How to plan a cloud migration for an agency

Cloud migrations fail on planning far more often than on technology. The systems that matter are usually the ones nobody documented, and the constraints that decide the architecture are usually regulatory rather than technical.

This sequence assumes an agency with a mixed environment: some on-premises servers, several vendor-hosted platforms, and a productivity suite already in the cloud.

Steps

1. 1Inventory what is actually running

   Catalog servers, applications, databases, integrations, and the business process each supports. Include the systems only one person knows about. Record data classification for each, because that drives everything downstream.

2. 2Map dependencies

   Find what talks to what, including scheduled jobs, file transfers, and database links. Undocumented dependencies are the leading cause of failed cutovers.

3. 3Decide a disposition for each workload

   For each system choose rehost, replatform, refactor, replace with a product, retain, or retire. Record the reason. Retiring unused systems is often the highest-return decision in the whole program.

4. 4Design the landing zone first

   Establish account or subscription structure, identity federation, network topology, centralized logging, and preventive guardrails before any workload arrives. Define it as infrastructure as code so it is reviewable and repeatable.

5. 5Document the shared responsibility split

   For each service, write down which controls belong to the provider, which to the agency, and which to any managed service partner. This document answers most audit questions later.

6. 6Set cost controls before the first bill

   Tagging standards, budgets and alerts by department, non-production scheduling, and a monthly review with finance. Consumption pricing and annual appropriations do not mix without these.

7. 7Sequence the migration in waves

   Start with a low-risk workload to prove the process. Group later waves by dependency and by business calendar, avoiding billing cycles, election periods, and fiscal year end.

8. 8Rehearse each cutover

   For every wave, write the cutover runbook, the validation checks, and the rollback criteria. Rehearse with real data volumes. Decide in advance who can call a rollback.

9. 9Validate, then decommission deliberately

   Confirm functionality, performance, backups, and monitoring in the new environment before retiring anything. Record data sanitization and asset disposition for audit.

10. 10Hand over operations

    Deliver runbooks, monitoring dashboards, and training so agency staff operate the environment. A migration that leaves the agency dependent on the migrator is not finished.

What you end up with

An inventory and disposition record, a documented landing zone defined in code, migrated workloads with tested cutovers, cost controls that fit an annual budget, and staff able to run what was built.