Article By Glenn McPherson, Regional VP Australia, Snowflake
If there is one sector that isn’t shy in adopting technology, it’s manufacturing. From building robotics to deploying those same robots on the production line, manufacturing has often led the way when it comes to adopting cutting-edge technology.
When these technologies were initially adopted, there was anxiety. For one, would the CAPEX be too large compared to the status quo’s OPEX? In a high-cost manufacturing nation like Australia, these bottom-line concerns are always front of mind.
These anxieties aren’t unique to manufacturing, nor are they unfounded, yet all have been quelled by productivity gains, bottom-line benefits, and an improved speed to market. They’re the same anxieties we’re hearing about AI today, so it stands to reason that the sector would be quick to adopt the latest and greatest.
But in a recent cross-industry survey of generative AI adoption, manufacturers ranked second to last in putting the technology to work: just 37 per cent reported using gen AI across more than a handful of use cases. When it comes to proving the payoff, the industry trails everyone: only 56 per cent have quantified a positive return, the lowest reading of any sector, and the average return they report, 38 per cent, is also dead last.
At first glance, these figures paint a picture of an industry struggling to realise AI’s value. But they don’t tell the full story. A deeper dive into the same survey reveals that 90 per cent of manufacturers say AI has delivered tangible operational benefits, from strengthening supply chains and predicting equipment failures to reducing waste across production lines.
The issue in manufacturing is that when looking at the benefits AI can bring to the organisation, they’re myopic.
Manufacturers are more likely than companies in any other industry to see gen AI purely as a way to take cost out of operations: 57 per cent name operational efficiency as their primary driver versus 50 per cent elsewhere. That instinct is understandable in a business of thin margins, but it’s also self-imposing a ceiling on AI’s potential.
Now, the data does show one more hurdle to uptake; manufacturers were more likely than other industries to name their own workforce’s AI skills and experience as an obstacle at 41 per cent, against an average of 34 per cent elsewhere. This may be a key reason why manufacturers have the strongest preference for buying prebuilt AI agents from vendors, 68 per cent, versus just 31 per cent who want to build their own.
Off-the-shelf AI is fine, but it doesn’t leverage a manufacturer’s own internal data and IP. A company has years, if not decades of expertise and processes at its disposal, but too often off-the-shelf AI can’t leverage any of it.
It’s why those 31 per cent who want to build their own are on the right track, but how does this work when the perception of a skills gap is so pronounced in the sector? The key to building one’s own AI is leaning on partners that can set a solid data foundation and build the AI from there.
A prime example is Siemens Energy. The global energy player possessed a huge amount of valuable technical knowledge that was literally sitting on paper: a staggering 700,000 pages of proprietary R&D documents. It wanted to make that information available to its R&D team, but to read all of those documents would take a single person roughly four years.
Instead, Siemens Energy built an AI chatbot leveraging retrieval-augmented generation architecture to quickly surface and summarise those 700,000-plus pages of internal documents and accelerate research and development. In practical terms, the team digitised the paper records and then applied an AI chatbot on top of those datasets, making hundreds of thousands of pages easily queryable.
Why is this important? This use case demonstrates how AI can amplify R&D capabilities rather than simply reduce costs. It also empowers the people closest to the work with direct access to data, enabling faster insights and better-informed decisions.
If manufacturers can build a strong data foundation, the opportunities extend far beyond efficiency gains. Consider demand forecasting: AI models can combine historical sales, market trends, and external signals to improve forecast accuracy and reduce production bottlenecks. They can also analyse manufacturing performance, surface insights faster, and even anticipate equipment and process issues before they occur.
The anxiety surrounding AI is real, and manufacturing is not alone in grappling with it. But the sector’s relatively slow pace of adoption is surprising given its long history of driving innovation and embracing bleeding-edge technology.
To compete, Australia’s manufacturers need to shift from the mindset that AI’s sole benefit is to cut costs to seeing it as something that can create opportunity.
