Cloud migration gets sold as a decision about technology. In a company under a hundred people it is mostly a decision about money and risk, and the technology part is the least interesting bit. Here is what we tell clients before they commit, including the parts that argue against doing it.
First, work out what you are comparing
The honest comparison is not “server cost” against “cloud bill.” It is everything you currently spend, including the parts that do not appear on an invoice.
Hardware amortised over its real life, not the accounting life. Power and cooling. The maintenance window somebody works on a Sunday. The warranty renewal. The replacement you are going to need in eighteen months. Put those in one column, then get a costed cloud estimate for the other. Sometimes the answer is that you should stay put for another two years, and that answer is legitimate.
Not everything should move
The instinct is to lift the whole room. Resist it. Every migration we have run has a list of things we deliberately left behind, and that list is usually the most valuable page in the plan.
Software that is out of support, a licence tied to a physical machine, a device that talks to a specific bit of hardware in the building. These get replaced, retired, or left where they are on purpose.
The bill can drift, and you should model it
Cloud pricing punishes carelessness. Storage grows quietly, a test environment gets left running, egress charges appear when somebody starts moving large files around.
Ask for a model with three scenarios rather than a single number: what you expect, what a busy quarter looks like, and what happens if data volume doubles. If a provider will not produce that, they are selling you a migration and not a plan.
Rehearse before you commit
A dry run on a Saturday morning with the old system still live is the cheapest insurance in the whole project. It surfaces the things nobody wrote down. It has never once been a waste of a morning.
The rollback procedure matters just as much. If your provider cannot describe, in writing, how you get back to Friday’s state by Monday, do not schedule the cutover.
Plan the boring human parts
Mapped drives, printers, the shortcut on somebody’s desktop that points at a path which will no longer exist. Staff judge a migration entirely by whether their morning routine survived it.
Getting this right is unglamorous and it is most of the perceived success. People do not notice that the infrastructure is modern. They notice that the folder is where it was.
Timing
Six to eight weeks is normal for a company of twenty to forty staff, and shorter estimates usually mean the inventory phase has been skipped. Cut over on a Friday evening so there are two days of margin.
Avoid the week before a deadline in your own business, which sounds obvious and is ignored constantly. Do not migrate an accounting practice in March.
The question worth asking last
What breaks if this goes badly, and who is affected? If the answer is that thirty people cannot work on Monday, then the plan needs a rehearsal, a rollback and somebody’s phone number. If the answer is that a reporting job runs late, you can be more relaxed about the whole thing.