Most organisations making decisions about AI right now are doing it on data they would not stand behind if they had built it themselves. That pressure to move fast is real, but the pattern behind it is not new, it was just slower last time.

That is what AI governance in Australia keeps circling back to: the risk was never really about the model. It is about what the model is standing on, and this business has watched that story play out before.

Cast your mind back to when Tableau became widely used. People who had never built a reporting system could suddenly build dashboards, and they did. Marketing built them. Finance built them. Operations built them. Within a year, most parts of the business had their own version of the truth.

Then the questions started. Why does this dashboard say revenue is up 12% and that one says 8%? Which one is right? A tooling conversation became a trust conversation, because once people stop trusting the numbers, they stop trusting the system behind them.

Omnia's data teams spent a lot of time in that phase: walking into organisations where reporting had scaled faster than the data underneath it could support. Hundreds of dashboards, all looking useful, none of them consistent with each other. The work became less about building more dashboards and more about reducing them, back to a smaller set people could actually rely on. The tools were not wrong. The foundations had not kept up.

AI is following the same pattern, faster, with a much lower barrier to entry. Building something that looks like an AI solution no longer needs a data team. In a lot of cases it needs a tool and a credit card.

So people are building, because there is pressure to move, because boards want to see progress, because nobody wants to be the organisation that fell behind. ADAPT estimates Australian organisations are spending around $28 million a year on AI, and 72% say they have not achieved measurable ROI. Gartner's read is similar: a large share of AI projects without AI-ready data will eventually be abandoned. Investment goes in, activity increases, and outcomes do not land, usually because the issue was never the AI. It was what it was sitting on – and the stakes are higher this time.

With Tableau, the downside was a trust problem: conflicting dashboards, frustrated analysts, a few awkward board meetings. What is sitting on top of these foundations now is not dashboards. It is recommendations, decisions, and in some cases automated actions.

From December 2026, updates to the Australian Privacy Act will require businesses to disclose when AI systems make automated decisions affecting individuals, particularly relevant for financial services and government organisations already carrying the heaviest disclosure obligations. That context matters, but it is not the main point. The main point is that the same organisational patterns that caused problems during the Tableau wave are playing out again, with more at stake.

What the organisations that handled it well did differently. The lessons from that earlier wave transfer directly. The businesses that navigated the Tableau shift without much pain did three things.

• First, they assigned ownership early. Not to the platform, not to the consultancy: someone inside the business owned the metric definitions, the logic, and whether the output could be trusted. The organisations that struggled were usually the ones where dashboards spread faster than accountability did.

• Second, they set a reference point before things got complicated. During the Tableau wave, that meant agreeing on core metrics, revenue, customer count, margin, before dozens of teams started calculating them differently. With AI, the equivalent is pressure-testing outputs against known scenarios before people start relying on them. "Build a small set of golden questions," says one of Omnia's senior leaders in data and AI. "Three to five examples where the business already knows the correct answer. You need some way to recognise when the system is confidently producing the wrong thing."

• Third, they separated experimentation from operational dependency. A dashboard built for exploration was one thing; that same dashboard becoming the number in board reporting was something else. AI is going through the same transition now, and a lot of organisations drift between the two states without ever formally deciding to make the shift, which is usually when governance starts lagging behind adoption.

Once AI starts influencing decisions, it stops being a technology question. It becomes a question of who is responsible for what comes out of it.

Not sure where your AI projects sit on that spectrum

The Omnia Collective's AI governance advisory team runs a structured readiness assessment that looks at data foundations, ownership gaps, and where experimentation has quietly become operational dependency, whether that is Claude-powered workflows or another platform entirely. It takes a couple of hours and gives you a clear picture of what is ready and what is not before it becomes a problem. See our data and AI governance services  or get in touch.

This post is adapted from Episode 1 of our Nap Stack podcast on AI, data, and building a business. Listen on Spotify or Apple Podcasts.