Three moves decide whether AI lands with a team: start with the willing rather than the loudest, give each role one real task instead of a tool and a prayer, then make it safe to be bad at it for a fortnight. Most AI rollouts fail in the room, not the platform. The fear of being replaced, the fear of looking slow and one well-armed sceptic will sink better technology than yours.
And remember what you’re actually announcing. Your people are already using AI, quietly. The rollout isn’t an introduction. It’s a coming-out party for work that’s already happening, plus the rules and training that make it safe.
First, about that 70% statistic
You’ve heard it in every second conference talk: 70% of change programs fail. I’m not going to build this article on it, because that number is wobbly. It traces back decades, the original methodology is thin and researchers have spent years poking holes in it. It survives because it’s scary and round, which are two qualities evidence doesn’t need.
Here’s what holds up instead. Only 18% of employees actively support change from the outset; most sit neutral and wait. The top reasons people resist are lack of trust in leadership (41%), not understanding why the change is happening (39%) and fear of the unknown (38%). Read that list again: every one of those is a leadership communication problem wearing a staff-resistance costume. And McKinsey’s finding is the practical one: organisations that build skills through piloting and small experiments are three times more likely to succeed than the ones that mandate a broad rollout.
Why do mandates bounce?
Because AI isn’t like the last software you rolled out. The old accounting package worked the same for everyone who clicked the same buttons. AI output rises and falls with the skill of the person asking, which means a mandate without training doesn’t level the field. It makes your least confident people look worse in public, and they know it before you do. That’s where the eye-rolls come from. They’re not cynicism about the technology. They’re self-protection.
Meanwhile the baseline has moved under everyone’s feet: more than half of workers globally now use AI at work, up seven points in a year. The question your team is silently asking isn’t whether AI is coming. It’s whether it’s safe to be seen learning it here.
The three moves
- Start with the willing, not the loudest. Every team has two or three people already experimenting at home. Find them, give them cover and let them build the first proof. Visible wins from peers move a room faster than any leadership announcement, and projects with real employee backing run at a 30% higher success rate.
- Pick one real task per role. Not AI training in the abstract. The bookkeeper gets receipt sorting. The coordinator gets meeting follow-ups. Reception gets the enquiry inbox. Success on one small, real task is the most reliable change agent there is, because it answers the only question that matters: what does this do for my Tuesday?
- Make it safe to be bad at it for a fortnight. Say it out loud, in those words. Nobody’s output gets judged for two weeks; questions are free; the person who breaks something interesting buys the story, not the blame. Ongoing training through a change lifts adoption by more than half, and psychological safety is the ingredient that makes the training stick.
What the sceptic is telling you
There’s one in every room, usually senior, and the standard advice is to route around them. Don’t. Senior scepticism is usually readiness intelligence in disguise: they can see integration mess, client risk and workload reality that the enthusiasts can’t. Hand them the quality-control seat. Ask them to define what good enough looks like before anything AI-assisted goes near a client. You’ve just turned your biggest blocker into your standards department, and they’re usually right about at least two things.
And give the whole room the finding worth more than any adoption statistic: PwC’s 2026 jobs research shows the new tasks appearing in AI-exposed roles are two and a half times more likely to rely on empathy, judgement and creativity. The human skills don’t shrink as AI takes the routine work. They become the job. That’s the Augmented Workforce, and it’s a better story than efficiency ever was.
Where REIMAGINE starts
The three moves above are the front edge of my REIMAGINE framework, the nine-step change arc from my book AI & U: Reimagine Business. The first three steps do the heavy lifting for a team rollout. Relationships: map who’s affected and who’s willing before you touch a tool. Evaluate: get honest about what’s already in use, declared or not; the amnesty conversationdoes this in a month. Identify: choose the one real task per role, picked with the people who’ll do it rather than for them, because six in ten workers say they want a say before the plan is final, not after.
The other six steps carry a team from first experiments to a durable habit, and they’re the spine of how I run team programs. When Gymnastics WA brought their people through this arc, the shift wasn’t the tools. It was a team that stopped waiting for permission.
What to do on Monday morning
- Name your willing two. You already know who they are. Give them one task, cover and a fortnight.
- Say the safety sentence to the whole team, in plain words: nobody gets judged on AI output for two weeks, and questions are free.
- Put the sceptic in charge of quality. One conversation, this week, framed as I need your standards, not your silence.
And if you want a partner walking the arc with your team rather than a to-do list, that’s the work I do.
A partner for the arc, not a to-do list.
One real task per role, cover for the willing and a team that stops waiting for permission. That's the work I do with teams, from first experiments to a durable habit.
Work with me →Questions people ask
How long does it take a team to adopt AI?
First useful habits: a fortnight per person, per task. Team-wide comfort: three to six months with a steady training rhythm. Anyone promising a whole-company turnaround in a month is selling the announcement, not the adoption.
Should AI training be mandatory?
Split the question. Training on the rules (what never goes into AI tools, when a human reviews) is mandatory from day one; that's Practical Training on your guardrails and it isn't optional. Training on the tools starts voluntary with the willing, then rolls to everyone once there's local proof. Mandate the guardrails, invite the skills.
What if senior staff resist hardest?
Treat it as information, not obstruction. Senior resisters usually see complexity and client risk that early adopters can't. Give them the standards role: they define what good enough looks like before AI-assisted work ships. Most convert the moment their experience is treated as the asset it is.
Do we train everyone at once?
No. Pilot with the willing first; the piloting path succeeds at three times the rate of broad rollouts. Then move role by role, each wave trained on a real task with proof from the wave before. Slower on paper, faster in reality.
Human-led. AI-leveraged. My philosophy, my business, this article. The Augmented Workforce in action.
Drafted with Ada, my AI collaborator. Reviewed, shaped and signed off by me. How I work with AI· Tracy Sheen CSP