What migrations actually took

How often the clock restarted

We do not recommend retiring a source system until a shadow ledger has tied to its trial balance across three consecutive closed months. A material difference restarts the count.

Across migrations we have run: the clock restarted once in roughly half of them and twice in about one in six. Two restarts pushes a cutover by a quarter, and it is an uncomfortable conversation every time.

What caused the restarts is the interesting part. Roughly a third were our error — a mapping problem, a mishandled transaction type. The remainder were the customer’s existing books being wrong: an unbalanced period, a duplicate posting, a journal to a suspense account nobody had decomposed.

The migrations that never happened

This is the largest category and the one most worth knowing about. A substantial share of customers who engaged us intending to migrate never did.

They connected their existing ledger read-only, got dimensional reporting and consolidation in two to three weeks, and found that the problem which triggered the search was solved. The ledger was never the constraint.

We count those as successful engagements. They are worth a fraction of what a migration would have earned us, and recommending one first remains the right answer for most companies arriving with a reporting complaint.

What made the difficult ones difficult

• Data condition, every time. The migrations that ran long were the ones where the diagnostic was skipped or shortened. This is the single strongest predictor.
• Report parity. Consistently underestimated. Customers count the reports they use and find twice as many once they start looking.
• Customisation in the source system. SuiteScript, AL extensions, Acumatica framework work. None of it transfers, and where it is substantial we have told customers to stay.
• Decision latency. Where a chart of accounts decision sat with a committee, everything downstream waited. The projects that landed fastest had one named person who could decide within a day.

What we would do differently

Insist on the paid diagnostic before quoting a fixed price, in every case. We have occasionally quoted without one to accommodate a customer’s timeline, and those are disproportionately represented among the engagements that ran long.

And inventory reports earlier. It is a two-hour exercise that consistently changes the timeline estimate, and we now do it during scoping rather than during build.