Where AI is landing for Australian financial services. The numbers, and what they mean for the people in it.

The short version 

As AI agents reach human capability across a range of skills, it will matter how organisations and people direct that energy productively. US firms are tracking AI usage by token spend and are unsure of the return. Australia has built the guard rails and just started on the implementation journey. 

AI tools are excellent for programming and average for financial analysis so far. As technology changes, people and roles will adapt around the capabilities. Knowledge workers are around a third of the Australian workforce, and there is significant productivity potential available with appropriate use of the tools. 

On AI 

Anthropic now has the highest share of business adoption at 34.4%, according to Ramp. Companies with high AI spend grew revenue at 27% a year for three years. Those with no AI spend grew at 3% a year. 

In the US, the scramble to monitor employee AI usage rather than ROI has incentivised wasted compute. PwC notes that only 7% of Australian enterprises surveyed have redesigned workflows to incorporate AI, rather than simply adding AI tools, trailing AI leaders at 56%. Australia falls behind on every workforce measure. Its lowest score is on performance incentives that encourage employees to experiment with and use AI in their work: 13%, against AI leaders on 59%. 

On financial services 

Vals.ai benchmarks AI model performance. While agents routinely score near 90% on difficult coding or software engineering tasks, accuracy on entry-level financial tasks sits around 50%. Models handle simple retrieval well, but still struggle with harder, multi-step financial work that relies on precise numbers and industry convention. 

That capability gap may explain why technology executives are spending hundreds of billions on data centre capex while many investors reference the dot-com era. Different professions see different utility from AI tools, and so far the tools keep improving. 

On people 

Leadership and judgement are critical to directing new capabilities, as MIT notes. METR, which researches the capabilities of frontier AI models, found that technical people reported a median 2x improvement in the value of their work using AI tools. 


And because models are still not fully accurate, Microsoft found the human skills that matter most as AI takes on more work are quality control of output and critical thinking about the decisions made from those outputs. As agent use increases, human involvement does not disappear, it changes shape. What declines is tactical, step-by-step execution. What rises is the need to set direction, define standards, and evaluate outcomes. 

With thanks to Darren Lee for this guest piece. 

References 

Every source below was reviewed by the author. Developments are drawn from the week of publication. 

Stanford HAI, 2026 AI Index Report 

The annual benchmark of where AI capability and adoption actually stand. hai.stanford.edu

Ramp AI Index, May 2026  

Anthropic passed OpenAI on business adoption for the first time, reaching 34.4%. ramp.com/leading-indicators

CNBC, May 2026  

Almost every Fortune 500 now tracks AI usage, but few track whether the spend delivers a return. cnbc.com

HCA Magazine, May 2026 

Amazon staff are inflating an internal AI-usage leaderboard, a case study in what happens when a metric becomes a target. hcamag.com/au

PwC Australia, April 2026 

Just 7% of Australian enterprises have redesigned workflows around AI, against 56% for global leaders. pwc.com.au/media/2026

Vals.ai, FABv2 Benchmark

No AI model clears 52% on entry-level financial analyst tasks, even under generous scoring. vals.ai/benchmarks

MIT Executive, May 2026 

AI changes the tools; leadership and judgment still shape the impact. executive.mit.edu/blog

Microsoft, May 2026 

As agents take on execution work, the human skills that rise in value are quality control and critical thinking. blogs.microsoft.com