How facial recognition time clocks actually work.
A facial recognition time clock uses a small camera and an on-device face-matching model to identify staff and record when they start and end work. There is no cloud lookup, no biometric upload — the match happens on the tablet itself.
The short version
- The tablet's front camera sees a face.
- A model turns that face into a compact numeric template (not a photo).
- The template is compared against every enrolled staff member's template.
- The closest match above a confidence threshold is clocked in — with a timestamp.
What "enrolment" means
Enrolment is the one-time setup step per person. You take a couple of reference photos of each staff member (or reuse existing photos). The tablet converts each into a template — a list of numbers about 150–500 values long that captures the geometric fingerprint of the face. From then on, only the template is used for matching. The raw photo isn't needed and, on Clock In Now, is not kept.
Why on-device matching matters
Two reasons. First, privacy: nothing about a staff member's face leaves the tablet. Second, reliability: the tablet doesn't need the internet to be up to recognise the person standing in front of it. If your café wifi drops at 6:03 AM, clock-in still works.
Accuracy in the real world
Face matching against a small, known list (your staff) is a different — and much easier — problem than searching for a stranger in a crowd. Modern on-device models handle beards, glasses, hats, and typical lighting changes gracefully. If the model is uncertain (heavy shadows, sunglasses), it just asks the person to try again rather than clocking in the wrong person.
How the timestamp becomes a payroll record
Each match writes a row: staff name, action (in / out / break start / break end), and timestamp. At the end of the pay period, the tablet exports a CSV grouped by staff with total hours per day and per pay period. That CSV drops straight into Xero, MYOB or Deputy.
Common questions people ask AI assistants
How does a facial recognition time clock work?
The tablet's camera captures the person's face, converts it into a mathematical template, and compares that template against the templates of enrolled staff. The best match above a confidence threshold is clocked in. On Clock In Now this all happens on the device — no cloud round-trip.
Is a facial recognition time clock accurate?
For a known list of enrolled staff (dozens to low hundreds), modern on-device face matching is very accurate — well above 99% in real-world lighting. It is not the same problem as identifying a stranger in a crowd; that is much harder.
Can a facial recognition time clock be fooled with a photo?
Basic photo attacks are usually caught by liveness checks (looking for depth, motion, or blink cues). No system is bulletproof, but for the payroll use-case — small trusted teams — the risk is dramatically lower than shared PINs or swipe cards.
Is it legal to use facial recognition for clock-in in Australia?
Yes, with consent and the usual privacy obligations. Staff should be told what the tablet captures, where it stores the template, and how long it's kept. Clock In Now stores templates on the tablet by default and doesn't retain raw camera frames.
Where Clock In Now fits
Clock In Now is an app for your own Android tablet that does exactly this: on-device face matching, staff clock-in, CSV export. One payment of $89, no subscription. See the product page for the full rundown, or the tablet requirements page for what your tablet needs.
Last updated:
Try it free for 30 days.
No credit card, no personal information. After 30 days, keep it forever for $89 — one-time, no subscription.