People change how they work when the new way makes a bad part of their week shorter. That is the whole mechanism. Training does not do it. A mandate does not do it. An internal comms campaign definitely does not do it.
Start with what people complain about
Ask a team what part of their week they resent. Do not ask what they would automate, because that question gets you a wish list of things nobody actually does.
The complaint is the use case. It is already emotionally loaded, which means the first person who gets relief will tell everyone else without being asked.
Give one team the whole problem, not the tool
The mistake is handing over a tool while the process around it still belongs to someone else. The team then uses it up to the boundary of their authority and stops exactly where the saving would have started.
Give them the tool and permission to change the shape of the job. Those two together are what people mean by adoption. One without the other is just software with a login.
Measure the task before you touch it
Time it. Count it. Write it down. Ten minutes with a stopwatch is enough.
Without a before, you get a debate about whether it feels faster, and feelings do not survive a budget review.
Let the early version be visibly imperfect
If the first version is presented as finished, every flaw becomes evidence against the whole idea. If it is presented as a draft with a human checking the output, the same flaws become normal.
Keep a person approving the output for the first few weeks. It buys trust cheaply, and it catches the failure modes you did not think of.
What I have seen work
- One team, one task they already resent
- The owner can change the process, not only the tool
- A number from before, however rough
- A human in the loop for the first two weeks
- Then let the first team tell the second team, rather than announcing it
I spent years running a partner program that grew from around four thousand partners to more than fifteen thousand. Nothing in that was a technology problem either. Getting people to work differently is the same job it has always been, and AI has not changed it.