Sarah went to the Copilot workshop. Then she was told to get on with it.

By Glen Maguire | AI Consultant & Change Practitioner | 24 October 2026 | 9 min read

I've run a lot of AI workshops. Most of them go well. People lean in, the demos land, and the feedback is kind. For a long time, I didn't ask the uncomfortable question: six months later, is anyone actually working differently?

We've trained hundreds of people across New Zealand and Australia to use Copilot and other AI tools, in sectors from construction to government. The sessions are practical and tailored, and people leave with real ideas. But if you asked me whether those people are working differently six months later, my honest answer, in most cases, is that I don't know. Often nobody does, because nobody was set up to find out.

After 25 years in digital transformation, I should have known better. Technology doesn't change organisations. People changing how they work does. That realisation is what took me back into the classroom to complete my Prosci change management certification. It's also why I want to tell you about Sarah.

Sarah isn't a real person. She's a composite of people I've met in workshops, team meetings and corridor conversations over the last two years. If you lead HR or people experience, you'll recognise her.

The email

Sarah is an HR coordinator at a 300-person New Zealand business. She's been there 15 years. She knows where every policy lives, and she's the person managers ring at 4.55pm on a Friday when something has gone wrong.

On a Monday morning, an email arrives from the CEO. The company has invested in Microsoft Copilot for all staff. From 1 November, everyone is expected to use it in their day-to-day work. Training details will follow.

Sarah reads it twice. She's seen the headlines about AI and administrative roles. That night she doesn't sleep well.

Her manager, Mark, forwards the email to the team with two words: "FYI, below." Mark has his own questions about what this means for his team, and for him. He doesn't ask them either.

The workshop

The workshop is good. The trainer is engaging and the demos are impressive. Sarah watches Copilot summarise a policy in 20 seconds. That job would normally take her most of an afternoon.

Around her, people laugh and say "wow". Sarah feels something closer to dread. If it can do that, what am I here for?

This is the part most AI training misses. Nobody asks how people feel about what they've just seen.

The quiet retreat

Six months later

The licence renewal paper goes to the executive team. Usage is well below expectations, and the CFO asks whether it's worth continuing. Someone points out that the training went really well. Everyone nods.

Sarah now uses Copilot to summarise long email threads, and that's it. She's quietly concluded it isn't meant for people like her.

What actually went wrong

Change practitioners use a model called ADKAR, from Prosci. It describes five outcomes people need to achieve for a change to succeed and last. They build on one another, although people don't always move through them in a straight line. When someone hits a barrier at an earlier stage, the later ones become hard to reach, so practitioners look for the earliest barrier and deal with it first.

Sarah is fictional, so this is an illustration rather than a formal assessment. But the pattern is common: her organisation focused almost entirely on Knowledge, through a workshop, and skipped almost everything around it.

In my experience, fear about job security is one of the biggest barriers to AI adoption, but it isn't the only one. People also hold back when they don't understand why a change is needed, don't trust the output, or can't see how it improves the work they already do. None of those concerns is irrational. (I've written more about this in AI in the Workplace: Why Change Management Is the Key.)

The CEO who thought they'd sponsored it

The CEO wasn't careless. They approved the investment, funded the training and wrote the email. In their mind, that was sponsorship.

This is extremely common. In Prosci's research, around half of participants reported sponsors with less than an adequate understanding of their change sponsorship responsibilities. Most sponsors aren't unwilling. They simply assume that once they've signed off and funded something, the job is done.

It matters more than almost anything else. In every Prosci benchmarking study since 1998, active and visible sponsorship has been the number one contributor to success.

Real sponsorship is turning up, more than once. It's being honest about what the change means, and visibly using the thing yourself.

There's an AI-specific trap here too. Copilot usually arrives as a software licence, so IT often ends up owning it. An AI initiative needs an accountable executive sponsor with the authority to influence business outcomes, supported by a coalition of leaders across the affected functions. IT has an important role, but owning the technology isn't the same as leading the change.

The manager who forwarded the email

Mark isn't the villain either. He was promoted because he's good at his job, not because he knows how to lead people through change. Nobody told him his team would look to him for answers. Nobody gave him the words, or the time.

Yet people want to hear what a change means for them personally from their direct manager, not from the CEO. Mark was the most important person in Sarah's adoption, and the least prepared.

Prosci identifies five roles people managers need to play during change. Mark wasn't prepared for any of them:

That isn't a failure of character. It's a failure to equip managers for a responsibility the organisation depends on.

I hear the same story from other change practitioners and in my own client work: managers who mention it in a team meeting, or simply forward the email. It isn't the exception. It's the norm. And when people have nowhere proper to take their concerns, they take them somewhere else: to the lunch table, to each other, sometimes to the union.

Rewind: the same six months, done differently

Same Sarah, same Copilot licence, same workshop. A few different decisions.

None of this requires a major transformation programme. But it does require deliberate investment in leadership time, employee support and measurement, which is exactly why it gets skipped.

Would Sarah have become an enthusiastic Copilot user? Possibly. Would every concern about her future have gone away? Certainly not. But she would have understood the change, had a voice in it, and had the support to make an informed transition.

The measurement gap nobody talks about

Microsoft has invested heavily in Copilot analytics, and other AI platforms offer their own versions. But there's a difference between measuring activity and knowing whether someone can use AI well.

So when a licence renewal paper says "usage is up", that tells you very little. Sarah summarising email threads every day counts as an active user.

Access alone isn't enough. The harder question is whether people are using AI well, and most organisations can't yet answer it. Measuring effectiveness means looking at what actually matters in real work. Can someone judge whether an output is good enough? Do they check facts before they rely on them? Do they know what information shouldn't go in, when to escalate, and when not to use AI at all?

That's the gap I'm increasingly focused on: measuring what people can actually do with AI, not just how often they use it. The approach I'm testing is a simple cycle:

It's early, and I'll share what I learn. But I'm convinced this is the missing piece.

Four questions for HR and people leaders

You're closer to this than anyone. You own manager capability, workforce impact and the employee experience. Before your next AI rollout, or as a check on the one you've already started, I'd ask:

If the honest answer to any of these is "not sure", that's where your adoption risk sits.

Where I've landed

I still run workshops, and I still believe in them. Knowledge matters. But a workshop on its own mainly supports one of the five things people need.

The question clients ask me most now isn't "how does Copilot work?" It's "what comes next after the training?" So I've changed how I design the work. AI capability building now comes with the change built in. That means working with sponsors before the email goes out, equipping managers for the conversations only they can have, and being clear about which tasks stay human-led, which become AI-assisted, and which may eventually be handled by agents. And it means measuring whether any of it has actually stuck.

I can't do the sponsor's job, or the manager's job, for them. Nobody external can. But I can make sure they know what that job is, and give them what they need to do it.

Most organisations have a Sarah. Usually more than one.

Book a conversation Explore AI Change Management

Prosci® and ADKAR® are registered trademarks of Prosci, Inc. Microsoft 365 Copilot is a trademark of Microsoft Corporation. Sarah and Mark are fictional composites.