When companies hire me to run an AI workshop, most expect me to open ChatGPT on day one. I don't. For enterprise rooms I almost always start with Microsoft Copilot — the tool already inside their security boundary, already approved by IT, and usually already paid for.
This is a deliberate choice, and it changes how adoption plays out.
The shadow IT problem
Every enterprise I have worked with has the same open secret: employees are already using AI, on personal phones, with personal accounts, uploading company data into consumer tools with no data protection. At one Hong Kong manufacturer the IT lead had found exactly that before we started, and it set the rules for the whole program.
If I open ChatGPT in a corporate workshop, I validate that behavior. I am teaching a tool IT might block next month. If I open Copilot, I am teaching inside the fence. There is no debate about "is this allowed?" It is allowed, it is integrated with their email, their documents, and their calendar, and the session goes on the work instead of the argument.
The adoption curve is steeper with ChatGPT
ChatGPT is a standalone tool. You open a new tab, switch context, copy data in, copy results out. Copilot lives inside Word, Outlook, Excel, and Teams. The gap sounds small. In a room of finance staff at four in the afternoon it is the difference between a tool that gets used on Thursday and one that gets admired on Tuesday.
When I worked with PolyU's finance team, the breakthrough came when AI stopped being a separate activity and became part of the Excel file and the Outlook draft they were already in. That integration is what turns occasional users into daily users.
When I do teach ChatGPT
I am not dogmatic about this. ChatGPT and Gemini are better at some things — creative brainstorming, complex analysis, image generation. At Chow Tai Fook's design workshop we used four different AI tools, because jewelry design needs capabilities Copilot does not have.
The principle is to start with the tool that has the lowest barrier to sustained daily use, then add specialized tools for specialized workflows. For most enterprises the starting point is Copilot.
One caveat I have learned to say out loud: check who actually has a license. "Already paid for" is only true if someone paid for it, and I have designed Copilot-only programs for rooms where most seats had never been assigned. Count them before you book the trainer.
The real lesson
The tool choice is not really about the tool. It is about designing for adoption rather than demonstration. A workshop where everyone leaves saying "wow, ChatGPT is amazing" and nobody uses it next week has failed. A workshop where everyone leaves with Copilot wired into the email they were going to write anyway has done its job, even if nobody was amazed.
