Your Copilot licenses are fine.
Nobody is using them.
The most common call I get is not "help us choose an AI tool". It is "we bought Copilot 4 months ago, the launch went well, and now almost nobody opens it". Same story, different logo, over and over.
This page is what I have learned from the inside of more than 120 organizations as of mid-2026. Not a list of best practices. The specific reasons rollouts go quiet in month 2, and the sequence that actually pulls them back.
Start with the number
Recon Analytics surveyed more than 150,000 people in the US and found a workplace conversion rate of 35.8%. Roughly 1 in 3 employees who have Copilot actually use it.
Read that as a finance person for a second. At 30 dollars per user per month, an organization with 500 licenses is spending 180,000 dollars a year, and by that benchmark something like two thirds of it produces nothing. Nobody notices, because the invoice looks identical either way.
Why it always goes quiet in month 2
Month 1 looks great. There was a launch, there was energy, people tried it. Then the novelty runs out and everyone falls back to the way they already know how to do the job. That is not an AI problem. That is how every tool rollout has ever behaved.
What is specific to Copilot is that it does not force the change. Nobody has to open it. Teams did not have Copilot last year and shipped fine, so skipping it costs nothing today. A tool that is genuinely optional stays optional unless something makes it not.
The 5 reasons I actually see
01 Training was skipped, or it was a product demo
Someone from the vendor showed the features for an hour. Features are not the gap. Knowing what to type to get something useful back is the gap. If people leave a session without having produced something real on their own work, nothing transfers.
02 It was taught generically
A finance team and a communications team both live in Word and Outlook, and the examples that land for one do nothing for the other. Generic training gets polite feedback and no behaviour change. This is the most common thing I get called in to redo.
03 Nobody visible is using it
Adoption spreads by watching a colleague do something faster, not by email. If no one on the team has visibly solved a real problem with it, there is nothing to copy. And if the executives who mandated it are still working exactly as before, everyone reads that correctly.
04 The security question was never answered out loud
People quietly wonder whether it reads their mail, whether their documents train the model, whether IT can see their prompts. Unanswered, that becomes avoidance. Answered in 5 minutes with specifics, it disappears. (The answers exist. See the Copilot, GDPR and EU data residency page.)
05 Permissions were never cleaned up first
Copilot only surfaces what a user could already open. If your SharePoint has sites everyone can reach, someone will find that out in week 1, and the reaction is usually to slow the whole rollout down. Fixing this before launch is unglamorous and it is the thing that protects the project.
What actually moves the number
I will be honest about what does not: another all-hands announcement, a longer feature demo, or a dashboard that shows leadership how bad adoption is. Measuring the problem is not the same as fixing it.
Teach the specific team, on their own documents
Before a session I ask what the group does all day and collect real examples of their work (with sensitive content removed). We work on those. When someone sees Copilot handle something they recognise, it lands differently than any demo.
Spend real time on prompting
Knowing Copilot exists is not the gap. Knowing how to ask and how to iterate when the first answer is wrong is the gap, and it transfers to every surface and keeps working after the session.
Aim at the repeated task, not the impressive one
The weekly report, the recurring email, the meeting nobody wants to write up. Habits form on things that repeat. Nobody builds a habit around an impressive one-off.
Give managers something to model
Not a mandate. One or two things they do visibly, in front of the team, that the team can copy. This costs nothing and it is the single cheapest lever on this list.
Come back once, a few weeks later
The questions people have after 3 weeks of real use are completely different from the ones they have on day 1, and they are the questions that decide whether it sticks. A single follow-up session does more than doubling the length of the first one.
If you are already stalled
The instinct is to relaunch to everybody. That rarely works, because the people who already decided it is not for them will not come back for a second announcement.
What tends to work is smaller. Pick one team with a real repeating pain. Get that team genuinely using it, on their own work. Let the rest of the organization hear about it from them rather than from IT. It is slower on paper and faster in practice.
Questions I get asked about this
Is 35% adoption actually bad?
It is normal, which is not the same as fine. Normal is what happens when a tool is deployed without changing anything about how people work. The organizations that do better are not the ones with better licenses.
Should we just buy fewer licenses instead?
That is a reasonable question and sometimes the answer is yes, at least for a while. Assigning licenses to a group that is actually being supported beats spreading them thin across people who will not open it. I would rather tell you that on a call than sell you training you do not need yet.
How long does it take to see a change?
Weeks, not months, for a specific team that gets trained on its own work and has one follow-up. Organization-wide numbers move more slowly because they are an average of teams at very different stages.
Do you do this remotely?
Yes, and it is the default. I deliver workshops and rollout advisory live in English over Zoom for teams across Europe and elsewhere. Hands-on workshops are capped at 25 participants so I can actually help people who get stuck. Lectures have no such cap.
What if the problem turns out to be permissions, not training?
Then training is the wrong first spend and I will say so. That happens, and it is better to find out on an intro call than after a workshop.
Want a straight answer on where yours actually is?
An intro call takes 30 to 45 minutes. We go through what you rolled out, what the usage looks like now, and what the realistic next step is. Sometimes that is training. Sometimes it is not, and I will tell you that.