How we score · Suburbs

Buyer fit and confidence

Every suburb profile carries buyer-fit scores for five household types and a confidence rating. They sit next to each other and look like a pair, but they are built in completely different ways — and the confidence rating is not about the buyer-fit score. This page explains both, including the parts that are easy to misread.

TL;DR

  • Buyer fit is a judgement, not a calculation. An AI reads the suburb’s Census data and rates it — there is no underlying formula.
  • Confidence grades our data, not the suburb. It measures how complete the Census extraction was — never whether the buyer-fit number is right.
  • High confidence + a low buyer-fit score is a normal, trustworthy combination — the two numbers are not connected the way the layout suggests.
  • See the coloured bands below — Low confidence is a prompt to verify anything that matters, not a warning about the place itself.

1. Where buyer fit comes from

The buyer-fit number is a judgement, not a calculation. We give a language model the suburb’s ABS Census data and ask it to rate the suburb out of 100 for each household type, and to justify each rating with specific figures. There is no formula underneath. We would rather say that plainly than present a judgement dressed as arithmetic.

The five ratings and the Census material behind each:

RatingWhat the model is looking at
FamiliesHousehold composition, school-age population, dwelling sizes, ownership.
ProfessionalsCommute patterns, employment and industry mix, education levels.
First-home buyersPrice levels against incomes, rental share, dwelling types.
InvestorsRental share, weekly rents against prices, population direction.
RetireesAge profile, ownership outright, household size, health access.

Why not a formula

Because “good for families” has no objective weighting. Any formula we published would be our opinion about how much school access matters against block size against commute — expressed as arithmetic, which would make it look more neutral than it is. A stated judgement, with its reasons attached and the figures cited, is more honest and more useful than a spurious equation.

What we do check

Each rating is validated before it is stored: the score has to be a number in range, and it has to arrive with at least a minimum count of supporting reasons. Ratings failing those checks pull down the suburb’s confidence rating. Note what this is — a check that the output is well formed, not that it is correct. Nothing verifies that 75 was the right number.

2. Reading a buyer-fit score

Three habits make these useful rather than misleading.

Do this

Read the reasons first

Every rating carries four or five reasons citing real Census figures — “43.4% hold bachelor degrees against 29.2% for Victoria”. Those figures are checkable and are the actual finding. The score is a summary of them, and the weaker half of the pair.

Do this

Compare across the five, in one suburb

A suburb rating 80 for retirees and 45 for first-home buyers is telling you something real about who the place suits. That contrast is more reliable than either number on its own, because both came from the same reading of the same data.

Not this

Don’t split hairs between suburbs

A 72 in one suburb and a 75 in another is not a meaningful gap. They were generated independently and there is no calibration between them. Treat differences under roughly ten points as noise, and look at the reasons to see whether the two are actually different.

3. What confidence measures

Confidence is calculated, and it measures the quality of the data behind the page. It is assembled in two halves.

Half one — the Census data

How complete and sound the ABS extraction was for this suburb:

  • Completeness40%How many of the expected ABS fields actually came through.
  • Validation30%Whether values pass range checks — a median age of 3 fails.
  • Consistency20%Whether related figures agree with each other.
  • Outliers10%Whether values sit inside the range seen across Victoria.

Not every field counts equally. Core fields — population, median age, income — weigh three times a merely useful one, and six times an optional one. A suburb missing its median income is penalised far harder than one missing a minor category.

Half two — the written insights

How complete and specific the generated profile turned out:

  • Input quality35%The demographics score above, carried forward.
  • Completeness30%How many of the expected insight fields were produced.
  • Specificity20%Whether the text cites actual figures rather than generalities.
  • Format validity15%Whether the model returned clean structured output first time.

Note the first row: insight confidence inherits the Census score. Poor source data caps the confidence of everything written from it, however polished the writing.

The four bands

BandScoreWhat it tells you
High85 and aboveNear-complete Census coverage, values passing every check, and clean, specific insight output.
Medium70–84Good coverage with some gaps, or insights that are complete but less specific.
Fair50–69Noticeable gaps. Read the profile, but expect some sections to be thin or absent.
LowBelow 50Sparse underlying data. Treat everything on the page as indicative and verify anything that matters.

4. What confidence does not mean

High confidence does not mean the buyer-fit score is right. The two are not connected in the way the layout suggests. Confidence says the Census data was complete and the profile came back well formed. It says nothing about whether rating this suburb 75 for families was a sound call.

Three more things it is not:

  • Not a rating of the suburb. A quiet, well-documented suburb can score high confidence and still be wrong for you. Confidence is about our inputs.
  • Not a measure of how current the data is. Census data is from 2021 everywhere. A high-confidence profile is as old as a low-confidence one.
  • Not comparable to Discover’s scores. Different system, different scale, different meaning. See how Discover ranks properties.

Read it as a completeness warning, which is what it is. Low confidence is a prompt to verify anything that matters before relying on it. High confidence removes one source of doubt, not all of them.

5. Limits

Census data is from 2021 and a suburb can change a great deal in a few years. Where we have more current figures — sale prices, school zones — they are labelled with their own dates.

Suburb averages hide streets. Every figure here describes a whole locality. A suburb can be affordable on average and unaffordable on the street you want.

A language model wrote the prose and the ratings, from real data, and it can still be wrong. Every claim it makes should carry a figure you can check — that is the point of the design, and if one does not, that is a bug.

None of it is advice. These pages describe places; the decision is yours. See our terms.

Found a rating that does not match its own reasons? Tell us — that one is genuinely a bug.

Data sources

Demographic figures from the Census of Population and Housing © Australian Bureau of Statistics, used under Creative Commons Attribution.

Narrative summaries are AI-generated from the underlying data and may contain errors.

Sale figures are aggregated from recorded transactions and shown only as medians and counts.

Informational only — not financial, legal or planning advice. Verify zoning, overlays and catchments with the relevant council before relying on them. Data vintage: 2021 Census.