1 The situation
The old system would be switched off on a fixed date. After that, anything not already extracted was gone. That changes the order of work completely.
So extraction came first, and only extraction. Every investigation that came up went onto a written out-of-scope list until the source data was safely out.
2 What we did
We pulled everything out of the old system, then built and loaded the new one: about 130 of 143 ledger accounts, 446 vendors and 57 customers. Open receivables were re-entered at invoice level and open payables at bill level, so the new books could be worked from day one.
We mapped 1,044 payments to the bills they paid. That link existed nowhere in the exported ledger. It would have been lost with the old system.
Before access ended we archived 1,755 attachments, 2,274 files in all, with a manifest that proves no file has changed since.
3 The tie-out
Every account ties to the cent at 3 dates: two year-ends and the freeze date. It ties for every month, by customer and by vendor, by class and by department.
The client received an 18-sheet workbook and a memo. 721 reconciliation rows and 1,946 matrix cells, all reading 0.00.
4What the migration uncovered
- Two entities in the old system were one legal entity that had been set up twice. Their books did not eliminate against each other.
- The receivables ageing did not tie to the ledger, and that was real data, not a report problem. The opening balance needed three components, not one.
- Exports are snapshots, not archives. A checksum proves a file did not change. It never proves the file was right.
5What the AI did, and what people did
The AI
- Pulled the exports from the old system
- Built the import files and ran the loads
- Wrote the verification scripts
- Produced the archive manifest and drafted the client package
Our accountants
- Made the 3 gating decisions: which book is primary, the cut-over date, the fixed-asset register
- Set the scope
- Held every conversation with the client
- Made every upload click
