Two Valuation Firms Inspecting the Same Property Agree Within 10% Just 78% of the Time. The AVM Managed 77% — and RG 273 Never Mentions Valuations at All
Cotality benchmarked about 1.17 million lending records to see how digital valuations stack up against a full inspection. The answer reframes a conversation brokers have most weeks — but only once you read the price bands.
- The question behind “can we get a proper one done?”
- What Cotality actually measured
- The benchmark has its own error band
- Where the method genuinely matters: the price bands
- What the dataset does not cover
- The compliance line: RG 273 is silent on valuations
- Where an 8% variance actually bites
- The 2027 valuation wave
- What to review this week
A valuation lands short. The client asks whether you can get “a proper one” done. Underneath that question sits an assumption almost nobody tests: that the human inspection is the accurate number and the digital one is the approximation. A dataset released on 19 November puts that assumption against about 1.17 million lending records.
The question behind “can we get a proper one done?”
Every broker has had the call. The automated valuation on a refinance comes in below what the client believes the property is worth, the LVR tips over a band, and the pricing or the LMI premium changes with it. The client wants a valuer to physically attend. You order the upgrade, or you escalate, or you move the file to a lender whose panel will.
That reflex is reasonable. It is also, until now, largely untested in public. Cotality’s Valuation Accuracy Benchmarking Report, summarised in a research paper published on 19 November 2026, is the first dataset this size to compare how close digital and hybrid valuations land to a full inspection of the same property — and, critically, how close two full inspections land to each other.
The findings will not change who orders the valuation. That remains the lender’s call, governed by its own panel arrangements and credit policy. What the findings should change is how confidently you explain the number to a client, and where you spend your escalation energy.
What Cotality actually measured
The study drew on approximately 1.17 million eligible records — 1,170,800 in the published table — from anonymised lending activity over a rolling two-year dataset. Each digital or hybrid assessment was matched to a full inspection valuation of the same property completed up to two years before or after it, with values adjusted to the benchmark inspection date using Cotality’s AVM postcode medians index for houses or units.
Three Cotality products were tested against that benchmark: the IntelliVal automated valuation model (AVM), Cotality Desktop, and SMARTval. The report also compared separate full inspections of the same property carried out by different valuation firms.
Two methodology points matter before any broker quotes a figure from it. First, as the report states plainly, “The comparison is with a full inspection valuation, rather than a subsequent sale price.” This is a study of agreement between methods, not a study of whether any method predicts what a property later sells for. Second, off-the-plan assessments, “as if complete” assessments and multiple dwellings on one title were excluded.
A note on what we read. The figures below come from Cotality’s public research summary. The underlying Valuation Accuracy Benchmarking Report sits behind a request form, so the detailed data tables in the full report have not been reviewed here. Every figure quoted in this article appears on the public page.
The benchmark has its own error band
Here are the national results, as published.
| Method | Sample | Within ±5% | Within ±10% | Within ±15% |
|---|---|---|---|---|
| IntelliVal AVM | 576,100 | 50% | 77% | 88% |
| Cotality Desktop | 230,800 | 48% | 76% | 88% |
| SMARTval | 48,800 | 53% | 80% | 91% |
| Full inspection | 315,100 | 51% | 78% | 89% |
Read the bottom row again. When two separate valuation firms each sent a valuer to physically inspect the same property, their figures landed within 10% of each other 78% of the time. Which means that roughly one time in five, two qualified humans who both walked through the same house disagreed by more than 10%.
The AVM, working from data alone, managed 77%.
Cotality is direct about why: “Full inspections provide the benchmark, but two valuers can inspect the same property and arrive at different values.” Their choice of comparable sales and their judgement about the property’s features both move the result. At ±5% and ±15% the picture is the same — the report notes “just three to five percentage points separating the highest and lowest figures at each.”
This is the single most useful thing in the report for a broker, and it is not the headline most coverage led with. A valuation is not a measurement with a right answer that one method finds and another misses. It is an estimate with an error band, and the error band on the human benchmark is roughly the same size as the error band on the model.
One caveat the report volunteers and brokers should carry with them: SMARTval posted the strongest numbers, but on the smallest sample at approximately 48,800 assessments, and Cotality says it “cannot be concluded that SMARTval is inherently more accurate” or that it would outperform the other methods on every lending application. Do not go into a client conversation naming a winner.
Where the method genuinely matters: the price bands
The national averages flatten something important. Break the dataset by price and the methods separate — in both directions.
Between $500,000 and $1 million, the choice barely matters. Every method scored 80% or better within ±10%, with the single exception of Desktop at 79% in the $500,000–$700,000 bracket. These two bands hold 57% of the dataset, approximately 670,700 assessments. For the bulk of residential files written in this country, escalating from a digital valuation to a full inspection is buying you a difference measured in one or two percentage points of hit rate.
Above $1.5 million, the gap opens. In the $1.5–$2 million band, the AVM landed within ±10% just 64% of the time, against 70% for Desktop, 73% for full inspections and 74% for SMARTval. That is a nine-percentage-point spread between the AVM and the physical inspection — a materially different proposition from the national average. The $1–$1.5 million band sits between the two worlds: all four methods still cleared 70%, with SMARTval and full inspection both at 77%.
At the bottom, everything struggles. In the $200,000–$300,000 band, the AVM hit ±10% only 47% of the time — but the full inspection managed just 54%, and SMARTval 58%. The sample here is small (about 7,000 assessments across all methods) and this end of the market skews regional and to units, where comparable sales evidence is thinner for a model and a valuer alike. The lesson is not “order an inspection”; it is that sub-$300,000 security carries valuation uncertainty whatever method you use, and your file should anticipate it.
The practical reframe: method choice is close to irrelevant in the $500,000–$1 million core and meaningful at the top and bottom tails. If you are going to spend credibility escalating a valuation, spend it on the $1.6 million purchase, not the $720,000 refinance.
What the dataset does not cover
Worth knowing before you lean on it. The IntelliVal AVM and Cotality Desktop figures are limited to properties with an upper threshold of $2 million, and the published price-band table stops at the $1.5–$2 million bracket. SMARTval and full inspection were drawn from houses and units valued between $200,000 and $5 million.
So if you are writing a $3 million purchase, this research says nothing about your file. Given that the clearest trend in the data is accuracy degrading as value rises above $1.5 million, the absence of evidence above $2 million is not a reason for comfort.
Cotality also notes that the price-band analysis “alone does not establish which method is appropriate for an individual property,” pointing to the property’s specific characteristics, available comparable sales evidence and the lender’s own risk requirements.
The compliance line: RG 273 is silent on valuations
Here is something that surprises brokers who assume the best interests duty reaches everything. ASIC’s Regulatory Guide 273, Mortgage brokers: Best interests duty, issued in June 2020, does not mention property valuations anywhere. Not “valuation”, not “valuer”. The phrase “loan-to-value ratio” appears once, in a worked example.
That makes sense. The valuation method is the lender’s decision under its own credit policy, not a product recommendation you make. The duty attaches to your credit assistance.
But three things in RG 273 do touch this, and brokers should understand them as general information rather than as a compliance rule anyone has written down for valuations specifically.
RG 273.77 deals with features unrelated to cost that may still be highly relevant — approval speed for a time-sensitive settlement is ASIC’s own example — and closes by saying ASIC expects “that any claims of this nature you make will be evidence-based and able to be substantiated.” If you tell a client that a physical inspection will produce a more accurate figure than the automated valuation, that is a claim. Until this week there was little public evidence to substantiate it. There is now some — and at $500,000 to $1 million it does not say what most brokers assume.
RG 273.60 recognises that credit providers take “different policies and approaches” to matters including lenders’ mortgage insurance. Valuation policy is where that becomes concrete: the same property, valued by two lenders’ panels, can sit on either side of an LMI threshold.
RG 273.42 is the hard boundary. A broker who provides incomplete or inaccurate information as part of a home loan application will not be acting in the consumer’s best interests, even where doing so improves the chance of approval or the terms obtained. An owner’s estimate supplied to support a valuation challenge is information in the application.
None of this is legal advice, and none of it is a rule ASIC has articulated about valuations. Brokers should raise valuation escalation policy with their licensee or aggregator compliance team, and seek independent advice where an obligation under the National Consumer Credit Protection Act 2009 is genuinely in question.
Where an 8% variance actually bites
Accuracy bands are abstract until they cross a policy threshold. Work it through on a single file.
A client is refinancing. They believe the property is worth $900,000 and they owe $720,000 — 80% LVR on their number. The valuation returns $850,000, a 5.6% variance well inside the band where three-quarters of valuations land. Their LVR is now 84.7%. Depending on the lender, that is an LMI premium that did not exist in the conversation you had last week, a pricing tier change, or a decline.
Nothing has gone wrong. No one has made an error. A 5.6% gap is an unremarkable outcome in a dataset where two physical inspections disagree by more than 10% roughly one time in five. But the client experiences it as a mistake, and the way you framed the number beforehand determines whether they think the mistake was the valuer’s or yours.
The operational implication is about sequencing, not escalation. Where a file’s viability turns on landing inside an LVR band, the variance belongs in the first conversation, not the recovery conversation after the number arrives.
The 2027 valuation wave
There is a commercial dimension building behind this, and it is worth getting the facts right because the detail is widely garbled.
The Federal Government’s negative gearing and capital gains tax reforms, announced at 7:30pm AEST on 12 May 2026, take effect from 1 July 2027. From that date the 50% CGT discount for individuals, trusts and partnerships is replaced with cost base indexation plus a 30% minimum tax rate on real capital gains accruing from 1 July 2027.
For an asset owned before 1 July 2027 and sold after it, the value at that date has to be established. The Government’s factsheet says an asset’s value at 1 July 2027 “will be determined by taxpayers as part of their tax return”, and that one option is to “seek a valuation of the asset as at 1 July 2027”. Importantly, it is not the only option: the alternative is a specified apportionment formula estimating the value from growth over the holding period, with ATO tools provided.
Cotality’s research paper estimates that approximately 2.4 million residential investment properties “may require a valuation as at 1 July 2027, subject to confirmation of the scope and requirements”. That figure is Cotality’s own and carries no attribution on the published page, so treat it as an industry estimate rather than an established number — particularly given the ATO formula exists as an alternative path.
Separately, note that the negative gearing changes do not involve a valuation at all, and that properties held before the 12 May 2026 announcement are exempt from them. Brokers fielding investor questions should be careful not to merge the two measures, and should send clients to their accountant rather than opining on cost base.
What to review this week
- Audit your own language. Listen to how you describe a valuation to clients. If the phrase “a proper valuation” is in your vocabulary, it is doing work you may not be able to substantiate in the $500,000–$1 million band.
- Flag the band risk at the first meeting. On any file where the deal turns on an LVR threshold, say out loud that valuations carry a variance and that a result 5–10% either side of expectation is ordinary, not an error.
- Reserve escalation for where the evidence supports it. Above $1.5 million the method gap is real. Between $500,000 and $1 million it is one or two points.
- Treat sub-$300,000 and above-$2 million security as uncertain by default. One is poorly served by every method tested; the other was not tested at all.
- Check your lender panel’s valuation and escalation policies. This is where the difference actually lives, and RG 273.60 recognises those policies differ.
- Raise it with compliance before you change a process. Nothing here is a regulatory requirement about valuations. Your licensee’s position governs.
- Across about 1.17 million records, two full inspections of the same property by different firms agreed within ±10% in 78% of cases. The IntelliVal AVM reached 77%.
- In the $500,000–$1 million core — 57% of the dataset — every method cleared 80% within ±10%, bar Desktop at 79% in one bracket.
- At $1.5–$2 million the AVM fell to 64%, nine points behind full inspection at 73%. That is where escalation is best justified.
- The AVM and Desktop data stop at $2 million, so the dataset says nothing about higher-value files.
- RG 273 never mentions valuations. RG 273.77 does say ASIC expects the claims a broker makes to be evidence-based and able to be substantiated.
The strategic point
The broker instinct to escalate a disappointing valuation comes from a good place — it is advocacy, and clients value it. But advocacy aimed at the wrong target costs credibility with lender BDMs and burns time that the file needs elsewhere.
What this dataset offers is not a reason to stop escalating. It is a reason to escalate with a sharper sense of where a different method will plausibly produce a different answer, and where you are simply asking a second estimator to re-roll the same dice. In the middle of the market, that is mostly what an upgrade buys. At the top of the market, it is not.
The brokers who get value out of this will be the ones who move the valuation conversation forward in the process — from the recovery call to the first meeting. That costs nothing and changes how every subsequent number is received.
Breaking news for modern brokers
Lender policy shifts, regulator moves and market data — read in the time between appointments.
More at The Broker Times →- Cotality, Benchmarking Digital Valuation Accuracy research paper, 19 November 2026 — cotality.com
- ASIC Regulatory Guide 273, Mortgage brokers: Best interests duty, issued 24 June 2020 — asic.gov.au
- Australian Government, Negative Gearing and Capital Gains Tax Reform factsheet — budget.gov.au
Disclaimer: This article is for general information and professional development purposes only. It does not constitute legal, compliance, or financial advice. Brokers should consult their aggregator's compliance team and, where required, seek independent legal advice regarding their obligations under the National Consumer Credit Protection Act 2009 and ASIC's responsible lending guidelines.

