Getting punches from a device into payroll correctly — and handling the exceptions that cause most of the disputes.
A biometric attendance system captures employee punches by fingerprint, face or card and pushes them to an HRMS, which converts raw punches into attendance, applies shift and overtime rules, and feeds payroll. The device is rarely the problem. The problems are in the rules layer, and in what happens on the days the device did not capture a punch.
Attendance disputes are the most common payroll complaint, and they almost never concern the biometric device. They concern a missed punch, a shift that crosses midnight, an employee who worked at another site, or an overtime rule applied differently from how the supervisor explained it. The integration has to handle exceptions gracefully, because the exceptions are routine.
Punches are pulled or pushed on a schedule. A device offline for a day must backfill, not lose the data.
In and out punches are paired into shifts. Night shifts crossing midnight are the classic place this goes wrong.
Grace periods, late marks, half-days, overtime eligibility and weekly-off handling are applied to the paired data.
Missed punches, manual regularisation and supervisor approval, with a record of who approved what.
Approved attendance becomes payroll input. The cut-off has to be clear, or late regularisations land after processing.
A factory installs a single fingerprint device at the main gate for 400 workers on a shift. Verification takes about three seconds when it works first time, and roughly one in eight attempts needs a retry — wet hands, worn fingerprints, dust.
The arithmetic is unforgiving. At an effective four seconds per person, 400 people take over 26 minutes to clock in through one device. Shift start becomes a queue, workers arrive earlier to avoid it or clock in late through no fault of their own, and supervisors start waving people through, which defeats the system.
The fixes are ordinary: more devices, placed where people actually enter rather than at one gate; a second method for workers whose fingerprints read poorly, which in manual trades is a significant minority; and a regularisation workflow that treats a failed punch as routine rather than as an exception requiring explanation. The common thread is that biometric attendance is a throughput problem first and a technology problem second, and sizing it on the happy path produces a queue.
| Term | What it means |
|---|---|
| Template | The mathematical representation of a biometric, stored instead of the image itself. |
| False acceptance rate | How often the wrong person is accepted. The security-facing error. |
| False rejection rate | How often the right person is rejected. The experience-facing error, and the one that causes queues. |
| Punch | A single attendance event, captured at a device. |
| Regularisation | Correcting a missing or wrong punch through an approval workflow. |
A shift starting 22:00 and ending 06:00 belongs to one shift, not two days. Getting this wrong misstates overtime.
Missed punches are inevitable. Without an approval path they become WhatsApp messages to HR.
An unsynchronised device clock shifts every punch. Time sync is a maintenance task, not a one-off setup.
Regularisations arriving after processing create retrospective corrections every month.
| Metric | What it tells you | What distorts it |
|---|---|---|
| Verification success at first attempt | Device and enrolment quality. Low rates create queues and resentment. | |
| Peak clock-in throughput | People per minute against shift size. The number that determines how many devices you need. | |
| Regularisation requests | How often the system fails the employee. A rising trend is a hardware or enrolment problem. | |
| Attendance-to-payroll exceptions | Records needing manual intervention before payroll can run. |
We build and support the systems this page describes. Start with a free working proof of concept — no cost, no commitment.