A genuinely zero-downtime ERP cutover is rare, because the financial ledger has to stop moving long enough to be reconciled. What is achievable is a freeze window of a few hours instead of a long weekend — and that window is bought in the months before go-live, not on the night itself.
4-Phase Near-Zero Downtime ERP Cutover Architecture
Initial Historical Load & Master Freeze
Historical GL years & inactive data migrated. Master records (COA, Customers, Items) frozen.
Open Transaction Sync
Open Sales Orders, Purchase Orders, and Pending Work Orders transferred via differential ETL.
Legacy Freeze & Delta Push
Legacy ERP locked read-only. Nightly delta changes executed. Final trial balance extracted.
0.00 Variance Reconcile & Live
CFO signs off on exact penny reconciliation. DNS and user sessions switched to new ERP.
What zero honestly means
The traditional big-bang cutover stops the business on Friday, runs migration scripts across the weekend, and hopes the numbers agree by Monday. For an operation that runs continuously, that is not a plan.
The alternative is not magic. It is arithmetic: the freeze only needs to cover work that genuinely cannot happen earlier. Everything else — history, closed periods, inactive records, master data — can be moved and reconciled weeks in advance, while the business keeps trading.
Continuous load instead of one dump
| Mechanism | When it runs | What it moves |
|---|---|---|
| Initial load | Weeks before go-live | Closed years, historical orders, inactive master records. |
| Differential sync | Nightly from initial load onward | Records created or changed in the legacy system since the last run. |
| Final delta | Inside the freeze window | The last few hours of activity, once the legacy system is read-only. |
A note on terminology, because it causes real scoping errors. In the Microsoft ecosystem, Dual-write is a specific product: the native bidirectional infrastructure linking Dynamics 365 Finance & Operations with Dataverse. It is not a generic name for mirroring writes into a legacy system during migration. Teams that use the word loosely end up budgeting for something they are not deploying.
Staggering the loads
- T-14 days — master data. Chart of accounts, customers, vendors, items. From this point changes in the legacy system are logged and applied to both.
- T-2 days — open documents. Open purchase orders, open sales orders, unapplied cash.
- Cutover day — balances only. Closing trial balance, AR and AP ageing, physical inventory counts.
The runbook
The weekend runs off a minute-by-minute document that names an owner for each step and a duration measured during the mock cutover. The four gates that matter:
- Lock. Legacy ERP goes read-only. Timestamp recorded.
- Final delta. The last differential run executes and is confirmed complete.
- Reconciliation checkpoint. Identical trial balance reports run in both systems. They agree exactly, or the next gate does not open.
- Release. The named sponsor signs. Only then do users get the new system.
Rollback, decided in advance
Before the weekend begins, write down the point of no return and what triggers an abort — typically a reconciliation variance that cannot be explained, or a critical integration failing smoke tests. Name the person who makes the call.
Rollback is credible only because the legacy system was held read-only and not decommissioned. A team that has already started switching off the old system has no rollback, whatever the plan says.
What it comes down to
A short freeze is not a technique applied at go-live. It is the visible result of months of pipeline work, staggered loads and a rehearsed runbook. Teams that go looking for zero downtime in the final week are asking the question far too late.