Leaving Microsoft 365: the order most projects get wrong

We have now accompanied enough of these projects to know where they fail. It is never the technology. It is always the order.

The usual mistake looks like this: management decides to leave, IT starts with the largest item — the mailboxes — and in week three half the workforce is at the door because a calendar invitation to an external partner did not arrive properly. After that the project is politically dead, no matter how well the migration went technically.

The order that works

Stage 1: video meetings. No data migration, no risk. A different link in the calendar. If OpenTalk or Nextcloud Talk does not work for you, you find out in the first meeting and switch back. The effort is an afternoon; the benefit is visible proof that a European tool is not automatically worse. You need that proof for everything that follows.

Stage 2: file storage, but only the cold half. Every company has shares nobody co-edits: archives, closed projects, contract folders, image libraries. Those move without friction. The hot half — the folders where three people work on the same spreadsheet at once — you leave alone for now.

Stage 3: chat. Politically trickier than video, because history and habit hang on it. Technically solved with Matrix and Element. Expect old history not to come across, and say so beforehand.

Stage 4: email and calendar. Only now. IMAP sync is a solved problem, the MX record moves last, and you begin with a department that is cooperative — not with sales.

Stage 5: document editing. The hardest part, which is why it goes last. This is where the macros, the templates and the external correspondence live.

What to test first

Take twenty real documents out of your estate. Not twenty test files — twenty that are actually used. The proposal with the company layout, the spreadsheet with the nested formulas, the form that goes to a regulator.

Open them in LibreOffice, in Collabora and in ONLYOFFICE. You will find that ONLYOFFICE stays closest to the original on OOXML and that LibreOffice wins on anything meant to run locally without a server. That result is your tooling decision, and it takes half a day.

What gets worse, and why to say so

Co-editing in Collabora is slower than in Microsoft 365. Search across the whole file estate is weaker. The phone apps are rougher. And there is no department in the building that has never needed an Excel macro.

Hide that and you lose credibility at the first incident. Say it up front and you get patience — and patience is the scarcest resource in a project like this.

The timeline

For a company with 100 to 300 seats, the realistic frame is twelve to eighteen months for all five stages while the day job continues. Stages 1 and 2 are doable in eight weeks.

And if you stop after stage 3 because document editing does not pay: also fine. Three of five layers in European hands beats a matter-of-principle debate that ends at zero.