There's a version of AI adoption that almost always fails, and I see it constantly: leadership buys licences, sends an all-hands email, and waits for transformation. Three months later usage is a trickle, a few enthusiasts aside, and someone concludes the tool "didn't take." The tool was fine. The rollout was the problem, because there wasn't one.

Adoption is a behaviour-change project that happens to involve software. Treat it that way and the first 90 days look completely different. Here's an approach I've seen work across very different organisations, drawn from years of running change programmes, including a large intrapreneurship effort at ICI Pakistan where getting hundreds of people to actually do something new was the entire challenge.

Why the first 90 days decide it

Habits form or fail to form early. If people experience a concrete, personal win in the first few weeks, they pull the tool into their workflow on their own. If their first experience is a blank box, vague instructions and no obvious payoff, they conclude it's not for them, and re-engaging a sceptic is far harder than engaging a newcomer. The first quarter is where the trajectory gets set. Spend it deliberately.

Days 1-30: a narrow, real win

Resist the urge to roll out everything to everyone. Pick one team and one painful, high-frequency task they all recognise, the report nobody enjoys writing, the inbox triage that eats every morning. Narrow and real beats broad and abstract every time.

Then don't just hand over a login. Show them, concretely, how to do that specific task better with the tool. Sit with them. Let them feel the difference on work they actually care about. The goal for month one is simple: a meaningful number of people in one team having a genuine "oh, that's useful" moment on something that matters to them.

The principle

People don't adopt tools. They adopt better ways of doing work they already do. Lead with the work, not the technology, the technology is just how the better way arrives.

Days 31-60: build the habit

Now widen carefully. Take what worked with the first team and extend it, more tasks for them, the same proven approach to an adjacent team. This is also when you surface your internal champions: the people who took to it naturally. Give them a bit of visibility and a bit of time to help others. Peer enthusiasm travels further than any top-down mandate.

Crucially, start capturing what good use looks like in your specific context. The prompts that work. The tasks worth handing over and the ones to keep human. The early wins worth retelling. You're building the beginnings of an institutional playbook, not just running sessions.

Days 61-90: systematise

By now you should have real evidence, not vibes, but specifics about what's working and what it's worth. Use the final month to turn that into something durable. Light-touch guidelines so people know what's encouraged and what's off-limits. The governance guardrails that should have been sketched on day one, now firmed up. And a simple, honest read on impact: where is this saving real time, and is anyone actually doing their job differently?

That last question is the one that matters. Deployment is easy to claim. Adoption is the thing you're actually after, and by day 90 you should be able to point to specific work that genuinely changed.

The mistake to avoid

The single biggest error is treating the tool as the deliverable. "We rolled out AI" is not an outcome, it's a procurement event. The outcome is people doing better work, and that only happens when you invest in the human side as seriously as the technical one. Buy the licences in an afternoon if you like. But budget real attention for the first 90 days, because that's where adoption is genuinely won or quietly lost.