AI change management: getting your team to actually use it

The data says the same. Deloitte Access Economics and Amazon found in November 2025 that two-thirds of Australian SMBs use AI, but only about 5% are “fully AI-enabled” - strategy embedded, staff trained, data organised. The rest have access and shallow use. Meanwhile MYOB’s April 2026 figures show AI-using SMEs growing 2.8 times faster than the rest, with 54% reporting real time saved. The gap between having AI and getting value from it is almost entirely change management.

So treat it as a change-management project, not an IT rollout. Here is what actually works.

Name champions in each team

Peer proof beats a top-down email every time. People adopt a new way of working when someone they respect shows them it saves time.

Pick one champion per team. They do not need to be technical - they need to be respected and genuinely keen. Give them a little extra training and some protected time, and make them the first port of call for “how would I do this with AI”. Adoption then spreads sideways through the team, which is far stronger than spreading only downward.

Make AI the default for specific, named tasks

Vague encouragement to “use AI more” goes nowhere - it asks people to invent their own use cases on top of a full workload. They won’t.

Instead, name the tasks. “First drafts of client emails start in Claude.” “Meeting notes get summarised by AI before they go in the file.” “Quote responses begin from the AI draft, not a blank page.” Pick one, make AI the standard way to do it, let it bed in, then add the next. The unit of adoption is one default task, not the whole business at once.

Train on real work, not toy demos

A generic “intro to AI” webinar where everyone writes a haiku teaches nothing that sticks, because it has nothing to do with their day.

Run hands-on sessions on the team’s own actual work. Bring the real emails, quotes and reports, and have people do their genuine tasks with the tool, with the champion and a facilitator in the room. The moment someone watches a task they hate take two minutes instead of twenty - on their own work - is when adoption becomes real.

Remove the friction

Every extra click, login or moment of “wait, am I allowed to put this in here” is a reason to fall back to the old way. Friction is where good intentions quietly die.

Three things to fix:

• The right tools. Make sure people have the paid tool that fits their job, on a no-training business tier, not a personal free account they’re nervous to use for work.
• Easy access. The tool should be one click from where they work, not buried behind three steps. Pin it, integrate it, make it the path of least resistance.
• Clear permissions. Write down plainly what data can go into which tool and what is off-limits. Ambiguity makes careful people freeze. Clarity sets them free to use it.

Leaders model it visibly

If the founder and the managers do not visibly use AI, the team reads that as the real signal, whatever the email said. You cannot delegate the culture change.

Use it in front of people. Show the draft you started in Claude in the meeting. Mention the analysis you ran this morning. Talk about what worked and what did not. Visible use from the top gives everyone permission and a model to copy. Quiet non-use gives them permission to ignore the whole thing.

Measure actual usage, not licences bought

Licences purchased tells you what you spent. It tells you nothing about what changed.

Look at the real signals in your AI tool’s admin console: weekly active users, which teams are live versus dormant, whether the named default tasks are actually being done with AI. Pair that with the outcome that matters - wage-equivalent hours saved or cycle-time on the workflows you targeted. Fifty licences and eight weekly active users is an adoption problem you can now see and fix, not a tooling decision to second-guess.

Address the fear honestly

Two fears sit under most quiet non-adoption, and both deserve a straight answer.

The first is job security. Some people reasonably suspect that using AI well makes their role easier to cut. Pretending the fear does not exist just drives it underground. Be honest about the actual intent. For most SMBs that is removing the tedious parts of a job so people do higher-value work, not shrinking the team. Say it plainly, then act consistently with it.

The second is “is this even allowed”. Careful staff self-censor when the rules are unclear, so they avoid the tool to stay safe. The fix is the clear, written permission from the friction section: what data goes where, what is off-limits, who to ask. People use a tool freely once they trust it will not get them in trouble.

The honest bottom line

Mandates without enablement fail. “Everyone must use AI” with no defaults, no training and no friction removed simply teaches people to say they are using it while they quietly are not - and now the dashboard lies to you too.

The expectation from leadership gives permission and direction. The enablement - champions, defaults, real-work training, low friction, honest answers - makes it happen. You need both, and the second is the part almost everyone skips.