Data Migration and Cloud Migration Projects: Planning, Risks and Cutover
Moving applications and data to the cloud, or from an old system to a new one, is among the most common IT projects and among the most underestimated. The technical work is real, but projects usually go wrong on discovery (nobody knew what was running), data quality (the data was messier than anyone admitted) and cutover (the weekend was not long enough). This lesson gives project managers a practical structure for data migration and cloud migration projects, with the main strategies, phases and risks.
Two Related but Different Projects
| Aspect | Data migration | Cloud migration |
|---|---|---|
| What moves | Records and files from a source system to a target system | Applications, servers, databases and workloads to a cloud platform |
| Typical trigger | System replacement, merger, consolidation | Data-centre exit, cost, scalability, resilience |
| Main challenge | Data quality, mapping and reconciliation | Dependencies, performance, security and cost control |
| Typical success test | Data is complete, accurate and usable in the new system | Workloads run reliably, securely and at expected cost |
They often overlap. Moving a database to the cloud is also a data migration, and a new SaaS system usually needs data loaded from the old one.
Cloud Migration Strategies
Cloud providers and industry bodies describe a set of options for each application, often called the “R” strategies. The exact list varies by source, but a common version is:
- Rehost (lift and shift): move the application as it is to cloud infrastructure. Fastest, but gains fewer cloud benefits.
- Replatform: make limited changes, such as moving to a managed database, to gain some cloud benefits without redesign.
- Refactor (re-architect): redesign for cloud-native capabilities. Highest effort and potential benefit.
- Repurchase: replace with a SaaS product.
- Retire: switch off applications that are no longer needed.
- Retain: keep on-premises for now, because of cost, compliance or readiness.
- Relocate: move virtualised infrastructure to a cloud equivalent with minimal change, where such a service exists.
The project manager does not choose the strategy alone, but must make sure a decision is made per application, with reasons, early in the project. A common finding in discovery is that a meaningful share of applications can be retired.
Migration Phases
- Discovery and assessment. Inventory applications, data stores, dependencies, owners, volumes and licences. Include hidden dependencies such as batch jobs, integrations and shared services.
- Planning and business case. Choose strategies, group applications into migration waves, estimate costs, define success criteria and confirm the security and compliance approach.
- Foundation build. For cloud: set up the landing zone, with accounts, network, identity, security controls and monitoring. For data: agree the target data model and tooling.
- Pilot. Migrate a small, low-risk application or data subset to test the process and tools and to learn.
- Migration waves. Move groups of applications or data in planned batches, with repeated testing.
- Cutover. Switch live use to the new environment (see Go-Live, Deployment and Cutover Planning).
- Optimise and decommission. Tune performance and cost, shut down old systems and close contracts.
Data Migration Steps in Detail
- Profile the source data. Measure completeness, duplicates, invalid values and formats before promising anything.
- Define scope. Decide what to migrate: all history, or only active records plus an archive. Less data means less risk.
- Map and transform. Document how each source field maps to the target, including transformation rules and defaults.
- Cleanse. Fix or flag poor data, with business owners making the decisions about what is correct.
- Build and test the load process (often an extract-transform-load pipeline) with repeatable scripts.
- Rehearse. Run full trial migrations to measure duration and error rates, usually at least two or three.
- Reconcile. Compare counts, totals and sample records between source and target, and get business sign-off.
- Plan rollback. Decide in advance how to revert if the cutover fails.
Data quality is a business responsibility, not just a technical one. Appoint business data owners and give them time to review exceptions, or the schedule will slip.
Typical Risks
| Risk | Response |
|---|---|
| Hidden dependencies discovered late | Automated discovery tools, interviews, a dependency map reviewed by owners |
| Poor data quality | Early profiling, cleansing owners, agreed quality thresholds |
| Cutover window too short | Rehearse, measure timing, consider phased or incremental migration |
| Performance worse than expected | Load testing, right-sizing, network analysis |
| Security or compliance gaps | Security design review, data residency checks (see Security, Compliance and Data Privacy in IT Projects) |
| Costs higher than forecast | Cost modelling, tagging, budgets and alerts, regular cost reviews |
| Skills gap in the team | Training, partner support, pilot to build experience |
| Business disruption | Schedule around peak periods, communicate, provide hypercare |
Record these in the risk register with owners (see Project Risk Management).
Cost Management in the Cloud
Cloud costs are usage-based, so they continue after the project ends and can drift upward. Plan for the run cost as well as the migration cost: include it in the business case, tag resources by project and owner, set budgets and alerts, and review usage regularly. The practice of managing cloud spend collaboratively between finance, engineering and the business is often called FinOps. A migration that finishes on budget but doubles the monthly bill has not necessarily succeeded.
Testing
- Functional testing that the application behaves as before.
- Integration testing of connected systems and interfaces.
- Performance and failover testing in the new environment.
- Data validation including reconciliation reports.
- User acceptance testing with real business scenarios (see Quality Management and Testing in IT Projects).
Worked Example: Moving a Finance Application
A company plans to move its finance application and database to the cloud over a holiday weekend. Discovery reveals three undocumented batch jobs and a reporting tool that reads directly from the database. The project manager adds these to the dependency map and schedules a rehearsal migration the previous month. The rehearsal takes 31 hours against a 24-hour window. The team tunes the data load, uses an incremental approach for historical data, and re-rehearses, bringing the window to 19 hours. At cutover, reconciliation reports are signed off by the finance controller, and the old system stays read-only for two weeks as a fallback.
Common Mistakes
- Starting migration before discovery is complete.
- Assuming the data is clean because the old system “works”.
- Skipping rehearsals and discovering the real duration during cutover.
- Ignoring run costs and security design until late.
- Decommissioning the old system immediately, leaving no way back.
- Discovery and dependency mapping determine whether the migration succeeds.
- Choose a strategy per application: rehost, replatform, refactor, repurchase, retire or retain.
- Profile, cleanse, map, rehearse and reconcile data, with business owners involved.
- Plan rollback, security and ongoing cloud costs from the start.