Most companies we onboard are not starting from zero. They have a seat-based contract, a shadow deployment nobody sanctioned, and no way to show a regulator what happened. Here is how the move works.
You have a good model behind one interface, and a per-seat bill that grows with headcount rather than usage. What you do not have is a named agent per person, tier-enforced routing, or an export your DPO can read. We keep the model — it is one of ours too — and put a gateway and a ledger in front of it.
Deep in Office, shallow everywhere else, and routed wherever Microsoft decides. Teams stays as your chat surface. What changes is that your CRM, ad accounts and Nordic finance tools become readable, and Tier 3 work stops leaving your jurisdiction.
Usually the hardest conversation, because the tool works. The problem is the paperwork: sub-processors you cannot enumerate, logs you cannot export, and a DPA written for a different continent. We run alongside it until the evidence gap is obvious.
You do not cut over. You run in parallel, compare, then let the old contract lapse on its own terms.
We map what AI is actually in use, including the personal subscriptions. Then we stand up your workspace and put agents next to the incumbent for the same work. Nobody loses a tool. Two weeks is usually enough to see the difference in output and in evidence.
Counter-intuitive, but correct. The regulated workflows are where the incumbent is weakest and where tier routing pays for itself immediately. Once Tier 3 is provably inside European jurisdiction, the rest is a preference question rather than a risk one.
Personal subscriptions get cancelled because the sanctioned tool is better, not because IT sent an email. When the incumbent renewal comes up, you have a full ledger of what the alternative did instead of a hunch.
| Thing | Transfers? | Detail |
|---|---|---|
| Your files | Nothing to move | Agents read in place from Drive, SharePoint, OneDrive and the rest. No copy into a separate store unless you add it to the vault deliberately. |
| Chat surface | Yes | Slack or Teams, whichever you already use. Your admin approves one app. |
| Custom GPTs / Copilot agents | Rebuilt, not imported | There is no honest import path. We rebuild the ones that earn it as skills, which are versioned and auditable. Most turn out to be two or three prompts. |
| Prompt libraries | Partly | Paste them in; they become skills. Expect to keep about half — the rest were working around a limitation you no longer have. |
| Conversation history | No | Deliberately. Importing a history you cannot attest to would poison the ledger on day one. Old history stays where it is until you delete it. |
| Your incumbent contract | Runs its course | We do not ask you to break it. Parallel running is the point. |
| SSO configuration | Yes | SAML against Okta or Entra ID. Usually the fastest part of the whole exercise. |
Thirty-day notice, standard-format exports on day one, exit clause in the DPA. We would rather you read it before signing than discover it later.