Ask four sources what an AI engineer earns in Australia and you will get four different answers. Hays puts the typical figure at $190,000, with a range of $165,000 to $250,000.1 Glassdoor’s average for a machine learning engineer is $137,500.2 SEEK publishes advertised ranges that cluster around $132,500 in most Victorian markets.3 Specialist recruiter guides quote a hiring band from $115,000 to $225,000-plus.4
The gap between the lowest and highest of those is more than $110,000, for what is nominally the same job. That is not noise. It is four incompatible measurements being reported in the same unit, and if you are setting a salary band off any one of them you are probably wrong in a direction you cannot see.
The numbers do not disagree because the market is volatile. They disagree because they are not measuring the same thing.
What each number is actually counting
Every published salary figure is the output of a collection method, and the method determines the answer far more than the market does.
| Source | What it measures | Systematic bias |
|---|---|---|
| SEEK | Salary ranges employers voluntarily disclose on job ads | Only captures roles where the employer chose to publish a number, and SEEK notes some disclosed salaries include superannuation while others do not3 |
| Glassdoor | Salaries self-reported by employees | Self-selection; skews to those who choose to report, and lags current offers |
| Hays and other recruiters | Placements the firm made | Reflects that firm’s client mix, and is published by a party with a commercial interest in the number1 |
| Job title averages | Whatever was labelled “AI engineer” | The title spans a Python developer wiring up an API and a researcher shipping production models |
The superannuation point alone is worth $12,000 to $27,000 on these bands, and it moves silently between sources. Before comparing any two figures, establish whether either includes it.
Nobody publishes a financial services number
This is the part worth being blunt about. There is no credible published salary benchmark for AI and machine learning roles specific to Australian financial services. Not from Jobs and Skills Australia, not from SEEK, not from the recruiters. Anyone showing you a banking-specific AI salary table has extrapolated it from general market data, or built it from their own placements, which is a legitimate thing to do and a different thing from a benchmark.
We would rather say that than publish a table we cannot defend.
The shortage signal is genuinely confusing
The demand picture is as contradictory as the salary picture, and for a similar reason.
On one hand, Australia’s technology workforce shrank 0.3% in 2025, to roughly 967,000 people, the first recorded decline.5 Jobs and Skills Australia removed software engineer from the national shortage list entirely: as of the 2025 list it is not in shortage in any state or territory, and the share of assessed occupations in shortage fell to 293 of 1,022.6 Demand for ICT occupations dropped 49% over the period.7
On the other, LinkedIn ranked AI Engineer the fastest-growing job title in Australia in its 2026 Jobs on the Rise list, at roughly 150% growth, and found AI literacy to be the single most in-demand skill across every job on the platform.8
Both are true. The reconciliation is that government shortage data is built on ANZSCO occupation codes, which have no granular category for AI-specific roles, so that demand is invisible in the official statistics, absorbed into broader software categories that are, in aggregate, easing. The macro tech market has loosened. A narrow band of AI roles inside it has not.
What this means for a band
If you benchmark an AI role against general software engineering data, you will land low, because the general market genuinely has softened. If you benchmark against AI-engineer headline figures, you may land high for a role that is really a data engineering job with a model attached. The title is not the unit of analysis.
AI is not shrinking the jobs it was supposed to
One more piece of context, because it changes how candidates read your offer. The Office of the Chief Economist published the first Australian government analysis of AI’s employment effect in July 2026. It found employment in the occupations most exposed to AI grew 5.6% between November 2022 and February 2026, that software development employment is up 25% over the same window, and that there is no evidence to date of broad AI-driven labour market upheaval.9
Candidates in these roles are not negotiating from fear. Any offer strategy built on the assumption that AI anxiety softens candidates is built on something the data does not support.
What to do instead of trusting a table
- Price the sub-skill, not the title. “AI engineer” spans at least four distinct markets: applied research, ML platform, LLM/retrieval engineering, and analytics with a model attached. They do not clear at the same price.
- Normalise for super before comparing anything. State package or base explicitly in your own band, and check what any external figure includes.
- Use your own offer data as the primary source. Your last three offers, accepted and declined, are worth more than any published table, because they are the only data measured on your brand, your stack and your location.
- Treat declines as the real signal. A band is too low when good candidates stop returning calls, not when a table says so.
- Ask what the counter-offer looked like. The number that matters is what it took to keep them where they are, and it is the number no survey captures.
None of this is a reason to ignore published benchmarks. It is a reason to treat them as four rough bearings rather than one map.