The hour-by-hour mechanics of go-live weekend — what freezes and when, the reconciliation gate that closes each wave, four-corner financial reconciliation report by report, what rollback actually means once the first customer invoice posts, and the governance and hypercare model that carries the programme from go/no-go to handover to run.
Chapter 2 covered the engine that moves the data. This chapter covers the weekend it moves it for real: the state everything is in before, during and after the blackout window, the specific reports that prove the numbers agree, what "rollback" actually means once real transactions exist only in the new system, and the governance and…
Before anything else, cutover planning has to answer a deceptively simple question for every object class: at each stage, is SAP open for transactions, closed to users but still readable, or fully frozen — and is D365 not yet live, mid-load, or live?
The blackout window — SAP locked, D365 not yet live — runs approximately 24 hours for a mid-size organisation, and that number is communicated to the business at T-7 days. It is not a fixed constant: GL opening balances alone can take up to 180 minutes per the wave manifest, and a higher-volume financial entity can extend the window…
"Four-corner" reconciliation means proving the same financial truth from four independent angles — the AR, AP, inventory and fixed-asset subledgers — each tied back to the general-ledger trial balance that anchors them. Every corner pairs a named SAP report on one side with its D365 equivalent on the other:
If a No-Go is called at T+24h, rollback is a defined, seven-step procedure, not an improvisation: trigger the abort_run kill switch in the engine's control plane; unlock SAP business users, reversing the SU01 lock;
Each layer of testing exists to catch a defect class the layer above and below it cannot:
Five roles carry the programme: the Programme Manager owns overall timeline, budget and escalation; the ERP Architect owns technical design, architecture sign-off and canonical model governance; the Data Migration Lead owns wave execution, the extraction pipeline and reconciliation-gate management;
Hypercare begins at T+48h, once the go/no-go decision has cleared and end users have had a working day on the new system, and runs for thirty days with a daily drift-dashboard review throughout.
A cutover weekend that works is the product of an explicit state matrix, gates that are binary and named rather than felt, and a governance model that never lets "the team that did it" also be "the team that approved it."