Why Copilot Usage Collapses Around Month Three

Here is the pattern we keep seeing.

Launch week looks busy. People click around. A few power users dig in. Leadership sees logins and calls it progress.

Then the quiet drop. By around month three, novelty has worn off. Seats stay assigned. Daily work looks a lot like it did before the rollout. Licenses still show as “live.” Usage does not.

That is usually not a tools problem. And it is usually not a people problem in the way leadership first names it.

Novelty is not enablement

A kickoff, a demo, and a seat assignment create a spike. They do not create a way of working — or clear ownership for what happens after the meeting ends.

Enablement that sticks starts from the job: a real report, a real dataset, a process someone already owns. Not from another tour of features. When the starting point is the product, Tuesday does not change. When the starting point is the work, people have somewhere to put the tool.

Policy and council work the same way. They answer the organizational question. They do not answer Monday morning for the person who still does not know what “good” looks like in their world — or who owns the outcome if something goes wrong.

More on that gap: We Gave Everyone ChatGPT. Now What?

Month three is when search-only use becomes visible

Early on, almost any activity looks like adoption. Open the tool. Ask something light. Feel like you participated.

A few months in, the split is clearer. Some people are using Copilot inside real work. Many are using it like search when they remember it exists — or not at all. The higher-stakes work stays untouched because nobody showed what safe, expected use looks like on the job that matters.

From the outside, the license report still looks okay. From inside the building, the work barely moved.

That quieter cost — confidence eroding while outcomes stay stuck in pockets — is what we meant here: The Hidden Cost of Uneven AI Adoption.