88% of Firms Use AI, Only 11% Have It Embedded in Real Work

Eighty-eight percent of Australian and New Zealand businesses use AI, but only 11% have it deeply embedded in how work gets done. That gap comes from CPA Australia's survey of 1,679 accounting and finance professionals across six markets, and it frames a problem now landing on senior leaders' desks months after the Copilot, ChatGPT and Claude licences went live.

The pattern is familiar to change-management practitioners. The calls they take are no longer about implementing AI. They come from organisations that already have. IT has approved the platform. There has been a webinar, sometimes several. Then someone senior asks the question that prompts the call: why isn't this being used more, and why hasn't the work changed?

The easy diagnosis is resistance. In most engagements, that isn't the first barrier. When teams are asked why they aren't using AI, the most common answer is "I don't know what I'd use it for." Workshop polls point the same way.

But that answer is rarely the whole story. Unclear rules, unsuitable tools, competing priorities and limited time to practise all contribute. In practice, relevance, permission and practice tend to fail together.

What separates access from adoption?

Licences, logins and webinar attendance tell you whether people can use AI. They don't tell you whether it has changed how work gets done. That distinction is the real story in the CPA Australia numbers: near-universal access in Australia and New Zealand, yet last place among the six surveyed markets for broad integration.

Practitioners diagnose stalled rollouts with structured change models. One Prosci-certified practitioner uses ADKAR — Awareness, Desire, Knowledge, Ability and Reinforcement — to describe what a person needs before a change sticks. Many rollout programmes concentrate on announcing the change and explaining the tool. They leave motivation, practical capability and reinforcement to chance.

Three things need to work together: purpose, permission and practice. Here's what each one means in practice.

Purpose

People need a task worth changing. Generic examples, such as summarising an email, rarely motivate a site engineer or a payroll officer, because they don't look like their job. Desire comes from seeing AI help with the work in front of you.

Permission

People need to know what's safe. Employment Hero found 37% of New Zealand workers feel guilty using AI at work. That doesn't prove guilt causes underuse, but it suggests workplace attitudes and expectations matter. Unclear rules cut both ways: some people hold back, while others use AI quietly, outside approved tools.

Practice

Knowledge isn't Ability, and the distinction matters. What lifts uptake most clearly, practitioners report, is in-person workshops built on the organisation's own workflows. People bring a real task — a tender response, a monthly report — and build the AI-assisted version in the room, using approved tools and suitably sanitised material. They leave having done it, not just having watched a demonstration.

One recent example makes the mechanism visible. A clinician's administrator had watched several generic AI videos. They left her more confused, and more wary, than before. After one personal session built around her own tasks, she built and deployed two Copilot agents that have lifted her productivity.

The technology hadn't changed. What changed was that she could see where it fitted her job, and felt safe to try. The room matters less than the relevance, practice and feedback — which is why larger organisations train managers and champions in person first, then have them take it into their teams.

Why does governance matter for usage?

Governance earns its place not as a brake, but as what makes safe use straightforward. MyHR reported that just 5.7% of AI-using employers in its research had an AI policy. Where policies do exist, they're often too vague to act on. "Use AI responsibly" tells nobody what to do on a Tuesday afternoon.

Employees need guidance they can actually use, ideally on one page:

• which tools are approved
• what information can and can't go in
• which outputs need a human check
• that it's fine to say you used AI
• who to ask when something goes wrong

That page is the visible layer of governance, not the whole of it. Behind it, leaders need to have decided:

• which tools and use cases are approved, and who approves new ones
• how data is classified and where it may go
• where human review is required, and who is accountable for the output
• how incidents are reported and escalated
• when a pilot is good enough to scale

Not every use needs the same scrutiny. Drafting a routine internal summary is different from using AI to assess an employee, advise a customer or inform a consequential decision. Good governance makes low-risk use easy and keeps tighter controls for high-risk use.

Done well, this removes the uncertainty that stops people trying, and gives managers something concrete to reinforce. It won't create demand on its own. A well-governed tool that solves no worthwhile problem will still sit unused.

Who actually owns adoption?

Adoption falls between functions, which is often why it stalls. The CIO owns technical assurance and enablement: the platform, security, data controls and integration into the systems where work already happens. If the approved tool is slower than the workaround, people will use the workaround. No workshop or policy fixes that; integration has to come first.

HR helps people build the capability and confidence to use AI, including manager expectations, role impacts and consultation. But business leaders must own the changes to workflows and the results those changes deliver.

Managers are the link. Their job isn't to mandate AI use. It's to identify suitable tasks, give people time to experiment, set quality expectations and be clear where AI adds risk or little value.

For HR, that also means considering consultation, changing role expectations, job security and the psychosocial risks that poorly managed change can create — which Australian work health and safety rules now require employers to manage.

Should you measure logins or changed work?

Higher usage isn't automatically better. An organisation can increase active users while producing more rework, exposing sensitive information or applying AI to decisions it shouldn't touch. The aim is valuable, well-governed use.

So measure what changed:

• which workflows now use AI routinely
• output quality and rework
• time saved on specific tasks
• staff confidence
• incidents or near misses

That also gives reinforcement something to work with. Teams sharing one real example in an existing meeting, and managers recognising better work rather than prompt counts, help new habits last.

What should leaders take from this?

Buying access was the easy part. AI adoption isn't simply a training problem. It's a leadership and operating-model problem. People need to know where AI adds value, what they're permitted to do, and who will help them change how work gets done.

A useful test for the next leadership meeting: can you name three workflows where AI is genuinely improving outcomes, explain the rules for using it safely, and identify the business leader accountable for the result?

If not, another webinar probably isn't the place to start. Given that Australia and New Zealand rank last of six markets for integration despite leading on access, the next competitive advantage will go to organisations that fix purpose, permission and practice — not the ones that buy more licences.