Every migration programme — ERP, core banking, policy administration, a CRM estate, a mainframe workload — converges on the same task. Before cutover, and during parallel running, someone must prove that the target system equals the source. Not approximately, not on two hundred sampled accounts, but in full: every record, every balance, every derived figure the business will be held to afterwards.
It is the least glamorous work in the programme and the part everyone remembers when it goes wrong.
How it is usually done
Extracts. A team pulls samples from both systems into spreadsheets, reconciles what it can, documents the differences it finds, and signs off on the rest. The sample is small because the work is manual, and it is manual because the two systems were never designed to be queried together — which is, not coincidentally, the problem this whole site is about.
Two things follow. The coverage is a fraction of the estate, so the defects that matter are the ones the sample did not touch. And the comparison is stale the moment it finishes, because both systems kept moving while the spreadsheet was open.
The same question, asked of both systems live
The alternative is to read both estates in place — the legacy and the target, side by side, read-only — and ask the same question of both, continuously, during parallel running. Every account, not a sample. Today's data, not last week's extract. The differences surface as an exception list: this record migrated with a rounding difference, this balance depends on a derivation the target implements differently, these four hundred records have not arrived at all.
Each exception carries its evidence, so the conversation with the migration team is about specific records and specific rules, not about whose spreadsheet is newer.
What you keep when the programme ends
The comparison is not throwaway. The connections, the entity definitions and the reconciliation rules built for the migration are the foundation of a governed data layer over both estates — which matters, because most organisations do not switch the legacy off at cutover. They run both, sometimes for years. The assurance apparatus becomes the operating picture of the estate the business actually has.
And the record of the proof itself — what was compared, what differed, what was accepted and by whom — is retained and re-runnable. When the regulator or the auditor asks, a year later, how the bank knew the migration was correct, the answer is a query, not an archaeology project.