Copilot licences give people access to AI inside Microsoft 365. Access is not adoption. The value shows up only when a specific workflow runs differently: fewer steps, less waiting, less rework. To get there, find out where your team's hours go, pick two or three workflows where Copilot or another AI tool fits, redesign them with the people who do the work, and measure them before and after.
What the numbers say about Copilot rollouts
Most companies on Microsoft 365 have tried Copilot by now. Few have turned it into something the board can see.
In a 2024 Gartner survey of IT leaders, 60% were running Microsoft 365 Copilot pilots, but only 6% had finished a pilot and were moving to a large-scale rollout. Respondents said employees were keen to keep access, yet 72% said people struggle to fit it into their daily routines, and 57% said engagement drops quickly (Gartner). A year later, in Gartner's 2025 survey of 187 IT leaders, 40% were still piloting, and of those that had finished a pilot, only 5% were moving to a larger deployment. Measuring the value remained the hard part (techpartner.news).
These surveys mostly cover larger organisations, and they ask IT leaders rather than the business. But the pattern is familiar to anyone who has watched a rollout: an exciting first month, a slow fade, and a renewal decision made on a feeling.
Why licences do not change how work runs
Copilot is an assistant. It helps when someone asks. It summarises the email you open, drafts the reply you request, finds the document you search for. That is useful, and for some people it saves real time.
But a workflow is not a series of requests. It is a sequence of steps that moves work from one person or system to the next: an application arrives, documents get checked, details go into the CRM, a file goes to an adviser. An assistant can make one of those steps a little faster for the person who remembers to use it. It does not change the sequence, remove the handover or move the information into the right system by itself.
Take a client onboarding workflow that costs 48 minutes per client. With Copilot, the person who reads the incoming email gets a summary and saves two minutes. The other 46 minutes are still there: the checking, the chasing, the retyping, the handover. Multiply that by a hundred employees and you get many small gains spread thinly across the company, and almost nothing a board can point to.
The adoption curve nobody plans for
Most rollouts follow the same curve. In the first weeks, people try everything. They summarise meetings and rewrite emails and ask questions about documents. Usage looks great.
Around week six, the curve bends. Using the assistant is an extra step in a day that is already full, and when the rest of the workflow has not changed, the extra step feels optional. People return to the habits that the rest of their work is built around. This is the drop in engagement that Gartner's respondents describe, and it is not a failure of the tool. It is what happens when a tool is added to work instead of work being changed.
Usage is not value
Microsoft's usage reports show who is active, how often, and in which apps. They answer a question about the tool. Boards ask a different question.
| What a usage dashboard shows | What the board wants to know |
|---|---|
| Active users per week | Which work takes less time now |
| Prompts and actions per user | How much faster clients get an answer |
| Usage per app | Whether rework went down |
| Licences assigned and unused | Whether the same team handles more work |
Usage is a useful signal. It shows where people are curious and where they are not. But it is never the result. The result is a workflow that runs differently, measured the same way before and after. We describe that method in how to measure AI ROI.
Where Copilot fits well, and where it does not
Copilot is strongest inside the Microsoft apps, for one person at a time. It fits well when someone needs to summarise a long thread, turn a meeting into actions, draft a document from files they can already see, or analyse a spreadsheet.
It fits less well when the work is a repeated, multi-step process that crosses systems: reading every new email in a shared inbox, checking each document against a checklist, filling the CRM and accounting system, and handing the case to the next team, the same way every time, without someone prompting it. That kind of work needs a designed workflow: clear steps, connections to the systems involved, a person who checks the output, and a log of every run. Microsoft's own agent tools can cover part of this, but they still need someone to design the workflow first.
The question is never "Copilot or not". It is: for this workflow, which steps should AI do, with which tool, and where does a person check?
How to turn licences into changed work
You do not need to start over. You need to start from the work instead of the tool.
- Find out where the hours go. Ask every team what takes the most time, where they wait for others, and what they copy, check or chase by hand.
- Pick two or three workflows. Choose work that repeats, costs hours and moves between people. See which processes to automate with AI first.
- Decide per step. For each step, decide: Copilot for the person, an automated step, AI with a human check, or no change.
- Redesign the workflow and put it where people work. In Outlook, Teams, SharePoint and the CRM, not in a separate place people must remember.
- Train inside the workflow. Show people how the new version of their own work runs, not how to write prompts in general.
- Measure before and after. Minutes per case, waiting time and rework. Let finance confirm.
Train in the workflow, not in the tool
Most Copilot training teaches the tool: how to write a prompt, what each app can do, tips and tricks. It is well meant, and most of it is forgotten within a month, because it is not tied to anyone's actual work.
Training inside the workflow is different. The onboarding team learns how the new onboarding runs: the AI checks the documents, they review the flags, they approve the email to the client. They do not need to remember a prompt, because the prompt is part of the workflow. That is what makes a change stick after the first enthusiasm fades.
Should you cut licences?
Not blindly. Some people get real value from Copilot every day, and cutting their access would cost more than it saves. Look at usage per team next to the workflows you mapped. Keep licences where they support a workflow you are changing or where people use them every day. Move or drop the rest. It is fine that not every employee needs an AI assistant. It is not fine to pay for access that changes nothing.
One thing to fix first: what Copilot can see
Copilot can see what each user can see. If permissions in SharePoint and Teams are too broad, Copilot will surface documents people were never meant to read. Gartner's 2024 survey found that oversharing and security concerns delayed rollouts by three months or more at 40% of organisations (Computerworld). Clean up access to the sites your first workflows use before you scale, and agree clear rules for what may go into AI tools. That is part of our governance work.
Frequently asked questions
Is Copilot worth the money?
For some people and some workflows, yes. Across a whole company, only if it changes specific workflows you can measure. Start there, then decide which licences to keep.
How do we measure Copilot ROI?
Not through usage. Pick a workflow, time it before you change it, and measure it again the same way after. The full method is in how to measure AI ROI.
We have Copilot and ChatGPT. Which should we use?
Decide per workflow, not per company. Choose the tool that fits the steps and the data, and settle the rules first: which data may go where, and who approves a new tool.
We barely use Copilot. Did we waste the money?
Not if you use this moment to map your work. The licences are a head start once you know where they belong.
Paying for Copilot without seeing the change? Book a 30-minute call. For the bigger picture, read where to start with AI in a mid-size company.