The First Quarter After a Change

Whatever changed — a new module, a rebuilt join, a migration — the first quarter determines whether the estate is trusted.

Watch weekly, not monthly

The reconciliations, all four, every week rather than every month.

Manual corrections, counted.

Interface failures and retries.

Questions arriving from managers and from workers, which indicate where the explanation did not land.

Anything unresolved at week three becomes the story of the change.

Run the old process alongside

For at least two full periods, compared per person.

Totals match while individuals are wrong in both directions, which is the trap and the reason for per-person comparison.

Do not switch off the old process until its replacement has reconciled cleanly for a full period, signed off in writing.

Fix visibly

When someone reports a wrong figure, fix it and say what caused it.

"The department code changed and the mapping had not been updated" is a better answer than "it has been corrected", because it tells people the estate is understood.

Speed matters more than elegance in the first month.

Be generous with corrections

Volume will be high because everything is new.

Approve readily and investigate the pattern afterwards, rather than querying each one — querying teaches people to stop submitting and absorb the error.

Tighten later, once the reconciliation is clean and an ordinary correction is distinguishable from a signal.

Watch for the workarounds forming

A spreadsheet appearing to bridge a gap.

Someone retyping something.

A manager keeping their own version of the rota.

Each is a signal that a join does not work, and each becomes permanent if it survives the first quarter.

Find them by asking, not by looking.

The review at week twelve

All four reconciliations, before and after.

Manual corrections, before and after.

What broke and what caused it.

And one honest question: does anyone still not trust a number the estate produces? Usually somebody does not, and the reason is specific and fixable.

Find the workarounds forming

By asking, not by looking.

A spreadsheet bridging a gap, someone retyping, a manager keeping their own rota.

Each is a signal that a join does not work.

And each becomes permanent if it survives the first quarter, at which point it is load-bearing and much harder to remove than to prevent.

Reproduce the workflow

For another way to make this requirement testable, consult the service-desk example. Reproduce the case with real roles, codes, failures and recovery steps.