Retire a Legacy Platform Without Making Cutover the Experiment
“We’ll know once we switch over” is not a migration strategy.
When a platform supports live customers, the cutover is the wrong time to discover whether the replacement behaves as expected. The team needs evidence before traffic moves, and a clear way to stop if the evidence is not good enough.
Across two platform retirements, the consistent principle was to define the decision before the pressure of the decision arrived.
Start with observable comparison
For a decision engine replacement, the legacy system needed to remain active until the new flow met criteria we had defined in advance. We ran shadow comparisons so the team could inspect how the replacement behaved alongside the existing platform before committing to cutover.
For a separate migration of more than three million accounts, we used phases, parallel processing, transaction-level accuracy monitoring, and rollback criteria for each phase. We required the accuracy gate before progressing to the next cohort. When an issue appeared mid-migration, the pre-agreed criteria made it possible to pause and fix rather than debate the policy under pressure.
The specific thresholds depend on the product. What matters is that the threshold, evidence, and decision owner are agreed before the rollout starts.
Make rollback a real option
A rollback plan that nobody can execute is not a safety net. Before migrating, teams should know:
- What signal triggers a pause or rollback?
- Who has authority to make that call?
- How quickly can traffic or accounts return to the prior path?
- What happens to work already in progress?
- How will teams communicate the change to support and operations?
Rollback criteria should be measurable and practical. They should protect the customer and the business, not simply reassure a steering committee.
Treat the sunset as part of the product
The replacement can be technically live while the operating model is not. Tiered support, access governance, alerting, escalation, and ownership need to be defined before the legacy system disappears.
That is especially important when the new platform has more than one channel or many downstream consumers. A cutover can be successful for the primary user and still break a reporting, servicing, or fulfilment dependency that nobody considered part of the launch.
Map the dependencies, assign owners, and confirm what each system can do today. A requirement stating that a downstream system is ready is not the same as evidence that it can handle the new product in production.
Learn without gambling on the customer experience
The two retirements completed with zero cutover incidents. That result did not come from assuming the migration would be safe. It came from creating ways to learn before making the change irreversible: parallel comparison, phased movement, pre-defined gates, and named operational owners.
A migration review can begin with three questions:
- What must be true before we move the first cohort?
- What evidence will tell us whether the new system is behaving correctly?
- What is the safest action if that evidence is missing or negative?
If the team cannot answer those questions, it is not ready to make cutover the test.