Australia’s productivity is flatlining, posting the worst vitals we’ve seen in 60 years^[1].

Politicians and chief executives are prescribing artificial intelligence (AI) like it’s the new penicillin – a wonder drug with almost magical healing powers. Prime Minister Anthony Albanese, Treasurer Jim Chalmers and the Productivity Commission^[2] all see AI as a key part of the plan to cure Australia’s productivity ills, with estimates that automation and AI could add A$600 billion^[3] to Australia’s annual economy.

Unfortunately, AI is no panacea. It’s more like physiotherapy after major surgery: it only delivers if you put in the effort, follow the program and work with experts who know which muscles to strengthen and when.

AI projects have high fail rates

AI is a broad suite of tools and techniques, of which generative AI such as ChatGPT is just the latest iteration. When implemented well, AI can undoubtedly lift productivity across a wide variety of applications. Unilever’s legal team reports generative AI tools save its lawyers^[4] 30 minutes daily on document review and contract analysis.

Other AI applications can deliver life-saving results at even greater efficiency. In a German study^[5], AI-supported mammography screening reduced radiologists’ reading time by 43% for examinations tagged as normal, while improving cancer detection rates.

The federal government is focused on improving productivity. In this five-part series^[6], we’ve asked leading experts what that means for the economy, what’s holding us back and their best ideas for reform. But the hard truth is that AI-driven productivity gains like these depend on both smart implementation and trusted adoption. Organisations that skip the tough part – such as staff engagement, training and good governance – often find the promised benefits never materialise. The numbers back this up: some 80% of AI projects end up failing^[7], twice the rate of traditional IT projects. Only one in four executives in a global survey report meaningful returns^[8] on their AI investments. We shouldn’t really be surprised. Other general-purpose technologies, such as electricity and earlier digital technologies followed a similar path. US economist Robert Solow famously said^[9]: “You can see the computer age everywhere but in the productivity statistics.” Workers don’t trust the technology Like the early days of the internet in the 1990s, the success of AI relies on adoption and trust. Without trust, uptake stalls and the benefits evaporate. That’s a big challenge in Australia, where public trust and optimism in AI remains comparatively low^[10]. Why? Australians also report^[11] lower levels of AI use, training and confidence. And people are less likely to trust what they don’t understand. Closing that trust gap means involving workers from the start. By listening to worker concerns and identifying existing pain points in processes, companies can deploy AI systems^[12] that help, rather than sideline employees. Conversely, when workers aren’t meaningfully involved, things don’t go well. Take Klarna. The Swedish fintech volunteered to be the generative AI platform OpenAI’s “favourite guinea pig”. It slashed jobs and claimed to have automated the equivalent of 700 employees^[13]. But CEO Sebastian Siemiatkowski now admits the shift to AI hurt customer service, forcing the company to rehire humans^[14]. Similarly, Duolingo recently faced a user backlash^[15] when it replaced 10% of contractors with AI.

Regrets? Bosses have a few

These aren’t isolated cases. Some 55% of UK executives who replaced workers with AI later regretted it^[17]. In the rush to automate, workers are often seen as expendable.

This attitude to AI leads to what US economists Daron Acemoglu and Pascual Restrepro call^[18] “so-so automation”, where technology displaces workers without delivering meaningful productivity gains.

Rather than trying to replace staff with AI, organisations should be deeply engaging with them. Engaging workers can dramatically boost the AI’s return on investment.

Like other general-purpose technologies, getting the most out of AI means transforming the way we work. And the data^[19] show companies that engage workers in organisational transformations are nine times more likely to succeed.

The companies that are unlocking the benefit of AI understand it works best when it amplifies human capability, rather than replacing it. Workers still know things that algorithms don’t. They deeply understand the practical realities of their jobs, which is crucial for designing AI systems that actually get things done.

Designing better solutions

Our own research^[20] confirms this. Australian workers feel AI is being imposed on them without adequate consultation or training. This not only creates resistance to adoption but also means organisations are missing the experience of the people who actually do the work.

Our most recent report^[21] shows worker engagement strengthens competitive advantage and profitability, and leads to better AI solutions rooted in workers’ problems and needs. When workers are involved in deciding how AI is used, the solutions are better designed, more effective and more widely adopted.

Australia’s new Industry and Innovation Minister, Tim Ayres, recognises this. In a recent speech^[22] he emphasised the need to work “cooperatively with workers and their unions” on tech adoption.

It’s a promising place to start. If AI is going to be an effective treatment for Australia’s productivity challenge, then workers must be an essential part of the recovery team.

References

1. ^{^} in 60 years (ministers.treasury.gov.au)
2. ^{^} Productivity Commission (www.pc.gov.au)
3. ^{^} add A$600 billion (www.mckinsey.com)
4. ^{^} save its lawyers (www.ft.com)
5. ^{^} German study (www.nature.com)
6. ^{^} five-part series (theconversation.com)
7. ^{^} end up failing (fortune.com)
8. ^{^} meaningful returns (web-assets.bcg.com)
9. ^{^} said (www.standupeconomist.com)
10. ^{^} comparatively low (assets.kpmg.com)
11. ^{^} report (assets.kpmg.com)
12. ^{^} deploy AI systems (www.uts.edu.au)
13. ^{^} equivalent of 700 employees (www.hrgrapevine.com)
14. ^{^} to rehire humans (www.fastcompany.com)
15. ^{^} user backlash (www.fastcompany.com)
16. ^{^} Summit Art Creations/Shutterstock (www.shutterstock.com)
17. ^{^} later regretted it (www.orgvue.com)
18. ^{^} call (pubs.aeaweb.org)
19. ^{^} data (www.mckinsey.com)
20. ^{^} Our own research (www.uts.edu.au)
21. ^{^} recent report (www.uts.edu.au)
22. ^{^} recent speech (www.minister.industry.gov.au)