I Tested a $33 AI Tool Against Our Buyer's Agent Team. The Results Were Uncomfortable.

I'm about to tell you something that could put me out of a job. Or at least, that's what some of my competitors think when they see what AI can do now.

Last month I sat down with a $33-per-month AI tool and fed it a single property address. Ten minutes later, it spat out a valuation report that covered comparable sales, council planning overlays, school catchments, vacant land transactions, commercial activity, and even REIV median data for the suburb. My team of five — people who drive 30,000 kilometres a month inspecting properties — would normally take two hours to compile something equivalent.

The tool got the valuation range right. It estimated $1.55 million to $1.65 million for a 900-square-metre block in Ferntree Gully, with an optimistic ceiling of $1.80 million. Our internal estimate was $1.60 million to $1.70 million. Close enough to make me uncomfortable.

But here's the thing most people will miss: knowing the price is only about 20% of the job. The other 80% is what the AI cannot see from its desk.

What the AI actually did in ten minutes

The tool I tested uses what Google calls "deep research" — it doesn't just search the web. It opens dozens of tabs simultaneously, reads planning documents, parses government databases, cross-references sold prices from Domain, CoreLogic, and agent websites, then synthesises everything into a structured report with citations.

For this particular property — a 900-square-metre block with older improvements — the AI pulled data from the listing agent's past sales history, the previous time this property was marketed (three years ago, different agent, different price), HTAG vacancy rate data for the specific street, council planning scheme overlays, nearby commercial activity, and comparable vacant land sales within 800 metres.

It even found the REIV quarterly median price for Ferntree Gully and calculated where this property sat relative to the distribution. The reference list at the bottom of the report ran to over 40 sources 1.

I won't pretend that wasn't impressive. Because it was. My team needs two hours and five people to cover that same ground. The AI did it while I made a coffee.

Where the AI got it right

The valuation range was solid. $1.55M to $1.65M with upside to $1.80M. Our team's assessment was $1.60M to $1.70M. The overlap is significant.

The AI correctly identified the 30-metre frontage as a rare asset — wide frontage on 900 square metres gives genuine subdivision potential in General Residential Zone areas. It flagged the school catchments, the walkability to public transport, and the proximity of the Boronia junction commercial strip. It even noted the suburb's current infrastructure pipeline 2.

Rental yield assessment was conservative but reasonable. The AI estimated $550 to $600 per week for the existing dwelling. We'd push that to $650 after a light renovation, but the baseline figure was defensible.

The development potential analysis was particularly good. The AI cross-referenced the planning overlay with vacant land sales nearby and correctly concluded that the land component alone was worth approximately $1.1M to $1.2M — meaning the existing house was being valued at roughly $400K to $500K as a structure. That's an important insight for anyone thinking about a knock-down-rebuild or subdivision strategy 3.

Where the AI fell over completely

And here's where I stop worrying about my job security.

The AI cannot walk the block. It cannot feel the slope. This matters because every metre of slope on a residential block adds approximately $50,000 to construction costs if you're building. A photo from the street shows a flat-looking property. Walking it reveals a 1.5-metre fall from front to rear. That's $75,000 in additional costs that the AI has no mechanism to detect 4.

The AI cannot check the Section 32. It cannot read the vendor's statement to identify easements that run through the centre of the block (killing subdivision potential), or a restrictive covenant that limits the site to a single dwelling. We've walked away from properties that looked perfect on paper because the s32 contained a single sentence that made the entire investment case collapse.

It cannot smell dampness in the subfloor. It cannot hear the highway noise from the backyard. It cannot notice that the fence line doesn't match the title boundary — which, in our experience across 350-plus transactions, happens on about one in twelve established properties.

And critically, it cannot negotiate. The AI told me the property was worth $1.55M to $1.65M. Great. But we regularly buy properties for 5% to 12% below what even our own valuation says they're worth. On a $1.6M property, that's $80,000 to $192,000 in savings — achieved through reading the vendor's motivation, timing the offer, structuring settlement terms, and leveraging our relationship with the listing agent 5.

A tool that tells you the price is useful. A team that buys below that price is valuable. The distinction matters.

How buyers should actually use this

I'll be blunt: if you're buying a property under $800,000 in a straightforward suburb with no development ambition, an AI valuation tool might genuinely save you money. You don't need a buyer's agent to tell you that a three-bedroom brick veneer in Cranbourne is worth $620,000 when Domain shows you fifteen comparable sales within 500 metres.

But if you're buying with development intent, if you're targeting off-market properties, if the property has any complexity — slope, overlay, heritage, dual occupancy potential — then the AI gives you a starting point, not an endpoint.

Here's my recommendation for buyers: run the AI report before your first inspection. Use it to set your baseline expectations. If the agent's price guide is 20% above the AI's estimate, ask why. If it's 15% below, figure out what the AI missed (or what the agent is hiding).

For sellers: run the AI report before you sign an agency agreement. If three agents tell you your house is worth $1.8M and the AI says $1.5M, someone is managing your expectations upward to win the listing. In Melbourne, this happens constantly. Agents quote high to win the listing, then spend six weeks "managing your expectations" downward until you accept a lower offer. The AI report gives you a second opinion that costs $33 instead of $3,300 6.

I've seen sellers sign with an agent who quoted $1.9M, only to sell at $1.55M four months later. If they'd spent $33 on an AI report first, they'd have known the $1.9M quote was fantasy.

The off-market problem AI can never solve

About 30% of the properties we purchase for clients never appear on REA or Domain. They come through our agent network — relationships built over years and hundreds of transactions. The listing agent calls us before the property goes public because they know we settle fast, we don't waste time, and we bring unconditional offers when the vendor needs certainty 7.

AI cannot access these properties. It cannot build relationships with agents. It cannot earn the phone call that says "I've got one coming up next week, do you want to see it before it goes live?" Our Hampton Park purchase at $590,000 — now valued at $670,000 with rent at $850 per week — was an off-market deal. The vendor was in financial distress. No AI tool would have found it because it was never listed 5.

This is the fundamental limit of any technology-driven approach to property buying. The best deals aren't on the internet. They're in someone's phone.

A real test: what I fed the AI and what came back

I want to walk you through the actual test, because the details matter.

The property was a 900-square-metre block in Ferntree Gully with original 1970s improvements. Listed for auction with a guide of $1.3M to $1.4M — which, if you know Melbourne auctions, means the vendor expects north of $1.5M. I copied the address into the AI tool and typed: "I'm the owner of this property. What do you think it's worth?"

Ten minutes of processing. The tool opened roughly 45 web sources. It read the listing on REA, pulled the agent's previous sales record from their website, found the property's sale history on Domain (it last sold twelve years ago for $680,000), accessed HTAG data for the street-level vacancy rate, reviewed Whitehorse Council's planning scheme, checked nearby commercial lease rates, examined vacant land transactions within 800 metres, and even dug into REIV's quarterly report for the suburb's median price movement.

The output was a 2,500-word report with a valuation range of $1.55M to $1.65M, a rental estimate of $550 to $600 per week, a development potential assessment noting the block could accommodate a dual-occupancy or three-lot subdivision under GRZ, and a risk section highlighting the property's proximity to a main arterial road and the age of the plumbing infrastructure.

I read it twice. The first time with grudging admiration. The second time looking for mistakes. I found three: it overestimated the likely rental income for the existing dwelling by about $50 per week, it didn't account for the property's actual condition (which you can only assess by walking through), and it assumed council approval for subdivision would be straightforward when Whitehorse has actually been tightening its planning controls recently. But those errors were marginal. The core analysis was sound.

The $50,000 question: when does professional help actually pay for itself?

Let me give you a framework for deciding whether you need a buyer's agent or whether the AI report is sufficient.

If your purchase ticks all five of these boxes, you probably don't need us:

1. Price under $800,000
2. Established suburb with high transaction volume (lots of comparables)
3. No development intent — you're buying to hold as-is
4. Standard residential zoning with no overlays
5. You're comfortable attending auctions and negotiating directly

If any of these apply, the AI report won't save you:

• You want to subdivide, build a granny flat, or convert to multi-tenancy
• The property has slope, easements, or heritage considerations
• You're targeting off-market properties that don't appear on any portal
• The purchase is in a trust, SMSF, or complex ownership structure
• You need renovation scoping and builder coordination post-purchase

The break-even calculation is straightforward. Our fee is roughly $12,000 to $15,000. If our negotiation saves you $30,000 (which is below our average), the service paid for itself plus put $15,000 to $18,000 back in your pocket. If we source an off-market deal worth $50,000 to $80,000 more than you paid — which happens regularly in our 350-plus transaction history — the return on the fee is 300% to 500% in the first year.

The AI tool costs $33 per month. Use it regardless. Use it before every inspection. Use it to cross-check agent claims. Use it to build your baseline knowledge. Then decide whether the execution gap — the physical inspection, the legal review, the negotiation, the network — is worth professional help for your specific situation.

I've made myself comfortable with this position: information should be cheap and widely available. Execution should be worth paying for. The AI makes information cheaper. That's progress. It doesn't make execution less valuable. If anything, it makes the distinction clearer.

What this means for the industry

I think lazy buyer's agents should be worried. The ones who charge $10,000 to run a comparable sales analysis and send you a PDF — yeah, a $33 AI tool does that better and faster.

But the agents who actually walk properties, who read s32 documents line by line, who have builder contacts that can quote a renovation within 48 hours, who maintain agent networks that deliver off-market stock, who negotiate with the understanding that every vendor has a different pressure point — those agents aren't going anywhere.

The AI makes the information gap smaller. That's good for consumers. But it makes the execution gap more visible. And execution — buying the right asset, at the right price, with the right renovation plan, and managing it properly afterward — is where value is created.

Our team manages properties at a ratio of one property manager to fifty properties. The industry average is one to one hundred and seventy. AI can help you find a property. It cannot manage your tenant, fix your plumbing at 2 AM, or take a non-paying tenant to VCAT 8.

Use the AI. Seriously, use it. It's $33 and it's genuinely good. But don't confuse information with execution. They're different skills, and only one of them makes you money.