Skip to content

Migration and cutover

After mapping sign-off, the migration is a controlled load. The agent shows you the target and a transformed sample, waits for your approval, loads in batches with checkpoints, and reconciles source and target.

  1. Ask the agent to load the target.
  2. Review what the agent shows: the target system, the connection it will use and a transformed sample.
  3. Approve the write. The agent asks once for permission to write the listed records before any write.
Permission request in a session: create three vendors in the Northwind NetSuite sandbox, with the request for each vendor and Allow and Decline buttonsPermission request in a session: create three vendors in the Northwind NetSuite sandbox, with the request for each vendor and Allow and Decline buttons
The agent lists each write before it runs. Nothing is written until you select Allow.

After approval, the load starts with the smallest useful batch. The agent checks the result, then continues in batches while preserving checkpoints and rejected records.

The agent stops on an unsafe partial failure. It resumes only from a verified boundary.

You can watch progress on the Runs page and abort a run in progress.

Each batch is a run. With Store run results on, the run keeps each step’s result and the records it wrote. That record is the audit trail of the migration.

Runs page filtered to the vendor load tool, with six successful batch runsRuns page filtered to the vendor load tool, with six successful batch runs
Each batch is a run. Filter the Runs page by the load tool to see every batch.

A load has no undo button. Records written to the target stay there until something removes them. Plan for this before the first load:

  1. Turn on Store run results before the load. Each run then keeps the step results, including the IDs that the target returned for the records it created.
  2. When a batch is wrong, ask the agent to remove it, for example “Delete the vendors that yesterday’s second batch created in the Intacct sandbox.”
  3. The agent reads the run, lists the records it will delete, and asks for permission before it deletes them.
  4. Fix the mapping, reload the batch and reconcile again.

Some records cannot be deleted after other records use them, such as posted journal entries or vendors with bills. For those, ask the agent for reversing entries or to make the records inactive. In a sandbox, a sandbox refresh is often faster than deleting records.

For production, agree with the client before the load how a wrong batch is reversed, and write it into the cutover runbook.

After the load, the agent compares source and target:

  • Record counts
  • Important totals, such as open balances by period
  • Rejected records
  • Duplicates
  • A sample of transformed fields

The agent reports in plain language and calls out only the discrepancies that need action.

Ask for a reconciliation artifact when the client needs a signed document. It covers the agreed scope, the final mapping, validation results, remaining risks and the cutover plan. It exports to PDF or Word.

The migration tools are saved and repeatable. The same saved tools run the sandbox load, the production load and the delta load at cutover, so every load applies exactly the agreed mapping.

A typical cutover is:

  1. A full load into the sandbox and client sign-off on the reconciliation.
  2. A full load into production at the cutover date.
  3. A delta load of records changed since the full load, using the same tools with a date filter.
  4. Reconciliation of production, exported as the handover document.

Plan the cutover as project tasks with owners and a timeline. The client then sees the freeze window, the load sequence and the go or no-go decision in one place. Ask the agent to draft the tasks from the migration scope.

Ask the agent for a cutover runbook with the tasks, owners and timings measured in the sandbox load.

Before the go or no-go call, ask for a summary of the reconciliation of opening balances and open items, and of the open issues. You and the client make the call.

Clients who keep the legacy system running, or whose other applications still feed it, need ongoing syncs. See Post go-live.

For the closing documents and the audit trail, see Handover.