I’m going to say something that won’t make me popular with training vendors: most of the money Australian companies spent on AI training last year was wasted.

Not all of it. But most of it. And I say that as someone who spent fifteen years running L&D at one of the big four banks, where I personally signed off on hundreds of thousands in technology training budgets.

The programs looked great on paper. Polished decks. Engaging facilitators. Strong satisfaction scores. Employees ticked the completion box, HR reported upskilling numbers to the board, and everyone felt good.

Then nothing changed.

The Classroom Fallacy

Here’s what keeps happening. A company decides it needs to “get its people up to speed on AI.” Someone sources a training provider. They book a half-day workshop. Staff attend, learn what a large language model is, write a few prompts, get a certificate.

Six weeks later, they’re doing their jobs exactly the way they did before.

This isn’t a content failure. The problem is structural. We’re applying a classroom model to something that doesn’t work in a classroom.

Think about it. When you learned to drive, you didn’t sit through eight hours of lectures and get handed the keys. You sat in the car, with someone beside you, and practised in the actual environment where you’d be driving. The instruction was a small part. The embedded practice was everything.

AI skills work the same way. You can’t teach someone to integrate AI into their procurement workflow with a generic ChatGPT demo. They need to be in their workflow, with their data, solving their actual problems.

The Numbers Don’t Lie

Jobs and Skills Australia has been tracking digital skills development across the economy, and the pattern is consistent: formal training alone produces minimal lasting behaviour change. The organisations seeing real adoption aren’t spending the most on training. They’re embedding AI into how work gets done.

An L&D director at a mid-size professional services firm in Brisbane told me recently they’d spent $45,000 on AI training for 120 staff. Three months later, fewer than 20% were using AI tools regularly. Not because the training was bad — because it was disconnected from daily work.

Compare that to a smaller firm in Melbourne that spent roughly $8,000 on a different approach. Instead of workshops, they paired each team with someone who sat alongside them for two weeks, building AI into existing processes. Adoption at three months? Over 70%.

The difference wasn’t budget. It was method.

Tools vs. Workflows: The Core Mistake

Most AI training programs teach tools. “Here’s how ChatGPT works. Here’s how to write a prompt. Here’s what Copilot does in Excel.”

That’s like teaching someone how a hammer works and expecting them to build a house.

What people need is workflow-level integration. Not “how does this tool work” but “how does this tool fit into what I do every Tuesday afternoon when I’m preparing client reports.” Generic tool training produces generic results — which is to say, almost none.

When I was at the bank, we learned this with every major tech rollout. The programs that worked embedded trainers into teams, watched how people actually worked, and helped them rebuild workflows with the new technology.

AI is no different. It’s arguably more dependent on this approach because its value is so context-specific.

What Actually Works

If you’re an L&D leader planning AI capability development, here’s what I’d argue for instead of booking another workshop.

Embed, don’t extract. Stop pulling people out of work to learn about AI. Put AI learning inside their work. Pair teams with someone who understands both AI and their business context. Have that person sit with them and co-build new workflows.

Start with the problem, not the tool. Don’t begin with “let’s learn ChatGPT.” Begin with “what takes you the longest every week?” Then figure out if AI can help. The motivation is completely different when someone connects AI to a pain point they already have.

Accept that it’s slower upfront. Embedded learning takes more time per person than a workshop for fifty. But the cost per person who actually changes their behaviour is dramatically lower. You’re paying for results, not attendance.

Invest in internal champions. Find the people who’ve already started experimenting with AI. Give them time and recognition. They’ll do more for adoption than any external trainer because they understand the work context intimately.

Working with experienced AI consultants Sydney can help design these embedded approaches — particularly when you need outside expertise applied inside real workflows rather than generic classrooms.

The Uncomfortable Truth for L&D

I know this is a hard message. Workshops and courses are what we know. They’re measurable. They’re scalable. They look good in board reports.

But if we keep running programs that produce impressive completion metrics and negligible behaviour change, we’re not developing capability. We’re performing it.

As Harvard Business Review has noted repeatedly, the gap between AI investment and AI value realisation is growing, not shrinking. Training that doesn’t translate to practice is a big part of why.

The organisations that’ll come out ahead aren’t spending more on AI training. They’re spending differently. Less on content delivery, more on context-specific support. Less on teaching tools, more on reshaping workflows.

It’s messier. It’s harder to put in a spreadsheet. And it actually works.

The money’s there. The intent’s there. It’s the model that’s broken.