Week two is where AI tools go to die.
Week one always looks good. The tool is new, someone senior is paying attention, and everybody is a little curious. Usage is high and the mood is positive.
Week two is the real test, and it is where most tools quietly stop being used. Not dramatically. Nobody announces it. Usage just tapers, and the story becomes "we tried it, it did not really fit us."
What changes between week one and week two.
Three things, all of them predictable.
The novelty is gone, so curiosity stops carrying it. Attention moves on, because whoever was watching has something else to watch now. And the first genuinely awkward case arrives, the one the demo did not cover, on a day when there is no time to work it out.
None of that is a flaw in the software. It is the normal shape of adoption, and it can be planned for.
The habit that predicts survival.
There is one question that predicts whether a tool is still in use at week eight, and it is not about features or training quality.
Does the tool sit inside work that someone already does, or beside it?
Inside means it appears in the flow the person is already in. The message arrives where they already look. The booking lands in the calendar they already open. Using it is the path of least resistance.
Beside means it is another tab, another login, another thing to remember. Anything beside the work depends on discipline, and discipline loses to a busy Tuesday every time.
This is why we prefer to fit AI into the tools a business already uses rather than replace them. Not because new tools are worse, but because a new tool has to win a habit war before it can deliver a single benefit.
How to check, honestly.
Do not ask the team whether they like it. They will be polite. Ask two concrete questions instead.
When did you last use it, and what did you do the last time it was awkward? The second answer is the important one. If they worked around it, you have found the exact thing to fix, and it is almost always small.
Look at real usage rather than opinion where you can. Not to police anyone, just to see whether the line held.
What we do about it.
We come back in week two. That is the whole intervention, and it is unglamorous.
By then the awkward case has happened and the workaround has started. Fixing it takes an afternoon at that point. Left alone for a month it becomes the reason the system failed, and by then nobody remembers it started with one confusing screen on a busy day.