Reconcile, and show only the exceptions

Reconciliation produces a long document that proves almost everything was fine. The proof is necessary and reading it is not, which is why the exceptions — the only part anyone needs — arrive at the end of a report people skim.

Invert it. Report the exceptions and say how many matched.

Say what counts as an exception before you start

Not everything that differs is a problem. A payment received a day either side of its expected date is normal. A rounding difference of a few pence is normal. A missing transaction is not.

Write your tolerances down. Without them the report will either flag everything or quietly absorb real differences, and you will not know which.

Ask for the short list

Every Monday at 7am, reconcile last week's accounts.

Report only exceptions:
  amounts that differ by more than [tolerance]
  entries present on one side and missing on the other
  anything dated more than [n] days from its expected date
  anything you could not categorise at all

For each: what it is, both figures where there are two, the dates, and
where it came from.

End with one line: how many entries reconciled cleanly, so I know the scope
the exceptions came from.

If there were none, say so and stop.

The count of clean matches is the one number worth keeping. Six exceptions out of six hundred is a good week; six out of twelve means something broke upstream. Without the denominator the exception list reads the same either way, and you will react to the wrong one.

Never let it write back

Reconciliation is a read. The moment the same job can also adjust entries, you have an automated system quietly making your books agree with themselves, and the reconciliation stops being evidence of anything.

This is the boundary in this playbook worth being inflexible about.

Expect the tolerances to be wrong at first

The first two or three runs will over-report. Each is a one-line adjustment, and the list should settle within a month at genuinely unresolved items.

If it never settles, the exceptions are real and the problem is upstream of this report.

What good looks like

A Monday list of four things instead of a forty-page statement, with a count telling you how much was checked to find them.

The measure is whether the same exception appears three weeks running. If it does, it is not an exception, it is a process nobody has fixed.