A waiter works a split shift on the Tuesday — lunch, a gap, then dinner. The clock-in system records two entries. The roster says one day. Payroll, which is a spreadsheet, is keyed from a printout.
At the end of the month the payslip is short by four hours and the waiter queries it. Establishing what actually happened means finding the printout, the roster and the clock records, deciding which is authoritative, and doing the arithmetic again by hand.
It takes a manager two hours. The discrepancy was worth forty pounds. And because the correction is made in the spreadsheet rather than at the source, the same thing happens again the following month.
The structural fix is that hours come from one calculator and payroll is generated from attendance rather than transcribed from it — so a timesheet and a payslip cannot disagree, because they are the same data.
What makes that survivable in practice is that generation is idempotent and repeatable: corrections arrive all month, "regenerate from attendance" can be pressed as often as needed, and any line a manager has deliberately locked survives every regeneration untouched.
Why this matters to you: payroll queries are expensive twice — the time to answer them and the trust they cost. In an industry with high turnover, being visibly accurate about pay is a retention feature.