Every couple of months a new AI-powered “buyer matching” platform lands in my inbox promising to change how agents qualify and connect with prospective buyers. I’ve been skeptical for a while because most of them seemed like glorified email automation with a model sticker on top. So in April I committed to actually testing four of them properly across my live Melbourne listings.

Here’s what I found, what worked, what didn’t, and what I’d actually pay for.

How I tested them

I ran each platform across 11 active listings spanning $580,000 to $2.4m, three suburbs, four property types. I tracked:

• Quality of matched buyers (subjective rating 1-5 based on fit, finance position, timeline)
• Time saved versus my baseline workflow
• False positive rate (matches that wasted my or the buyer’s time)
• Conversion to inspection
• Whether I’d recommend it to another agent

I’m not naming the four platforms publicly because the contracts I signed have evaluation clauses, but I’ll describe what each one does and whether it actually delivered.

Platform A: Behavioural matching from portal data

This one ingests anonymised browsing patterns from a major portal and tries to match active buyer profiles against my listings based on viewing behaviour, saved searches, and inquiry history.

What worked: It surfaced a handful of buyers I genuinely wouldn’t have known were active. For one Brunswick listing it identified six active buyer profiles within 48 hours of the listing going live, three of whom converted to inspections.

What didn’t: The “match score” felt arbitrary. A buyer rated 92% match was identical in every visible signal to one rated 67%. When I asked about the model, the response was vague.

Verdict: Useful for upper-funnel discovery. I’d pay $200/month for it if I were a high-volume agent. Less compelling for a 30-listings-a-year agent.

Platform B: Conversational AI qualification

This is essentially a chatbot that talks to inquiry leads on your behalf, asks qualifying questions, and books inspections. It plugs into my CRM.

What worked: Response time. Inquiries got a substantive reply within 90 seconds at any hour. For after-hours leads especially, this matters. The qualification questions were sensible.

What didn’t: It missed nuance constantly. A buyer mentioned she was selling a property to fund the purchase and the bot still pushed a finance pre-approval question that confused her. Two leads I’d have called immediately got handled by the bot in ways that cost me the opportunity.

Verdict: Promising direction but not quite there for prestige work. For high-volume mortgage-eligible inquiry on first-home stock it might be fine. For complex buyers I want a human in the loop.

Platform C: Off-market buyer database matching

A more enterprise platform that maintains a curated database of qualified buyers (verified finance, signed brief, exclusive engagement) and matches them against listings before they go to market.

What worked: When it matched, it really matched. Three properties got serious interest from properly qualified buyers before public launch. One sold off-market for above the expected campaign reserve.

What didn’t: Coverage. The buyer database is concentrated in particular price brackets and suburbs. For my listings outside that sweet spot it was useless.

Verdict: Genuinely valuable for the right segment. The economics make sense for prestige and unique stock where one match is worth a campaign’s worth of marketing spend.

Platform D: Lookalike modelling on past sales

This one analyses your past buyer database and tries to model who’s most likely to buy similar future stock. It then identifies similar profiles in the broader population.

What worked: Conceptually interesting. It surfaced patterns in my past buyers I hadn’t consciously noticed (lots of buyers from a particular adjacent suburb, an over-representation of buyers in certain professional clusters).

What didn’t: The “go find similar people” part was much weaker than the analysis. It identified profiles but had no way to actually reach them efficiently. The output was insight without action.

Verdict: Useful as analytics, not as a buyer-acquisition tool. I’d recommend it more to agencies thinking about marketing strategy than to individual agents looking for leads.

What the experiment taught me

A few broader observations after a month of this:

The model is not the moat. Several of these platforms are using the same underlying AI capabilities. The differentiator is the data they have access to and the workflow they fit into. A great model with poor distribution helps no one.

Human judgment still wins on edge cases. Every platform handled the obvious cases fine. They all stumbled on buyers with non-standard situations - separated couples splitting equity, expat returnees with foreign income, downsizers selling unencumbered, trust structures. Those are exactly the cases where good agents earn their commission.

Speed matters more than people admit. The biggest single win across all four platforms was just responding to inquiries faster. Whether that’s AI or a virtual assistant or just better CRM workflow, the lesson is the same. Slow response loses deals.

Be careful what you outsource to a model. A few firms I respect think hard about which decisions stay with humans and which are safely automated. Working with a thoughtful AI strategy advisor on the agent side has helped me think about this more clearly than the platform vendors do, since they have an interest in expanding what gets automated.

What I’m actually using going forward

After the trial I’m continuing with one platform (Platform C, the off-market buyer database) and dropping the other three. Platform A I might revisit if pricing changes. Platform B I’d want to see meaningful improvement on conversational nuance. Platform D I’d recommend to agencies, not individual agents.

The broader PropTech sector is putting out a lot of noise right now and a fair amount of substance. The trick is testing properly before you commit. A two-week trial with three live listings teaches you more than ten vendor pitches.

If you’re considering any of these tools and want to compare notes, I’m genuinely happy to swap observations. Drop me a line and let’s compare what’s working in your patch versus mine.