You bought Copilot and nobody uses it. Four causes and a 30-day plan

Where I know this from

At the company where I own Salesforce, I took part in rolling out AI in finance, accounting, administration and IT. The tools were good. The difference between a department where AI took hold within a month and one where the licences sat idle was not in the tool. It was in the four things below.

Cause 1: nobody has a task

“Use Copilot” is not a task. An employee opens the window, does not know what to ask, asks something generic, gets a generic answer and closes it. They will not open it a second time. AI takes hold where someone has a specific, repetitive chore that wears them down: the weekly report, the same customer questions, comparing two lists.

What to do: every employee picks 2–3 tasks from their own work that they would like to do faster. Not “AI in marketing” but “product descriptions for the shop, 20 a week”.

Cause 2: nobody has seen an example from their own work

The vendor demo shows AI writing a poem or summarising a board meeting. The accountant does not write poems. Until she sees AI reading a batch of invoices into a table, and where it gets that wrong, the tool is a curiosity to her.

What to do: show every department three examples from its own work, on its own (anonymised) documents, mistakes included. The mistakes matter: an employee who has seen AI make something up trusts it in the right way.

Cause 3: fear about data that nobody has defused

Staff have heard that “you must not paste data into ChatGPT”. Nobody explained to them how a business plan differs from a personal account, which data is fine and which is not, or how to anonymise. A sensible person in that position does not use the tool at all. A less sensible one uses a personal account, because “nobody is watching” there.

What to do: one page of rules: tools, accounts, data, responsibility, a contact person. An hour of training where everyone sees anonymisation done live. A business account for everyone who is going to work on company data.

Cause 4: nobody is driving it

A rollout whose only owner is the IT department ends at installation. In the department where AI took hold there was a person who asked every week “what did you do with AI, what did not work”, collected the prompts that worked and showed them to the rest. Not an IT person. Someone from the department who liked the tool and had an hour a week for it.

What to do: appoint such a person in every department, give them an hour a week and a simple channel (a chat, a document) for collecting examples.

The 30-day plan

Week	What happens	Who
1	A survey: who uses what, 2–3 tasks per person. One page of AI usage rules. Business accounts for everyone who needs one.	Management, IT, the champion
2	Training: a shared part on safety and verifying output, then group work on the tasks from the survey. Everyone leaves with a working prompt for their own task.	The whole team
3	Daily use. The champion collects what worked and what did not. Questions go to the trainer by email.	The team, the champion
4	A short meeting: a demo of 3–5 prompts that work, fixes to the ones that do not. A decision on what next: which tasks to automate, whether to extend to other departments.	The team, management

How you know it is working

• After a month staff bring new uses on their own; nobody has to ask.
• The company prompt document keeps gaining entries from different people.
• Someone reports an AI mistake they caught. That is a good sign: they are using it and checking.
• Tasks labelled “this could be automated” appear on the list. That is the moment to talk about implementation.

Weeks 1 and 2 of this plan are exactly how I build a training: the survey, the rules, a workshop on your own tasks. Weeks 3 and 4 are the 30 days of support and the session after a month.