Most AI job ads in Australian financial services are written to filter. That is what a requirements list is for. The problem is that several of the filters in common use remove strong candidates before they remove weak ones, and because you never see who did not apply, the failure is invisible.
Here are six requirements we see on most briefs, what each one actually excludes, and what to use instead.
1. “Degree in Computer Science, Mathematics or a related field”
Roughly 40% of people who recently entered Australia’s tech workforce did not come through university or VET. They arrived via workplace training (55,000), industry credentials (44,000), or self-directed learning (34,000).1 A hard degree requirement discards a large slice of the market, and does it at the automated-screening stage where nobody reviews the decision.
Instead: make the degree preferred, or drop it and test the underlying capability directly. If the role genuinely needs graduate-level statistics, say that and assess it.
2. “5+ years of experience with large language models”
This one is self-refuting often enough to be worth flagging. Requirements routinely specify more years in a technique or tool than the technique or tool has existed in production use. The candidates who match are the ones willing to overstate, which is the opposite of the selection you wanted.
Instead: anchor years to the durable skill and describe the recent work separately. “Eight years building production software, including recent work shipping retrieval or agent systems” selects far better.
3. The full vendor stack
Naming every tool in your environment (a cloud, a feature store, an orchestrator, a monitoring product, a vector database) and requiring experience in all of them produces a list almost nobody satisfies, while screening for exposure rather than judgement. It also biases toward candidates from organisations that happen to share your architecture, which is not a quality signal.
Instead: name two tools that genuinely matter and state the rest as your environment rather than your requirements.
4. Titles as a proxy for seniority
Title inflation and deflation are both severe in this market. “Senior Data Scientist” at a scale-up and at a tier-one bank can be six years of seniority apart in either direction. AI Engineer only became the fastest-growing job title in Australia recently. LinkedIn recorded roughly 150% growth in the role and ranked it first on its 2026 Jobs on the Rise list.2 That means the title is being applied to a very wide range of actual jobs.
Instead: describe the scope. Who does this person influence, what decisions do they own, what is the hardest problem they will face in the first year.
5. Assuming the market will forgive a low band
There is a persistent belief that AI-exposed workers are anxious and therefore pliable on money. The Australian evidence does not support it. The Office of the Chief Economist found in July 2026 that employment in the occupations most exposed to AI grew 5.6% between November 2022 and February 2026, that software development employment is up 25% since November 2022, and that there is no evidence of broad AI-driven labour market disruption in Australia.3
Instead: if the band is genuinely fixed, say so early and compete on the work, the data, or the autonomy. Candidates respect a constraint stated up front and resent one discovered at offer.
6. Defaulting to the office
Location and attendance requirements are the most expensive line in most AI job ads relative to how little thought they receive. Every additional mandated day narrows the addressable market, and in a specialism where the national pool is small to begin with, that compounds quickly.
Instead: decide what the attendance requirement is actually buying you, and price it. If it costs you a third of the pool, it should be worth a third of the pool.
The market context that makes this matter
It would be easy to read all this as advice for a red-hot candidate market. It is more specific than that.
Australia’s technology workforce contracted 0.3% in 2025 to around 967,000 people, the first decline on record.1 Jobs and Skills Australia removed software engineer from the shortage list in every state and territory, with the share of occupations in national shortage falling to 293 of 1,022.4 Broad technology hiring genuinely has become easier.
But the government’s shortage data is built on ANZSCO occupation codes, which have no granular category for AI-specific roles. AI demand is invisible in those statistics, folded into software categories that are easing in aggregate. Meanwhile Deloitte Access Economics estimates Australia still needs 259,000 additional technology workers over the next decade.1
You are hiring in a loose market for a tight skill. Job ad requirements calibrated for the loose market will cost you the tight one.
The practical consequence: filters that were reasonable when a hundred people could do the job are actively harmful when nine can. Every requirement should have to justify what it excludes.