
A cloud migration strategy in six steps: from inventory to the old server off
A cloud migration strategy rarely fails on the technology. It overruns on an application nobody had listed, a licence that counts differently in the cloud, or an old server that keeps running for years because nobody dares to switch it off. This plan puts the six steps in the order in which they cost least.
Start with an inventory of applications and their dependencies, not of servers. A server can only be moved once you know what relies on it.
Take a separate decision per application: keep, move as is, adapt, replace with a cloud service or retire. One decision for everything is usually the wrong one.
A cloud migration is only finished when the old server is off, the licences are cancelled and the disks are wiped. Plan that as a step, not as an afterthought.
Step 1: the inventory before you migrate
The first mistake in almost every cloud migration is starting with the servers. A server is a box; what counts is what runs on it and what is attached to that. Build the list per application, and answer these questions for each line:
- Who uses it and how critical is it? How long the organisation can do without it, and at which point in the month or year not at all.
- What does it depend on? A database on another server, a shared folder, an integration with the ERP system, a scanner that saves to a folder, a licence server.
- Who is the vendor? Do they support their software in the cloud, and do they offer their own cloud service that replaces the package?
- Which contracts come with it? Hardware maintenance, licences, the line into the building. Their end dates help decide when an application moves.
The dependencies are the most valuable part of the list. You do not find them only by looking at what is installed, but by measuring for a few weeks which network traffic runs between systems, and by asking each department what they open first thing in the morning.
Step 2: a cloud migration strategy per application
Not everything should go to the cloud in the same way, and not everything should go at all. Per application there are roughly five outcomes:
- Retire. What nobody uses any more does not move. Every inventory finds a few, and they are the cheapest migration there is.
- Replace with a cloud service. Email, files and many standard packages have a SaaS version. You then move data rather than a server, and the management of that layer disappears.
- Move as is. The server goes to the cloud as a virtual machine, unchanged. Quick, but you take the management and the quirks with you.
- Adapt. Part of it moves onto a cloud service, for example the database onto a managed database platform. More work up front, less management afterwards.
- Keep. A system that talks to machines in the building, or whose vendor does not support the cloud. That is a decision, not a backlog; write down the reason.
What decides the sum per application is not just the monthly cloud price but also what you cancel in return: hardware maintenance, power, a replacement cycle you skip. How to weigh your own servers against the cloud is covered in own servers or cloud.
Step 3: the foundation first, then the applications
Before a single application moves, the environment it lands in has to be ready. This is the part most easily skipped, because it produces nothing visible.
- Identity. One source for accounts, multi-factor authentication on everything, and separate admin accounts. This comes first, because every later step signs in somewhere.
- Network. How the office and the cloud talk to each other while both are needed, and which services may be reachable from the internet.
- Backup and recovery. A virtual machine in the cloud is not automatically backed up somewhere else. Sort it out before anything you cannot afford to lose is on it.
- Cost monitoring and naming. A budget alert, and a fixed way of naming resources and linking them to an owner. Working out afterwards who owns a running server costs more than agreeing it up front.
Step 4: a pilot with an application that is allowed to stumble
For the first migration, pick something that is genuinely used but where a day of disruption is not a disaster: an internal tool, a test environment, a departmental application. A pilot with something nobody opens proves nothing; a pilot with the ERP system is no longer a pilot.
Measure three things during the pilot that you will need for the rest: how long the move actually takes (including waiting for others), how the application performs when the users are in the office and the server no longer is, and what it costs per month compared with your estimate. Those three figures make the planning of the next waves honest.
Step 5: in waves, with a way back
Then move in waves: applications that depend on each other go together, applications that stand alone go separately. The inventory from step 1 sets the groups; the business calendar sets the moment. No finance system in year-end week, no tills in the busiest month.
- A fixed sequence per wave: copy while everyone keeps working, freeze the source, sync the difference, test with the users, switch over.
- A way back with a criterion. What has to go wrong before you roll back, and who decides. The old environment stays until that decision is no longer needed.
- Treat data separately. Where files and databases are concerned, what counts is proving that nothing is lost. How to do that is in our article on data migration without loss.
Step 6: actually switching the old server off
This is the step most often left undone, and while it is, you pay twice and manage two environments. Make it a step of its own, with a date:
- A read-only period. The old server stays available for an agreed time, with no more work done on it. After that it goes off, even if nobody asks.
- Cancel contracts and licences. Hardware maintenance, server licences you are not taking with you, the backup of the old environment. Check which licences you may reuse in the cloud; the vendor’s terms decide that, not the migration.
- Wipe or destroy the disks. A server that leaves the building leaves without data, with evidence per disk.
- Update the documentation. Network diagram, inventory and recovery plan describe the new state, not the old one.
Where our work stops
What we do. We build the inventory, set out the options per application, set up the foundation (identity, network, backup, monitoring), run the pilot and the waves, and switch the old server off, including wiping the disks. After that we manage the environment if you want us to.
What we do not do. We are not a cloud provider: the platform, its data centres and its availability belong to Microsoft or another provider, and your contract with them covers that. We do not convert the database of a third-party package to a new version; the software vendor or its implementation partner does that. And whether you may reuse a licence in the cloud is decided by the vendor’s terms; we look into it, but we do not give a legal opinion.
Questions we get about this
The five that come up most often.
What are the steps of a cloud migration strategy?
An inventory of applications and their dependencies, a decision per application, setting up the foundation (identity, network, backup, cost monitoring), a pilot migration, the cutover in waves with a way back, and switching off the old environment. The last step is the one most often forgotten, and it costs the most when it is left undone.
What are the common cloud migration approaches?
Broadly five: retire what nobody uses, replace it with a cloud service, move it as is, adapt part of it to a cloud platform, or keep it where it is. Most organisations end up using several of them, one per application, rather than a single approach for everything.
How long does a cloud migration take?
It depends on the number of applications and above all on how they are connected. An environment with email, files and a few standalone packages moves faster than one where everything hangs off the ERP system. The pilot gives the first honest figure; a plan without a pilot is an estimate.
What drives the cost of a cloud migration?
The number of applications and their dependencies, how much has to be adapted rather than simply moved, the volume of data and how much downtime the cutover may take. After that, compute, storage, outbound data transfer and licences set the monthly cost, against what you cancel in hardware, maintenance and power.
Can we take our existing licences to the cloud?
Sometimes, under conditions that differ per vendor and per licence. That belongs in the inventory before the business case is made, not after. The vendor’s terms are decisive; a migration partner can look into them, but does not give a legal opinion.
Further reading
What this plan connects to.
Shall we start your migration with the inventory?
Send us a list of your servers and the applications running on them. We will add a proposal per application and tell you which dependencies still need to be worked out.
Practical IT knowledge in your inbox
New guides on management, security and the workplace, written by the people doing the work. No sales talk, and you can unsubscribe in one click.