There’s a pattern that plays out in boardrooms across Australia every quarter. Someone senior comes back from a conference buzzing about AI. A strategy gets drafted. Budget gets allocated. And within six months, the whole thing quietly dies.

It’s not because artificial intelligence doesn’t work. It clearly does. The problem sits much earlier in the process — usually before a single line of code gets written.

The Strategy Document Nobody Reads

Most AI strategies start as a PowerPoint deck. They contain ambitious timelines, vague references to “data-driven decision making,” and a few vendor logos. What they rarely contain is a clear answer to one question: what specific business problem are we solving?

That’s the gap. Organisations jump to solutions before properly defining the problem. They buy tools before understanding their data. They hire data scientists before building the infrastructure those scientists need.

According to Harvard Business Review, around 80% of AI projects never make it to production. That number hasn’t budged much despite billions in investment globally.

The Data Problem Nobody Wants to Talk About

Here’s something uncomfortable: most companies don’t have “AI-ready” data. They’ve got spreadsheets on shared drives, legacy databases that haven’t been cleaned since 2014, and customer records scattered across five different CRMs that were never properly integrated.

Before you can do anything intelligent with data, you need to know where it lives, whether it’s accurate, and who’s responsible for maintaining it. That’s not exciting work. It doesn’t make for a good keynote. But it’s the difference between an AI project that delivers results and one that gets shelved.

Firms that get this right tend to invest in data governance first. They appoint data owners. They build pipelines. They create standards. Only then do they start exploring what’s possible with machine learning or automation.

Misaligned Expectations

Another killer: leadership expects AI to produce magic overnight. They’ve seen the demos. They’ve read the case studies from Google and Amazon. They assume their internal team can replicate those results with a fraction of the budget and none of the infrastructure.

The reality is that good AI development work takes time. It requires iteration, testing, and honest conversations about what the technology can and can’t do in a specific context. A proof of concept is not a product. A chatbot that works in a demo doesn’t mean it’ll handle the complexity of your actual customer queries.

Setting realistic timelines — and communicating them clearly to stakeholders — is one of the most important things a project lead can do. It’s also one of the most neglected.

Nobody Owns It

AI strategies fail when nobody truly owns them. If it sits with IT, the business units won’t engage. If it sits with the CEO, it becomes aspirational but disconnected from operations. If it sits with a “Chief AI Officer” who reports to nobody in particular, it drifts.

The organisations that succeed tend to embed AI initiatives within existing business units. They pair technical people with domain experts. They give small teams the authority to experiment and the accountability to deliver.

Australia’s CSIRO has published some useful frameworks on this, particularly around governance structures that balance innovation with responsibility.

What Actually Works

From what we’ve seen, the companies making real progress with AI share a few traits:

• They start small. Not with a grand transformation plan, but with a single, well-scoped project that proves value.
• They invest in people. Training existing staff matters more than hiring a rockstar data scientist.
• They’re honest about their data. If it’s messy, they clean it before building on top of it.
• They iterate. Version one is never perfect. That’s fine.
• They measure outcomes, not activity. Building a model isn’t the goal. Improving a metric is.

The Uncomfortable Truth

Most AI strategies fail because they’re strategies in name only. They lack specificity, ownership, and a willingness to do the unglamorous work that makes everything else possible.

If your organisation is considering an AI initiative, start by asking: do we actually understand the problem we’re trying to solve? Do we have the data to support it? And do we have someone who’ll still be pushing this forward in twelve months?

If the answer to any of those is no, that’s where you start. Not with a vendor pitch. Not with a hackathon. With the basics.

It’s less exciting. But it’s what works.