Your staff have already migrated — to AI, on personal ChatGPT accounts you can't see. Now you're expected to move them onto something sanctioned, and every instinct says mandate it. Resist it. Block the network and usage moves to phones. The only migration that sticks is the one people choose.
The real question
People revolt when you take away something that works and hand them something worse. The classic migration play — a network block plus a "please use the approved tool" email — fails twice. The block fails because personal devices exist: usage doesn't stop, it goes dark. The email fails because it asks people to trade a strong model and months of honed prompts for a weaker tool with neither. The revolt isn't loud: nobody argues in the town hall. They nod, and keep the personal account on their phone.
How to do it
Start by refusing to take the models away. Connect the subscription your organisation already pays for, so the models staff chose are the models they meet inside the governed workspace. The pitch changes from "give up the tool you like" to "same models, with governance you never have to think about."
Then widen choice instead of narrowing it. Run models from every provider your organisation approves — managed subscriptions, API keys, cloud-hosted options — and let people switch mid-conversation without losing the thread. A personal account ties everyone to one vendor; the sanctioned workspace should be the first place staff have genuine choice.
Finally, make the shared option actively better. Save your team's best prompts once and publish them as shared equipment — searchable, reusable, there for every new starter on day one. A personal account holds one person's tricks; a shared workspace compounds the whole team's.
What this looks like
In Pebble, this looks like: an administrator connects the organisation's existing ChatGPT subscription and enables its models through the same catalog controls as every other provider. Staff open the workspace and find the models they'd been paying for personally — alongside others from API keys or AWS Bedrock, switchable mid-conversation. The team's best prompt writer saves her brief-development prompt to the Prompt Library and publishes it to the team. The migration email then changes from "personal AI accounts must stop" to "the models you use are already here, with the team's prompt library loaded" — and that's the email people act on.
Why this holds up in a regulated business
- Admins add centrally managed ChatGPT and Claude subscription connections at organisation level; models are enabled through the same catalog controls as every other provider.
- Model choice spans API keys, AWS Bedrock, and managed subscriptions; users switch models mid-conversation without losing the thread.
- Admins manage which models are enabled and provisioned per organisation and per workspace.
- Saved prompts are searchable in the capabilities catalog and can be shared or published to teams.
- Deployed across US, Europe and Australia.
One honest limit: personal ChatGPT accounts and their chat history don't migrate — the connection is organisation-level, so people start fresh inside the workspace. That's exactly why the prompt library matters: day one starts with something the personal account never had.
Where to start
Connect the subscription before any announcement. Have each team publish its five best prompts. Only then send the email, leading with what staff gain. Migration by preference, not prohibition. Make the sanctioned option the one they'd choose anyway.
Pebble Powered AI.
