Preparing a cloud migration: the questions before anything moves
For: IT leads in smaller companies
Updated: 2026-09
Cloud migrations rarely fail on technology; they fail on preparation. Unknown dependencies, unresolved data classification, a cost model that only becomes visible after the move, and missing acceptance criteria against which success could be measured at all.
A move to the cloud is usually planned as a technical undertaking and then founders on non-technical questions: which systems actually depend on one another, which data may go where, and how we will know we are finished.
This list covers preparation rather than execution. It is written for mid-sized circumstances where no dedicated cloud team exists and the migration runs alongside daily business.
The checklist
- 01
Record the estate and its dependencies
List not just the systems but the connections between them. Planned migrations regularly stall on an interface nobody remembered.
Done when: For each system, what it depends on and what depends on it is named — confirmed by the relevant business unit.
- 02
Classify the data
Establish which data is particularly sensitive and what requirements apply to its location. This question decides the region and sometimes the provider.
Done when: Every data set carries a protection class and each class has a permitted region.
- 03
Choose a migration strategy per system
Move unchanged, adapt or rebuild — the decision is per system, not blanket. Moving unchanged is the fastest and rarely the cheapest to operate.
Done when: Each system has its strategy recorded with reasoning.
- 04
Model the costs before moving
Cloud costs arise differently from owned infrastructure: egress traffic, storage classes and deployment operations all count. An estimate beforehand prevents the most common unpleasant surprise.
Done when: An estimate of monthly running costs exists and explicitly includes data transfer.
- 05
Define the rollback route
For each system, establish how the pre-migration state is restored if something goes fundamentally wrong. Without a rollback, every incident becomes a crisis.
Done when: The rollback route is described and the window for using it is named.
- 06
Agree acceptance criteria in advance
What will measure success: response times, availability, completeness of data. Without criteria agreed beforehand, acceptance turns into a negotiation.
Done when: The criteria are agreed in writing and measured on the old system before the migration.
- 07
Rethink access and permissions
Grown permission structures should not be carried over unchanged. The migration is the only realistic moment when they can be cleaned up without extra resistance.
Done when: The permission model for the target environment is described and deliberately differs from the legacy one.
- 08
Settle operations and ownership
Who monitors, who responds to incidents, who owns cost. Responsibilities shift in the cloud, and the gaps surface at the first incident.
Done when: Named people are assigned to monitoring, incident response and cost ownership.
Common mistakes
- —Costs are estimated from compute while data transfer and storage classes make up the larger share.
- —Every system is moved unchanged because that is fastest. Running costs rise permanently with no offsetting benefit.
- —Permission structures are carried over unchanged, including accounts of people who left years ago.
- —Acceptance criteria are formulated after the migration, once it is already clear what works and what does not.
What this checklist does not cover
- —This list covers preparation. Execution itself requires a plan per system.
- —Provider selection is not included; it depends on your estate, your skills and regulatory requirements.
- —For particularly sensitive data or regulated sectors, supervisory requirements on outsourcing come on top.
- —Cost statements are estimates. Reliable figures only emerge in operation, so a recalculation after a few months should be planned in.
Parent service: Digital Transformation
Matching offers
Cloud migration assessment
A cloud migration assessment provides the basis for deciding on the move to the cloud: an honest evaluation of benefit, risk, cost and data protection — with a clear recommendation and a realistic migration path. The result is an informed decision rather than a leap into the unknown.
Digitalisation roadmap for SMEs
A digitalisation roadmap gives an SME a pragmatic plan: an honest current-state analysis, a prioritisation by actual impact, and an executable plan. The result is digitalisation focused on what genuinely moves the business forward — rather than expensive projects that miss the need.
FAQ
Is the cloud cheaper than owned infrastructure?
Not necessarily. It is more flexible and shifts investment into running costs. Whether it is cheaper depends on your utilisation profile and data transfer, and should be calculated beforehand.
Should everything migrate at once?
Rarely. Migrating one system with few dependencies first produces experience and makes the estimates for the following systems far more reliable.
What about data that must stay in Switzerland?
The major providers now operate Swiss regions. What matters is that the requirement is settled before provider selection rather than clarified afterwards.
More checklists
Replacing a legacy system: the groundwork that decides the outcome
Legacy replacements usually fail on three things: undocumented business logic nobody understands any more, data quality that only surfaces during migration, and a missing switch-off date, which leaves both systems running in parallel indefinitely.
Choosing a software partner: what to settle before you commission
Choosing a software partner is decided less by references than by four contractual points: who owns the result, who holds access to the systems, what a handover looks like, and what happens when the engagement ends.
EU AI Act conformity: a checklist for the first pass
A workable first pass through the EU AI Act has six parts: know which AI runs in the business, determine your role for each system, classify under Article 6, check transparency duties separately, ensure AI literacy, and file everything so it can be found under scrutiny.