Since ChatGPT’s debut in late 2022, NAB economists say job growth in heavily AI-exposed industries such as finance, insurance, ICT, software and professional services has lagged.
Roles most affected by generative AI are underperforming less-exposed occupations, with employment and hours worked in these high-exposure jobs sitting about 9% lower.
For every 10% of an occupation’s tasks that current generative AI can automate, employment growth in that occupation has been around 2% slower than in minimally exposed roles.
Overall, roughly 15% of Australian jobs fall into the highly or significantly exposed category.
Despite that drag, adoption across business is racing ahead.
NAB’s research finds 42% of Australian firms already use AI, with another 14% planning to adopt it.
Property services lead utilisation at 69%, followed by finance and insurance services at 64% then business services at 61%.
Other sectors sit on a spectrum of exposure, with pharma and R&D-heavy industries in the medium to high band, frontline retail, manufacturing and healthcare in the mid-range, construction and agriculture from low to medium, and hospitality plus personal services at the low end.
Those exposure patterns matter for wages and labour markets.
AI mainly targets white-collar, task-based roles, which could push employers to reprice labour and tilt relative wage growth toward blue-collar work.
NAB’s economists say it is difficult to isolate AI’s impact from a broader post‑pandemic slowdown, yet they argue the underperformance of AI-exposed occupations since 2022 is hard to ignore.
They also note that an 80% jump in business IT investment over the past decade has not translated into strong productivity gains, which suggests that simply spending on technology does not guarantee efficiency improvements.
NAB’s team is cautious on the long-term jobs outlook, noting that many assume AI means slower employment growth and higher unemployment, even as an ageing population shrinks labour supply.
They identify a key tension, since if AI lifts productivity across the economy that should boost real wages and aggregate demand, which in turn could increase labour needs rather than reduce them.
How this plays out will depend heavily on how effectively firms integrate AI into workflows, rather than just whether they adopt the tools.
That complexity is already shaping how central banks and policymakers think about technology, inflation and employment in the years ahead.
