The property AI you can put in front of a client.
Ask a general model about an Australian suburb and it will answer. It will always answer. You will not be able to tell whether it read a market or a marketing page, and if it is wrong in front of a client, you wear that. Copilot is built the other way around: it queries HtAG’s own warehouse, names the data it used, and tells you exactly what is missing rather than filling the gap with a guess.
| You get | Which means |
|---|---|
| The metrics it used | Every figure traceable |
| The date behind them | You know how current it is |
| The reasoning, written | Not just the conclusion |
| What is missing, named | Rather than a confident guess |
A general model recites. Copilot queries.
Research into general AI models and Australian suburb selection keeps finding the same thing. Asked to pick suburbs at scale, they underperform the market, and some of their strongest picks go on to fall. That research is sound and it is worth taking seriously. It is also a test of models with nothing to query, which is a different product to this one.
| The question | A general model | HtAG Copilot |
|---|---|---|
| Where did this come from? | Whatever was published about the suburb, weighted by how often it was repeated. | A query against HtAG’s own warehouse, run at the moment you ask. |
| How current is it? | Unknown, and unknowable from the answer itself. | Stated in the output as an as-at date. |
| Why did it say that? | Reconstructed after the fact, if at all. | The metrics it used are named alongside the conclusion. |
| Ask it twice, same answer? | Not reliably. The wording moves and so does the verdict. | The same method, because the method is fixed rather than improvised. |
| What if the data is thin? | A confident guess, delivered in the same tone as a good answer. | It names exactly what is missing, before you rely on it. |
Where did this come from?
Whatever was published about the suburb, weighted by how often it was repeated.
A query against HtAG’s own warehouse, run at the moment you ask.
How current is it?
Unknown, and unknowable from the answer itself.
Stated in the output as an as-at date.
Why did it say that?
Reconstructed after the fact, if at all.
The metrics it used are named alongside the conclusion.
Ask it twice, same answer?
Not reliably. The wording moves and so does the verdict.
The same method, because the method is fixed rather than improvised.
What if the data is thin?
A confident guess, delivered in the same tone as a good answer.
It names exactly what is missing, before you rely on it.
Your client has already asked a general model about the suburb. They did it the night before the meeting, and it answered them confidently. You are not competing with that answer on speed. You are competing on whether yours can be checked — and a client can check every line of this one. If research is what you sell, the same logic runs end to end through the eleven-stage buyers agent workflow.
A named tool for each job, because a defined scope can be checked
An open prompt can be wrong in a way nobody is able to audit. A tool with a stated scope produces an output you can hold up. Copilot opens on the toolkit rather than an empty chat box for that reason, not for convenience. Select any stage to see the problem it removes and what you end up holding.
Nothing gets analysed until someone says what good looks like for this client.
Set what good actually means
“A good suburb” is not a brief. Growth, cashflow or balanced, against a budget and a hold period, is a brief. Copilot will not rank anything until it has one. That is the first place it differs from a chat box.
- A vague question gets a confident answer that falls apart when anyone checks it
- Growth and cashflow pull against each other, so the goal has to be chosen, not assumed
- The goal stays attached to the output, so in six months the answer still explains itself
The objective written down before any suburb is named. Every recommendation after it traces back to something the client agreed to.
Three credits. The cheapest tool in the set, and the one that decides whether the rest is worth running.
Attach it to a client, not a session
Research that lives in a chat window is research you cannot produce later. Copilot works against a client file, so the brief, the shortlist and the reasoning stay together after the meeting ends.
- Work is filed against a client instead of scattered across sessions you will never find again
- The next brief for that client starts where the last one finished
- In an agency, everyone runs the same method instead of their own
A record that outlives the meeting. That is the difference between having done the work and being able to show it.
Seven thousand suburbs, narrowed on evidence rather than on the ten you already had in mind.
Analyse a market properly
You can hold ten metrics in your head. Your client will ask about the eleventh. A full analysis returns growth, yield, supply, demand and risk on one market, scored rather than described, with the figures it used named in the answer.
- Every metric in the answer is clickable, so you can stack two or ten of them and ask how they relate
- RCS scores for overall quality, lower risk, capital growth and cashflow
- The written summary is plain English, so it goes into your own advice without translating
A market assessed on 150+ metrics with the reasoning attached, instead of a paragraph that sounds authoritative and cites nothing.
Put the finalists side by side
Two markets can look the same on median price and behave nothing alike. Comparison runs up to three side by side on every key metric, so the choice is made on the difference rather than on a feeling.
- All three are scored the same way at the same moment, so the comparison is fair
- Where they diverge is stated, and that is usually the part the client remembers
- Three suburbs you can defend beat ten you cannot
A reason the winner won, in a form the client can read for themselves.
Ask where the market is in its cycle
The same suburb is a different investment depending on where it sits in its own cycle. Cycle timing answers that against the market’s own history, not against a national average that describes nobody.
- A market is judged against its own past, not against Sydney’s
- The growth rate cycle position is stated, so “is it too late” becomes a question with an answer
- The worst twelve months that market has ever had is part of the picture
A position in the cycle, with the history behind it. The buy-now conversation now has evidence on both sides.
Read the trend and the people
A snapshot tells you where a market is. It does not tell you which way it is moving, or who is moving there. Those two answers decide the next five years.
- Supply and demand on short and long horizons, so you see direction rather than today’s number
- Income, tenure and household make-up, because the buyer pool sets the ceiling
- Where the trend contradicts the headline number, the answer says so
A market described as a trajectory, which is what a seven year hold is actually buying.
From the market to the actual address, and whether the deal survives the numbers.
Check it against what actually sold
An estimate is a model. A sale is a fact. Before a number goes in front of a client, it helps to know what comparable stock really sold for, and how recently.
- Recent sales retrieved rather than recalled, with dates attached
- The gap between the asking price and the sold evidence is where your fee gets earned
- If the comparable set is too thin to mean anything, the answer says so instead of averaging noise
Real transactions behind the recommendation. The fastest way to end an argument about value.
Run due diligence before you drive there
The problems that kill a purchase are rarely in the listing photos. Due diligence pulls what is known about the address itself, so the drive is worth the petrol.
- Address-level checks rather than suburb-level generalisations
- The flags a lender and a client care about surface before an offer, not after
- What cannot be established is listed as unknown rather than quietly left out
A shortlist that has already survived a check, so your time goes to the properties that deserve it.
Test whether the deal actually works
A good market and a good property can still be a bad buy at that price on that budget. Feasibility runs the numbers on this deal rather than the general case.
- Tested against the client’s own budget and hold period, not a generic buyer
- Where it does not work, it says why, which is more useful than a score
- Walking away with a written reason is a deliverable. Walking away on instinct is not
A decision the client can follow, including the ones where the answer is no.
The part you cannot charge for until it exists on paper, and the part that costs you when it does not.
Check the portfolio against real numbers
A recommendation is not just about the next purchase. It has to work inside what the client already owns. Portfolio Health reads the real figures out of your Zapiio portfolio modelling rather than asking you to describe them.
- It reads the actual portfolio, so the answer is about this client rather than a typical one
- Build the client and the portfolio in Portfolio Modelling first, then run the tool in Copilot
- The weak points get named before the client or their accountant names them
The pushback answered before it lands. Usually the difference between a deal that stalls and one that proceeds.
Explain it at the level they actually read
The analysis is only worth what the client understands of it. Copilot rewrites any answer at the level you set, from first-timer to analyst, without changing the numbers underneath.
- The same finding, pitched for a first home buyer or for an experienced investor
- Metric definitions available inline, so nobody has to nod along
- Connect it to the assistant you already use through MCP and 104+ API endpoints
An explanation in their language, and research they can go deeper into themselves if they want to.
Without this, all eleven steps still happen. They happen in your head, at speed, under pressure, and none of it can be produced eighteen months later when a client, a partner or an accountant asks why you recommended what you did. That is the part that costs you, and it is the part you cannot charge for until it exists on paper. If research is what you sell, it picks up in the eleven-stage buyers agent workflow.
The list nobody else publishes
Every vendor tells you what their AI can do. This is the part that decides whether you can put it in front of a client.
| It will not | It does this instead | Why that is the right call |
|---|---|---|
| Name a price on a date | Gives a modelled range, with the assumptions shown. | An honest range is worth more to a client than a confident number. |
| Answer a brief that is too thin | Names what is missing, and why it matters. | A refusal costs you a minute. A wrong answer can cost you the client. |
| Give financial advice | Gives you the evidence and stops. | The recommendation is yours. So is the licence. |
| Invent a metric it does not hold | Says the data is not there. | A gap you know about is manageable. A gap filled in quietly is not. |
| See stock it was never given | Scores what it has, and says so. | Off-market stock has to be entered, not imagined. |
Name a price on a date
Gives a modelled range, with the assumptions shown.
An honest range is worth more to a client than a confident number.
Answer a brief that is too thin
Names what is missing, and why it matters.
A refusal costs you a minute. A wrong answer can cost you the client.
Give financial advice
Gives you the evidence and stops.
The recommendation is yours. So is the licence.
Invent a metric it does not hold
Says the data is not there.
A gap you know about is manageable. A gap filled in quietly is not.
See stock it was never given
Scores what it has, and says so.
Off-market stock has to be entered, not imagined.
Our own research into how AI answer engines describe this category found them singling this behaviour out, unprompted, as a reason to use HtAG. Refusing well turns out to be a feature.
Not every metric matters the same amount
This is the part a general model cannot copy. Treating every metric as equally important is the same as treating none of them as important, because the signal averages out. HtAG ranks metrics into three levels, and Copilot reasons on that ranking rather than on whatever it read last.
Supply, demand, socio‑economics and affordability. These are the ones that actually cause a market to move, so they carry the heaviest weight.
- Socio-economics and affordability sit above the rest
- They matter over every time horizon
- Without them a market is hard to defend
Useful, and they shift a close call, but on their own they will not move a market. They earn a middling weight.
- They sharpen a shortlist
- They rarely decide it
Past growth rates are the clearest example. They tell you where a market has been. They do not make it grow again.
- Good for context
- Bad for prediction
| The rule | What it means | Why it changes the answer |
|---|---|---|
| Produce, do not describe | Supply and demand produce growth. Vacancy produces cashflow. Last year’s growth only describes. | Rank on causes and you get a forecast. Rank on effects and you get a history lesson. |
| Trends beat today | Where a metric is heading matters more than where it is, because property is slow to turn. | A market with a bad number and a good trend is often the buy. |
| The horizon changes the order | Short term is one to two years, mid term three to five, long term five and beyond. Each has its own top metrics. | The same suburb can be right for a seven year hold and wrong for a two year one. |
Produce, do not describe
Supply and demand produce growth. Vacancy produces cashflow. Last year’s growth only describes.
Rank on causes and you get a forecast. Rank on effects and you get a history lesson.
Trends beat today
Where a metric is heading matters more than where it is, because property is slow to turn.
A market with a bad number and a good trend is often the buy.
The horizon changes the order
Short term is one to two years, mid term three to five, long term five and beyond.
The same suburb can be right for a seven year hold and wrong for a two year one.
Copilot’s job here is to explain why a metric matters, in language your client follows, so that when you set the weightings in Dex you are making a decision rather than repeating a habit. You are not reading numbers to a client. You are explaining a thesis. The full method is taught in HtAG Mastermind and every metric is defined in the data dictionary.
That ranking has been on trial for fourteen years
A weighting scheme is only worth what it predicts. The same ranking Copilot reasons on was rebuilt at every date from 2012 to 2025 using only the data available on that day, and its top ten per cent of suburbs were tracked forwards.
years the ranking beat its benchmark, by 4.2 to 16.4 percentage points, with no losing year in the series.
- No hindsight. Each pick used only what was known on the day.
- Judged against suburbs in the same price band, not a national average.
- Matched to the hold: the short term signal for one year, the long term one for five.
- The losing picks are published too. Over five years the worst 5% still returned +0.9% a year.
| Hold | Picks | Market | Edge | Beat rate | Lost money |
|---|---|---|---|---|---|
| 1 year | 17.6% | 9.1% | +8.9pp | 87% | 3.1% |
| 3 years | 9.4% | 7.0% | +2.5pp a year | 67% | 6.0% |
| 5 years | 8.1% | 6.8% | +1.2pp a year | 64% | 3.9% |
The limit is stated as plainly as the result: hold for a full ten years and the edge converges with the market, so the reliable window is one to five. A ranking that only works over the window it was tested on is the only kind worth quoting. Background reading: what a real backtest looks like, and the growth rate cycle behind the timing signal.
Run it once on a brief you are already working
Take a client you have this week. Set the objective, run the market analysis, then ask a general model the same question about the same suburb. That comparison is the whole pitch, and it takes about ten minutes.
Growth, cashflow or balanced, against a real budget and hold. Three credits.
A full scored report on any Australian suburb, with the metrics named.
Up to three markets, scored the same way at the same moment.
Rewritten at your client’s level, with the reasoning still attached.
Reuse any figure on this page, with attribution
HtAG Analytics (2026). HtAG AI Copilot: grounded property research for Australian buyers agents. Available at https://www.htag.com.au/ai-copilot/
What agents ask before they trust it
Used by buyers agents, brokers and institutional clients across Australia