What to check

An accountant's bookkeeping is done monthly for two hundred clients. In month one, every reconciliation was reviewed. In month six, nobody opened one. In month fourteen, a client's accounts were wrong by eleven thousand pounds, and the error had been repeated for nine months.

Nothing failed at month fourteen. Something stopped being checked at month two.

Checking is not distrust

Delegating work and then never looking at it is not trust. It is faith, and faith is a poor operating position for anything recurring. Checking is what makes delegation safe enough to keep doing.

The aim is not to redo the work. It is to know, cheaply and often, whether the work is being done the way it was described.

Sample, do not review everything

Nobody has time to check every instance of recurring work, and checking everything is the fastest route to checking nothing. Sample instead.

*Look at two or three each week.* Pick them without choosing: first, middle, last. If the same person always picks, pick somebody else occasionally.
*Look at the recorded output, not the effort.* The entry in the register, the filed document, the confirmed booking. The record is the evidence; the activity is not.
*Check the check occasionally.* Once a month, confirm that the sampling actually happened. A check that quietly stopped is worse than no check, because it is believed.

Set thresholds, then watch the exceptions

A useful check is usually a number with a line under it.

Every filed return has a confirmation number attached.
No vehicle goes off hire for more than five working days without a note.
No renewal starts later than its lead time allows.
Every document chase stops after four attempts and surfaces to a person.

Thresholds do something sampling cannot. They catch the case that looks normal and is not. A single renewal started late is noise. Nine renewals started late is a description of the whole process failing.

Three failure signals

Some patterns are worth acting on the moment they appear.

*Everything passes, every time.* Either the work is genuinely simple, or the check has stopped being real. Both are worth knowing, and only one of them is likely.

*Surprises arrive only at the deadline.* This usually means the intermediate states are not visible. Work that is going to be late is usually late six weeks before the deadline, and only becomes obvious at the end.

*The same exception repeats every week.* An exception that recurs is not an exception. It is a missing rule, and it should be written into the description rather than handled by hand each time.

Checking is not approving

The two are different jobs, and confusing them causes trouble.

*Checking* asks whether the work matched the description. Anyone reasonable can do it, and it can be done late, in a batch, and by somebody with no stake in the outcome.

*Approving* means taking responsibility for the decision. That needs the mandate, the knowledge and the judgement, and it happens before the act, not after.

An architect signing off a building control submission is approving. An assistant confirming that the submission contains the correct drawings is checking. If the same person is doing both, and particularly if they are doing it at speed, neither is happening.

How much checking is enough

Enough that a failure is discovered while it can still be fixed, rather than in a review after a year. For most recurring work that means a light, regular sample, a small set of thresholds, and one person whose job it is to look.

It should take minutes a week. If the check takes longer than the work did, the check is the wrong check, and the answer is a better threshold rather than more diligence.

Try it

Take the description you wrote in the last lesson.

Write two things about it that could be checked with numbers.
Turn each into a threshold with a line under it.
Write down who looks, and how often.

Five minutes. Then decide the harder question: what would make you stop checking, and is that reason honest.