Every large organisation currently has an AI strategy document. Very few have AI adoption. The gap between the two is not a technology gap. It is a management gap, and it is where I spent several years of my career inside a global retailer, leading AI adoption across teams that had not asked for it and mostly did not want it.
Here is what actually moves the needle, and what reliably does not.
Mandates create compliance, not capability
The fastest way to kill an AI program is to make it mandatory. People will attend the training, tick the box, and go back to exactly how they worked before, now with a slightly worse opinion of leadership.
What worked instead was finding the one genuinely painful task in each team's week and solving that first. Not the impressive use case. The annoying one. The report that took a day to compile, the transcript nobody wanted to summarise, the analysis that got skipped because it was tedious. When AI removes something people hate, they come back on their own, and then they start experimenting, and the experimentation is where the real value lives.
Adoption spreads through demonstrated relief, not through slideware.
Middle managers are the bottleneck and the unlock
Executive sponsorship gets an AI program funded. Frontline enthusiasm gets it tried. But it is middle managers who decide whether it survives, because they control the two scarcest resources: their team's time and their team's permission to work differently.
A middle manager who feels threatened by AI, or who is measured purely on this quarter's throughput, will quietly starve the program without ever opposing it. The fix is not another town hall. It is changing what those managers are recognised for. When “my team found a better way to do this” earns visible credit, managers become distributors of the change instead of dampeners of it.
If your adoption program has budget for either an extra vendor workshop or dedicated time with line managers, choose the managers. Every time.
“Am I training my replacement?” deserves a straight answer
The question is in the room whether you acknowledge it or not. Evasive answers (“AI will augment, not replace”) are heard as confirmation of the fear, because people can tell when they are being handled.
The honest answer is more useful: some tasks will disappear, most roles will change, and the people who learn to direct these tools will be worth more, inside this company or anywhere else. Then back it with something real: time protected for learning, and visible examples of people whose roles grew rather than shrank. Psychological safety is not a poster. It is a pattern of evidence.
Measure decision quality, not time saved
“Hours saved” is the metric everyone reaches for and it is close to meaningless. Saved hours are invisible, unverifiable, and instantly absorbed. Worse, it frames AI as a cost-cutting tool, which confirms every fear in the building.
Better questions: Are decisions being made with evidence that previously went unexamined? Are analyses happening that used to be skipped? Is the quality bar of routine output rising? When we redesigned how consumer insight flowed into decisions, the win was not that reports got faster. It was that questions which previously died in a backlog started getting answered at all.
The leadership shift underneath it all
The deeper change is that leaders themselves have to become literate. Not expert, literate. A leader who has personally used these tools for real work asks sharper questions, spots vendor nonsense faster, and models the behaviour they are asking for. A leader who delegates all understanding to a transformation team is sponsoring something they cannot steer.
That literacy matters more now than it did even a year ago, because the technology has moved from tools people use to agents that act. The management questions get harder from here: what is this system authorised to do, who is accountable when it acts, how do we supervise work we no longer perform? Organisations that never solved the adoption problem for chat-era AI are about to face the governance problem of agent-era AI with no muscle built.
The strategy document will not save you. The management work will.