no guarantee that their properties will be surfaced by the AI models unless they are using a model context protocol (MCP). So this means that real estate agents need to invest time and money to establish the MCP in the first place and then will probably have to pay each of the foundation models a fee for optimising the search... and if every real estate agent is doing this, where is the advantage? Given the cost to build the foundation models, will this really be cheaper than a centralised marketplace?

Real estate agents want to move properties fast. The foundation models have no incentive to sell your specific property without compensation to grease the wheels. RE agents pay REA to get the priority listing because they want turnover. Simply listing your property on the website and hoping for the best seems like a very indirect and passive way to sell a property. RE agents need to put off an aura that they are hustling for that juicy commission.

If you've ever used a cloud product, you have to consider your costs per query as part of your data architecture. Let's say there are 100,000 real estate agents in Australia who have all paid for special privileges. The LLMs would need to make 100,000 calls to each MCP which is highly inefficient and costly. Not to mention that the RE agents will incur cloud fees for the data they upload to their own website. There are two ways to solve this:

1. Have a centralised database to reduce the volume of queries. I believe this is the value add for the classifieds business which I will get to in a bit.
2. Make the models vastly more efficient. Could happen. Will it happen before the entire world has a paradigm shift to using AI as marketplaces? Don't think so.

AI is only as good as the data you feed it. RE agents would need to ensure they are maintaining proper data quality on their own websites to get the best results. This sounds like work that RE agents probably don't want to do.

From marketplace to data products

The classifieds sites solve all the above problems.

1. They are a centralised place for the AI models to query efficiently.
2. They provide services RE may not wish to do themselves.
3. They can provide enriched data by aggregating and creating new data for both the AI models and RE agents.

The classifieds sites will act as brokers of the data for the models, and they may charge RE agents by doing the listing on their behalf.

I could definitely see the classifieds businesses losing pricing power and margins as a result of AI, but I think they will adapt and adopt.

Maybe I misunderstand something, but keen to hear people's thoughts on this.

A good AFR article here - Is AI a threat to online property giants like REA

GCQ makes arguments too -=in this update - GCQ-Flagship-Fund-Class-P-November-24-Portfolio-Update-1.pdf

@Keyboardcat999 i think you have done a good job identifying the issues. Seems there are the AI is all BS camp, and AI will destroy the world camp. Truth is probably much more boring imo. I haven’t seen much , if any disruption, the other thing is most at risk are tech savvy and they have time, both critical to adapting. I’m viewing it from a buy opportunity, networks, enterprise software etc let’s see