Every business owner I speak to has one: a tool that was rolled out with great enthusiasm once and now sits unused on a server. Paid for, installed, trained on, gathering dust.
The reflex is that the software was no good. Usually the software was fine.
What actually happened
You bought something that was finished on day one. There was an afternoon of training, everyone nodded, and after that it was supposed to run.
That is not how it works with a new member of staff. You do not put someone on a job alone after an afternoon of explanation either. They watch for the first few weeks, do the work under supervision, make mistakes with someone there to see it, and only once you trust them do you let them get on with it.
A digital colleague deserves the same treatment, and for exactly the same reason: things go wrong in the first few weeks, and someone has to see it.
How we do it
On a mail routing system we built, the agreement was that a person approved every decision for the first two weeks. The AI proposed where the mail should go, someone confirmed or corrected it. Only above ninety percent accuracy did it go automatic.
In those first two weeks that cost more time than it saved. That is not a bug in the plan, that is the plan.
What you get back is that the team has watched the tool learn. They know what it is good at and where it goes off the rails. And that is the only reason anyone still trusts it three months later.
Three things that make the difference
Pick the job people hate. Not the job with the prettiest business case. Whoever gets rid of their biggest irritation turns into an ambassador all by themselves. Whoever gets a four percent saving on something neutral notices nothing.
Let the AI escalate when it is unsure. A tool that sometimes says "I do not know this one, have a look" gets trusted. A tool that always gives an answer, even when it is nonsense, gets ignored after the second blunder. Permanently.
Give it a name and an owner. Sounds soft. It is not. As long as nobody owns it, it is "that system", and nobody is bothered when it stops working.
Why this is a question for the board
If the problem is implementation, it is an IT job you outsource. If the problem is adoption, it is about how people work, what they dare to do and who they trust. That you cannot outsource.
That is why I have been putting it this way for a while: AI is not an IT project, it is a leadership question.
A tool nobody uses is not a failed implementation. It is a failed training period.