$97,500 under what the agent was pushing, bought before it ever launched

J and B came through a referral from a relative already on the books. He builds high-end homes, she teaches, and they had two young children and a build of their own underway in an expensive pocket of Sydney's north-west.

That build was the context for everything. They needed an investment that would offset two incomes immediately, not one that would come good in a decade. Dual income was the brief: a house with a compliant granny flat, so two tenancies serviced one mortgage.

Ceiling was $1.5 million. Nil flood, no fire zone, and land with a growth profile they could scale from later.

Twenty-eight properties assessed, fifteen physically inspected. The target was a property in Stella Street: a 1952 front house with a two-year-old granny flat behind it, always tenanted, on 676sqm.

Then the valuation problem. This property broke the data three separate ways. The automated models don't differentiate a granny-flat property from a plain house, so they undercount the second income entirely. The dataset couldn't cleanly split the land allocation between the two titles. And the outputs disagreed violently: the major providers returned $1.39M, $1.5M and $1.6M on the same asset, with a broad algorithmic range of $1.06M to $1.6M.

A half-million-dollar spread is not a valuation. It's noise.

So we abandoned the models and valued the land directly. The growth-adjusted comparable rate in that pocket was about $2,120 per square metre. On 676sqm, that implied roughly $1,433,000. Assessed fair value came out at $1,435,000, on a granny-inclusive basis.

Two risks then had to be closed before any offer went in. Was the granny flat legal, and was the income real? We chased and confirmed the Occupation Certificate, verified compliance, and had the lease and tenancy agreements attached to the contract. The rent was under market, which was upside rather than a problem.

The agent was guiding $1.4M to $1.5M while privately pushing $1,525,000, and said he was 85 to 90 percent confident of getting above $1.43M. The vendor wanted to launch publicly in about a week. Several developers had already walked the block for land value, and a third-party offer was live.

The first offer at $1,350,000 was rejected. Then $1,383,500. Then $1,402,500. Each step was anchored to the land rate, not to the agent's number.

Secured off-market at $1,427,500 — $2,112 per square metre, effectively exactly the comparable land rate and just under assessed fair value. That is $72,500 under budget and $97,500 under what the agent was pushing for.

It exchanged before the public launch, which meant the circling developers never got to bid the land up.

"Really happy, good buying. We will renovate that in a few years. The granny should be a good offset and income, in a few years re-develop. Really good job, we are really happy. We will settle this one down and then think about the next steps."

What decided it: refusing to negotiate against the agent's number. Every offer was justified against dollars per square metre of land, which is a figure that can be defended in a room. The agent's $1,525,000 never had that behind it.

The wider lesson: automated valuations are close to useless on dual-dwelling properties, they cannot see the second income and they cannot split a shared title. When the models disagree by half a million dollars, that is the signal to value the land yourself. And always confirm the Occupation Certificate before treating granny-flat income as real. An unapproved second dwelling is not an asset, it is a liability with a tenant in it.

J and B came back for a second purchase at Blue Haven the following year.