Employees use AI three times more than managers realise. Is yours one of them?

Employees use AI three times more than their managers realise, according to McKinsey's 2025 research. A PagerDuty survey from earlier this year puts a different number on it. 66% of office professionals across Australia, the UK, and the US admitted to using AI tools at work despite believing they weren't permitted under company policy. The methodologies differ and so do the figures, but both point in the same direction: unsanctioned AI use in the workplace is widespread.

At a recent presentation I ran on embracing AI in business, I asked the room whether they were using employer-provided accounts. Nearly every hand went up. That was genuinely refreshing to see. It means the conversation is happening in some businesses.

The pattern where it goes wrong is predictable, and it's really common. The tools arrived so quickly that policy hasn't had time to catch up. There's no clear direction on AI, so people make their own calls about what's safe. Someone discovers ChatGPT writes a first draft of their reports in two minutes. Someone else uses a free tool to summarise meeting notes. A third person is running customer emails through a chatbot. None of them are doing anything malicious. They're trying to get their work done faster. But without any visibility or structure around it, you have no idea what information is going in, where it's going, or what version of your business is being represented in the output.

The problem with free accounts.

Most shadow AI happens on personal free accounts, and this is where the real risk sits. On the free tier of most AI tools, your conversations can be used to improve the model, which means business information, client details, and internal processes could be included in that data. Free accounts have no admin controls, no usage visibility, and no central management. If the person using the account leaves, anything they've built or shared leaves with them. There's no audit trail and no way to know what's been passed through.

This isn't hypothetical. The terms of service for most free AI tools are explicit about how conversation data is handled. Most business owners haven't read them.

What a business account actually gives you.

On a paid or team account (Claude, ChatGPT, and most enterprise AI tools all offer this), your data is not used for training. You get admin controls that show who's using the tool, set boundaries on what's permitted, and give you a managed environment you can actually govern. Team accounts mean consistent tools, consistent outputs, and access that doesn't disappear when someone leaves.

The cost difference between a free account and a paid team account is typically $20 to $30 per user per month. The cost of a compliance issue, a data breach, or a client discovering their information was handled without care is considerably more.

Practical steps for getting ahead of it.

The first step is team alignment, and it's often the one businesses skip. Before you choose a tool or write a policy, your team needs a shared understanding of what AI is, what the business thinks about it, and what using it well looks like in your context. That conversation surfaces what people are already doing, builds trust, and means whatever comes next has buy-in rather than resistance.

From there, the focus shifts to guardrails. Most people assume guardrails slow things down. In practice they do the opposite. When your team knows how far they can go, they go further with confidence. When the boundaries are clear, people stop second-guessing every decision and start moving. A simple set of guidelines on what data can go in, which tools are approved, and how to handle something new takes an hour to write and removes a lot of uncertainty.

The third step is where the real work starts. Rather than trying to implement AI across the business at once, the businesses that get results pick one problem worth solving, map out how that process runs today, and assess which parts require genuine human judgment and which could be handled differently. That analysis is what tells you where AI will help, rather than where it sounds like it should help. And it's worth keeping in mind that AI won't fix a broken process. It'll just run it faster. The mapping step is where you catch that before it matters.

Done in that order, alignment then guardrails then process mapping and assessment, you end up with a team that's ready, a framework that's safe, and a clear first use case that delivers something measurable.

If you want to talk through where your business sits with AI right now, book a call.

Kate Fabian is the founder of Adoptech, a Cairns-based technology and AI adoption consultancy. She works with business owners across Far North Queensland on AI strategy, implementation, and training.

Sources: McKinsey, 2025. PagerDuty Shadow AI Workplace Survey, April 2026.