Key takeaways
- Decide per workload, not per company: each system has its own best path.
- Rehost is fastest, replatform captures cheap cloud benefits and refactor pays off only where redesign creates value.
- Retiring or replacing systems is often better than moving them.
- Plan cost control and data location before migrating, not after.
The main options
Migration approaches are commonly described as a set of strategies. The most relevant for most organisations:
- Rehost (lift and shift): move the system as it is onto cloud infrastructure.
- Replatform: move with small changes that use managed services, for example a managed database.
- Refactor: redesign the application to use cloud-native capabilities.
- Replace: switch to a software-as-a-service product.
- Retire or retain: switch the system off, or keep it where it is for now.
Rehost: fast, but limited gains
Rehosting gets you out of a data centre quickly with the least change, which suits tight deadlines. The trade-off is that you take your inefficiencies with you, and costs can be higher than expected if sizing is not revisited.
Replatform: the practical middle
Small changes, such as moving to a managed database or container platform, can cut operations work and improve resilience without redesigning the application. For many business systems this is the best balance of effort and benefit.
Refactor: only where it pays
Redesigning an application is the most expensive route and takes the longest. Reserve it for systems where scalability, speed of change or cost at scale justify the investment.
How to choose
For each workload, ask:
- How important is it to the business, and how long will it live?
- What does it depend on, and what depends on it?
- What are the data protection and location requirements?
- Is there a good software-as-a-service replacement?
- What is the effort and risk of each route, and what is the benefit?
Common pitfalls
Watch for these:
- No cost controls, so spend grows unnoticed.
- Moving first and thinking about security and identity afterwards.
- Underestimating data transfer and dependency complexity.
- Skipping a landing zone that sets baselines for every workload.