Cloud migration services have matured enormously in the last five years, but the failure mode has not changed: teams confuse lift-and-shift with modernisation, underestimate the network and identity work, and run out of budget halfway through. Here is a 12-week playbook that avoids that trap.
Weeks 1–3: Discovery and the right migration strategy
The first three weeks are not about moving anything. They are about knowing what you have. Asset inventory, dependency mapping, cost modelling, and picking the right strategy per workload from the classic seven Rs: rehost, replatform, repurchase, refactor, retire, retain, relocate.
Moving a database as-is to an IaaS VM is the most common way cloud migrations blow up the bill. Pick a managed service or plan a refactor — do not carry the anti-pattern into the cloud.
Weeks 3–5: Landing zones and foundations
- Multi-account/multi-subscription strategy — one account is a mistake you only make once.
- Identity and access baseline — SSO, roles, break-glass, tagging.
- Network architecture — transit, VPCs, private connectivity, hybrid routes.
- Security baseline — logging, detective controls, secrets management.
- FinOps baseline — tagging, budgets, anomaly alerts, dashboards.
Weeks 5–9: First migration waves
Start with a non-customer-facing, medium-complexity workload. Learn the cutover pattern on something forgiving. Then move to the harder workloads with the playbook tightened.
- Wave 1: internal tools, non-critical web apps, batch jobs.
- Wave 2: customer-facing stateless services.
- Wave 3: stateful services — databases, queues, caches.
- Wave 4: legacy monoliths and long-tail workloads.
Weeks 9–12: Optimisation, handover, and exit
- Right-sizing and reserved-instance/savings-plan commitments.
- Autoscaling, spot, and serverless where the workload fits.
- Runbooks, on-call handover, and DR testing.
- Decommission of old infrastructure — the hardest step to enforce politically.
The risks no migration brochure mentions
- Cost shock in month three because someone left dev at production size.
- Licensing transitions for Windows, Oracle, and SQL Server that were not modelled.
- Data egress bills when cross-cloud or hybrid topologies were not planned.
- Change management — your engineers need new muscle, fast.