Migrations do not fail on technology. They fail on the first exception.

We have now accompanied enough migrations to know where they tip over. It is rarely the technology. It is almost always a single exception that has become a precedent four weeks later.

The sequence

The company moves, everyone switches, and one person keeps their old tool. The reasons always sound sensible: travels a lot, the assistant maintains the calendar, an important project is running, right after that.

Four weeks later this is no longer an exception but a proof. The new tool evidently is not enough, otherwise that person would be using it too. Whoever wants an exception next no longer has to justify it — they point upwards.

From here the project is no longer steerable. IT can only administer which systems run in parallel, and the date for switching off the old one slides indefinitely.

What works

Management migrates in the first wave. Not symbolically, but with the same support as everyone else — and the same duty to report problems rather than sit them out. Start at the top and you need less persuasion further down.

Every exception gets a date. Not "for now" but "until 31 October", written down, with a person who follows up. An exception with no end date is a decision against the project, phrased politely.

Anyone with an exception is not consulted in the evaluation. Otherwise somebody assesses a product they do not use — and their assessment will outweigh that of the department which has been working with it for eight weeks.

The five most frequent tasks work on day one. Before cut-over, do not ask the department head; ask the people who perform those tasks. Everything else can wait. Those five cannot.

What happens if you ignore this

Shadow IT appears, within two weeks. If after cut-over everyone suddenly talks over a private messenger, the workforce has not failed. The migration has failed, and the workforce solved the problem nobody solved for them.

That is why Shadow IT is the most honest feedback a company can get: it points exactly at where the new tool does not do a job that used to get done. A ban only moves the usage somewhere you can no longer see it — turning a usability problem into a data protection problem.

What helps instead: set up an official path for external partners before somebody finds an unofficial one. And make reporting cheaper than working around — a channel where "this doesn't work any more" can be said without justification saves more than any policy.

The honest abort

The most uncomfortable point last. If management does not want to go along, that is a legitimate decision. In that case end the project, rather than running it for twelve months and afterwards blaming the technology.

We have seen both. The honest abort was cheaper every time — financially, and above all for the next attempt. A project that was ended cleanly can be restarted in two years. One that failed slowly over twelve months cannot.

And there is a middle outcome that is more often the right one than people think: stop after three of five layers. Video conferencing and file storage European, document editing not for now. That is not half a result. It is more than any debate of principle ever achieves that ends at zero layers.