I’m going to say something that might sound strange coming from someone who works with manufacturing AI: most Australian manufacturers shouldn’t be investing in AI right now.

Not because AI doesn’t work. It does. But because there’s usually easier, cheaper improvement available first—and skipping straight to AI often means building on a shaky foundation.

The uncomfortable truth about AI readiness

I’ve visited hundreds of manufacturing facilities over the years. Here’s what I typically find:

Data chaos: Production data scattered across spreadsheets, paper forms, disconnected systems, and tribal knowledge. Nothing integrated. No single source of truth.

Basic process problems: Obvious inefficiencies that don’t require AI to spot—unnecessary movement, poor layouts, unbalanced lines, inventory in the wrong places.

Maintenance neglect: Reactive maintenance culture where things get fixed when they break. No systematic approach to preventing failures.

Measurement gaps: Key process parameters that nobody’s tracking. Decisions made on gut feel because the data doesn’t exist.

Putting AI on top of this foundation is like putting a Ferrari engine in a car with bald tyres and broken brakes. The engine might be amazing, but the car won’t perform.

What to fix before AI

Let me be specific about the prerequisites that most manufacturers should address first.

1. Basic data collection

Before AI can analyse your operations, you need data to analyse.

Do you have:

• Accurate production counts by machine, shift, and product?
• Downtime records that distinguish planned vs unplanned and capture reasons?
• Quality data at the process level, not just final inspection?
• Energy consumption by major equipment?
• Maintenance records that include what was done and why?

If the answer to most of these is no, start there. Basic data logging systems cost a fraction of AI investments and are necessary prerequisites anyway.

2. Process stability

AI optimises within constraints. If your process isn’t stable, there’s nothing to optimise—you’re just fighting variation.

Signs of instability:

• Widely varying cycle times for the same product
• Quality that depends heavily on which operator is working
• Regular equipment failures that disrupt production
• Frequent rework and adjustment

Fix the basics: standard work, equipment reliability, training. Get to a stable baseline, then look for AI-assisted improvement.

3. Lean fundamentals

Traditional lean manufacturing approaches address many problems that people now want AI to solve.

5S gets workspaces organised. Standard work reduces variation. Visual management makes problems visible. Continuous improvement engages frontline workers. Value stream mapping identifies waste.

These approaches cost relatively little and often deliver significant improvement. A factory that hasn’t done basic lean work probably has more value available there than from AI.

4. Integrated systems

If your ERP doesn’t talk to your production systems, if your quality data lives separately from production data, if maintenance records are in a filing cabinet—you have an integration problem.

AI needs connected data. Building integrations between existing systems might be more valuable than adding AI on top of disconnected systems.

The hierarchy of manufacturing improvement

Think about it as a pyramid:

Foundation: Stable processes, trained people, organised workplaces, reliable equipment

Middle layer: Data collection, system integration, basic analytics

Upper layer: Advanced analytics, predictive systems, AI-assisted optimisation

You can’t skip levels. Companies that try end up with AI projects that fail for “data quality issues” or “change management problems”—which really means they didn’t have the foundation in place.

When AI does make sense

I’m not saying AI is never appropriate. It is—in specific situations:

You’ve done the basics: Your processes are stable, your data is solid, you’ve exhausted easy improvements.

The problem is genuinely hard: Optimising complex systems with many interacting variables, predicting subtle equipment degradation, classifying defects that are hard for humans to see consistently.

The scale justifies the investment: AI projects have real costs. The problem needs to be big enough that solving it justifies those costs.

You have the capability to implement and maintain: Either internal expertise or committed partners who’ll stick around.

A well-run facility that’s hit diminishing returns on traditional improvement? Yes, explore AI. A facility still struggling with basic operational discipline? Fix that first.

The cost of premature AI adoption

Jumping to AI before you’re ready doesn’t just waste money on the AI project itself. It has knock-on effects:

Distraction from fundamentals: Management attention goes to the shiny AI project instead of the boring-but-important foundation work.

Demoralisation: When the AI project struggles or fails, it poisons the well for future technology initiatives.

Vendor lock-in: Money spent on AI vendors can’t be spent on data infrastructure, training, or process improvement.

False conclusions: AI systems built on bad data can lead to confidently wrong decisions.

What I wish more manufacturers would do

Invest in data infrastructure: Spend the AI budget on getting your data house in order. Connect systems. Implement proper historians. Build dashboards so people can see what’s happening.

Train your people: Lean training, data literacy, basic problem-solving skills. A workforce that understands process improvement is more valuable than AI software they don’t understand.

Do the boring stuff: 5S, total productive maintenance, standard work. Unsexy but effective.

Start small with analytics: Before AI, can you do basic analytics? Can you answer simple questions from your data? Build this capability.

Fix your equipment: Reliable equipment is the foundation of everything else. If you’re spending your time fighting breakdowns, you’re not ready for optimisation.

The AI vendors won’t tell you this

Of course they won’t. Their job is to sell AI software.

And to be fair, the AI does work. If your foundation is solid, AI can deliver real value. The vendors aren’t lying about that.

But they’re not interested in telling you to spend two years getting ready before buying their product. That’s not their problem—it’s yours.

A different path

Here’s what a more sensible journey might look like:

Year 1: Stabilise processes, implement basic data collection, train people in lean methods

Year 2: Integrate systems, build analytics capability, address reliability issues

Year 3: Pilot AI in one area where foundation is solid

Year 4: Expand based on pilot learnings

This is slower than vendors want to sell and executives want to buy. But it’s how you build lasting capability rather than a shiny project that fizzles.

The bottom line

AI in manufacturing is real and valuable. I’ve seen it work. But it’s not a shortcut around the hard work of operational excellence.

If you’ve mastered the fundamentals and you’re looking for the next level, AI might be your answer. If you’re still fighting daily fires and wishing your systems talked to each other, you’ve got more basic work to do first.

The most successful manufacturers I know didn’t jump straight to AI. They built strong foundations over years, then added AI as one tool among many. That’s the path that actually works.