Investigating a Wrong Figure

A wrong payment is the visible end of a chain. Investigating it by asking each department whether they are right produces three yeses and no answer.

The sequence

Start at payroll and work backwards, not forwards.

What was paid, and what did payroll receive? If they differ, the cause is in payroll's own processing.

What did time capture send, and does it match what payroll received? If not, the join.

What does time capture hold, and does it match the actual? If not, capture or a correction.

What was scheduled, and does the difference have an explanation? If not, an unrecorded swap or absence.

Four steps, each a comparison rather than a question.

Why backwards

Because the error is visible at the end and the cause is upstream, and each step narrows the search by one system.

Asking forwards — is the schedule right? — starts with the system least likely to be at fault and takes three conversations to reach the one that is.

What to have ready

The chain for that person and period: scheduled, actual, exception decisions, classification, exported figure, paid figure.

Which is the record the time-to-payroll join note argues for, and this is why.

Without it the investigation is a reconstruction, and reconstructions take a week and end in a manual correction.

Fix the person first

Underpayment is urgent and frequently has a statutory deadline.

Correct it before finding the cause, which is a different task on a different timescale.

Tell them what happened and when it will be fixed, rather than waiting until the cause is known.

Then find the cause

A single instance can be a one-off.

Check whether anyone else at that site, in that pay group, or with that shift pattern is affected, which is the question that turns one correction into a fix.

The answer is yes more often than expected, and the others have not noticed.

Recording it

What was wrong, where in the chain, what caused it, what was changed.

Because the same error recurs, and the second investigation should take an hour rather than a week.

And a pattern of causes at one join is the signal to rebuild it rather than to keep correcting.

Check whether anyone else is affected

The question that turns a correction into a fix.

Same site, same pay group, same shift pattern.

The answer is yes more often than expected, and the others have not noticed.

Which means one reported error is usually a sample rather than an incident, and treating it as a sample is the difference between fixing a cause and correcting a symptom.

Turn the boundary into a test

For a concrete product reference, more information here can help turn this boundary into a test. Verify the current behaviour with representative fields and record what reaches the destination.

Independent reference

For a thematic point of reference, see the Cybersecurity and Infrastructure Security Agency. This popular specialist site offers a useful independent reference for the issue.