Cloud · 2026-06-12 · 11 min
The 2026 cloud migration checklist: what actually delays go-live
Most migrations stall on identity, data gravity, and cutover rehearsal — not on picking AWS versus Azure. Here is a field checklist we use with product teams.
Cloud migrations fail quietly. The architecture diagram looks clean, the proof of concept works, and then go-live slips because nobody owned DNS, SSO, or the one database that still talks to an on-prem license server.
We run a 14-point checklist before we promise a date. The items below are the ones that repeatedly move timelines by weeks, not hours.
What the data says
Flexera’s 2025 State of the Cloud survey still shows more than 30% of spend wasted on idle or oversized resources after lift-and-shift. Gartner has long warned that organizations without a FinOps owner see cloud bills climb 20–40% in year one.
Treat cost as a product metric from week one. A migration that “works” and then doubles your invoice is not done.
- Map every production dependency, including batch jobs and vendor VPNs.
- Decide identity first: SSO, service accounts, and secret rotation.
- Rehearse cutover with a freeze window and a documented rollback.
- Tag resources on day one so FinOps is not archaeology later.
A practical sequence
Start with a thin production slice — one customer-facing service — rather than a big-bang weekend. Instrument it. Then move data stores that the slice actually needs. Everything else can wait.
Rule of thumb: if you cannot name the person who will be paged, do not move that system yet.
When to stay hybrid
Latency-sensitive factory systems, licensed mainframe connectors, and data residency constraints are valid reasons to stay hybrid. The mistake is pretending the hybrid plane does not need the same observability as the cloud plane.