Fix the Data First: Your AI Pilot Didn't Fail Because of AI

**This story originally appeared in the April 2026 edition of AMT Magazine**

Australian manufacturers are curious about AI. Dabbling with chatbots, adding plug-ins to familiar software and experimenting with prompts. What most of them haven’t done is make it work.

The reason, says ARM Hub founder and chief executive Professor Cori Stewart, is almost never the technology.

“You can create an AI agent in an hour,” she says. “The problem is, if it’s not connected to data that works, it’s meaningless. People get to a proof of concept, roll it out across the organisation, and inevitably it falls over because it hasn’t been set up for success.”

Stewart leads one of four government-backed AI Adopt Centres in Australia, helping manufacturers and SMEs move from curiosity to commercial deployment. She’s seen the pattern repeat: companies invest in AI tools while ignoring the data infrastructure those tools depend on.

“What really matters to an organisation is to start to see itself as a data organisation,” she says. “We can harvest that data brain to work harder for the company.”

The spaghetti problem

Most manufacturers Stewart works with have the same issue. Their data lives everywhere: multiple SharePoints, a cloud system here, an ERP there, a CRM, an HR platform, safety software. None of them talk to each other.

“They might send it up for a report to the executive, but you’re not able to do deep science between one lot of data and the other if they’re in different systems,” she says.

Her prescription is to migrate to a modern Lakehouse environment, combining the flexible, low-cost storage of a data lake with the structured querying and data management capabilities of a data warehouse. The economics, she says, have shifted dramatically.

“It used to be you wouldn’t do this work for under a million dollars,” Stewart says. “Now you can migrate depending on your aspirations. Data is cheaper, but the systems are more sophisticated. Now there’s both the talent and the capability available to manufacturers to do this.”

What good looks like

Two ARM Hub portfolio companies show what happens when unified data meets a clear problem.

Microbio, a Brisbane diagnostics company, is building rapid tests for infectious disease, including a technology that detects 26 sepsis-causing pathogens from blood. The gold standard for detection was 24 hours. By pulling clinical, patient and hospital data into a single AI-enabled platform, they’ve cut that to three hours.

“Through unifying the data and overlying AI so it can report, they’re giving doctors and clinicians lifesaving support today,” Stewart says. The data infrastructure is also accelerating regulatory approval, replacing a manual FDA process with an automated one. Microbio now operates across five countries.

Urban Art Projects (UAP), which fabricates large-scale public art, faced a different problem: scheduling. Complex, bespoke projects drove around 3,000 different parameters, and pulling together a production schedule took more than two weeks.

Working with ARM Hub, UAP unified its scheduling data, trained a company-wide intelligence layer, and built a natural language interface on top. Now a scheduler inputs parameters, reviews scenarios, and presses enter.

“It saves them immediately about 10 days every time they schedule,” Stewart says, “and creates the confidence to take on new work.”

Who to trust, and who not to

When it comes to building a data environment, Stewart has a counterintuitive view on suppliers. She urges manufacturers to look beyond established CRM and SaaS providers, and to be wary of AI systems built for international markets.

“I’m actually encouraging, perhaps controversially, that companies look outside their established manufacturing providers,” she says. “All your systems must be accessible through API, so you don’t need them to access your data. They’re not overlords. Your data has been democratised. You are in charge of it.”

She’s also cautious about handing the project to internal IT teams.

“They are often quite wedded to that spaghetti, because they created it and they know how to use it. They may be quite fearful of what a modern data environment is.”

Speed matters

Looking across Australian industry, Stewart sees about 20% of manufacturers experimenting with AI, 30 to 40% actively investigating, and the rest not engaging at all.

“That’s something I think we should be concerned about,” she says. “It’s technology that’s moving faster than anything we’ve seen in the modern world.”

Her advice is simple: don’t wait.

“There’s no second mover advantage. You are slowing your learning, and you have to learn as an organisation. You can’t just adopt it from somewhere else. You have to go through what your company does, and you have to do it. There are no shortcuts.”