The choice is usually framed as accuracy. Commercially it is not. It is about whether a credential can be handed to somebody else, because that single property decides whether your attendance data is a record or an estimate.
The only difference that matters commercially
A card can be lent. A fingerprint cannot. Everything else — speed, cost, hygiene, hardware life — is secondary to that, because buddy-punching is the loss that attendance systems exist to remove and cards do not remove it.
So the honest starting position for a factory is biometric, and the interesting question is where that runs into practical trouble.
Where cards still make sense
- Contractors and visitors, where enrolling a fingerprint for a one-day visit is more friction than it is worth, and a numbered card returned at the gate is cleaner.
- Very high-throughput gates at shift change, where a tap is faster than a placed finger and a queue of two hundred people is a real operational cost.
- Gloved or sterile environments, where removing a glove to punch defeats the process it sits inside.
- Vehicle access, where a tag read from inside a cab is the only sensible option.
Note that all four are edge cases around a workforce, not the workforce itself. That is the pattern most plants land on.
Where fingerprint struggles on a shop floor
This is the part vendors gloss over, and it is the reason your floor is asking the question.
- Worn or abraded ridges. Genuinely common in manual trades — handling, masonry, chemicals, constant washing. A proportion of any manual workforce will read poorly no matter which sensor you buy.
- Dry, cracked or callused skin, which varies seasonally, so enrolment that worked in monsoon fails in winter.
- Oil, grease and dust on the finger and the sensor, degrading reads and requiring the platen to be cleaned far more often than anyone plans for.
- Throughput at shift change, where a failed read means a retry and the retries compound into a queue.
The mitigations are real but partial: enrol multiple fingers per person, clean the sensor on a schedule, and re-enrol seasonally for the affected group. Any vendor who tells you a sensor eliminates this problem entirely is overselling.
Face recognition, and what it changes
Face recognition sidesteps the skin problem completely, which is why it has taken over on manual shop floors. It is contactless, fast at shift change, and cannot be lent any more than a fingerprint can.
It brings its own constraints, and they are placement constraints rather than hardware ones: it needs consistent lighting at the terminal, it dislikes strong backlight from a doorway behind the user, and helmets, masks and safety glasses affect reads depending on the system. Terminal siting therefore matters as much as terminal choice — get it out of the glare of an open shutter.
It also raises a legitimate consent and data question with your workforce that a card does not. Deciding how face data is stored, for how long, and who can access it — and telling staff — is part of the implementation, not an afterthought.
Running both, and what to specify
Most factories we work on end up with a combination, and it is a coherent design rather than a compromise: face or fingerprint for the permanent workforce, cards for contractors and visitors, tags for vehicles, all reporting into one system.
What to put on the specification:
- Terminal type per gate, with the reason — throughput, environment, or user group
- Enrolment plan, including multiple fingers per person and who re-enrols seasonal failures
- Expected throughput at peak shift change, and the terminal count that supports it
- Export format into your payroll, tested before handover rather than promised
- What happens when a terminal fails — the fallback process, in writing
- Cleaning interval for contact sensors, as a scheduled line in the AMC
That fifth point is the one that causes disputes. A plant with no agreed fallback ends up with a paper register on the day a terminal dies, and that day’s payroll becomes an argument.
Common questions
What proportion of workers cannot use fingerprint reliably?
It varies enormously by trade and by season, and the honest answer is that you find out at enrolment rather than from a datasheet. The practical approach is to enrol the whole workforce, identify the group that reads poorly, and give that group an alternative - face or card - rather than fighting the sensor.
Does the attendance system integrate with our payroll?
Most export a file your payroll can import, but the format matters and so does how your payroll treats shifts, overtime and half-days. We test an actual export into your actual payroll before handover, because a mismatch discovered at month-end is a much larger problem than one found during commissioning.
Can the same terminal control the door as well as record attendance?
Usually yes, and on many sites that is exactly the right design - one credential, one record, entry controlled by the same event. Where it needs care is fire egress: a door on an escape route has to release safely, which is a life-safety requirement and should be confirmed against your fire clearance rather than assumed.