Leading AI Adoption: The Rollout Is a People Project

The rollout plan said the new AI tool would go live in March, and it did — for everybody, on paper. Six months later, a third of the team uses it every day, a third tried it twice, and a third has quietly routed around it. The software was identical in all three cases. What differed was everything around it: whether the case for change was believed, whether the rules were clear, whether the work itself was redesigned, whether anyone was helped past the awkward first month. Adoption is a people project with a technology component. Leaders who run it the other way around keep being surprised by the same result.

The honest case, made first

Adoption starts with why, and the why has to survive contact with a skeptical room. That means the honest case: what this tool will do, what it will not do, the full costs, the plain limits — and no guaranteed numbers waved in front of people who will be living with the actual results. People can forgive a hard change. They do not forgive being sold one, and the discovery that the case was inflated poisons every rollout that follows it. The skeptic in the third row is not the obstacle to this work; the skeptic is often the one who has found the real risk. A leader who hears skeptics early gets the risk while it is still cheap to fix.

Choosing where to start is part of the case. The pilots that teach the most begin where pain and feasibility meet: a process people already complain about, where the tool’s strengths fit the work and success would be visible to the skeptics in their own terms. Start with a showcase chosen for the announcement and you get a demonstration. Start where the work hurts and you get evidence — and converts, because nothing recruits the next team like the first team describing, unasked, what got easier.

Guardrails before pilots

Enthusiasm launches pilots; policy is what makes them safe to learn from. Before the first experiment: what never enters a tool, what data rules apply, who decides, and which named human owns every output the tool produces. Set while the stakes are low, guardrails are not brakes — they are what lets a team move quickly later without a public accident to explain. The organizations that stall are rarely the ones with the most rules. They are the ones where nobody knows what the rules are, so everybody quietly plays it safe, including with the uses that would have paid for the whole program.

Redesign the work, don’t just add the tool

Bolting AI onto an unchanged workflow buys the old process at a new cost. The adoption work is redesign: which steps disappear, which judgments stay human, how the handoffs change, what the job looks like afterward — said straight, including the uncomfortable parts about skills and roles, because the room is already thinking them. Training has to match that ambition: built for adoption rather than launch, practice on real work over weeks, help available inside the awkward middle. A launch-day webinar produces launch-day knowledge. It does not produce a changed Tuesday.

Measure like an auditor, scale like an investor

The last discipline is measurement with the vanity left out. Logins are not adoption; licenses are not value. What counts is use against a baseline: the work actually flowing through the new process, the quality held, the time honestly accounted for. Pilots that fail get owned out loud — each owned failure buys credibility for the next change, because the organization watches what leadership does with a failed pilot far more closely than what it says about a successful one. Then scale only what demonstrably works, and retire what doesn’t, visibly. That is the whole project: an honest case, clear guardrails, redesigned work, evidence over enthusiasm. The technology was the easy part. It usually is.

← All Insights