Turning a multi-shift factory's punch logs into audited exceptions
A controlled pilot evaluated a rules-based, traceable exception-review workflow for attendance records.
The manufacturer runs hundreds of staff across rotating shifts. Every month, HR reconciled attendance by hand from three systems that rarely agreed: punch logs, leave applications and overtime requests. The pass took most of a day, still missed cases, and was hard to defend when an employee disputed a deduction.
The rules that decided what counted as an exception lived in a few people's heads and a set of spreadsheets, so the same situation could be judged differently from month to month.
Three sources, never aligned
Punch, leave and overtime each told a different story for the same day.
Rules lived in people's heads
Applied by hand, they were inconsistent and left with whoever knew them.
A full manual pass each month
Slow, easy to miss a case, and hard to defend in a dispute.
One long undifferentiated list
Everything looked equally urgent, so nothing was actionable.
We encoded the rules HR already trusted into an engine that ingests the three daily sources and produces a typed, tiered exception list: high, medium and information, each with the punches, leave and overtime that triggered it attached as evidence.
Nothing is a black box. Every flag links back to the raw records, HR confirms a short list, and the rules are versioned: they only change when HR signs the change, so the engine never drifts from policy.
- Ingest punch logs, leave and overtime applications, and shift tables
- A rules engine with 19+ exception types, tiered by severity
- Evidence and an audit trail attached to every flag
- A sign-off-ready monthly report
- Rules versioned, changed only with HR approval
One expensive problem, proven before it scales. The A1 to POC method, run in weeks not quarters.
Our delivery practices define human decision ownership, source and output traceability, data boundaries, change control and review evidence.
The credential is held personally by Vincent Lin.
Have a problem like this?
Bring your messiest operational problem. We will examine fit, available evidence, the first pass/fail gate, and reasons not to proceed.
Discuss an operations problem