Face recognition and GPS attendance: a practical guide for Indian businesses
Registers get signed for absent colleagues and biometric machines break. Here is how selfie, face match and GPS check-in work, what they prevent, and how to roll them out without upsetting your team.
Ask any HR manager where the month-end arguments come from and the answer is usually attendance. A register signed by a friend, a fingerprint machine that stopped working on the 12th, a sales executive who "was at a client" - every one of these ends up as a disputed salary.
Phone-based attendance with a selfie, face matching and location solves most of these problems at once, and it costs far less than putting a biometric device at every site. This guide explains how it works and how to roll it out.
The three checks, and what each one stops
| Check | What it records | What it prevents |
|---|---|---|
| Location (GPS) | Where the phone was at the moment of the punch, and how far that is from the office | Checking in from home or on the way to work |
| Selfie | A photo taken with the same tap as the punch | Someone else punching on your behalf - there is now a picture of who did it |
| Face match | Whether the selfie matches the photo HR enrolled for that employee | Buddy punching, refused at the moment it is attempted |
You do not need all three for everyone. A common setup is location for all staff, a selfie for everyone, and face match switched on where proxy attendance has been a real problem.
How face matching works
HR enrols one clear, front-facing photo per employee. The app converts that photo into a face descriptor - a list of numbers that describes the shape of the face, not the picture itself. At check-in, the selfie is converted the same way and the two descriptors are compared on the server. If they are close enough, the punch goes through; if not, it is refused and the employee is told to try again in better light.
Because the comparison happens on the server, nobody can switch it off by editing the app, and every attempt is logged with its photo for HR to review.
Geofences: getting the radius right
A geofence is a circle around the office. Check-ins inside it are accepted; outside it, they are refused or flagged. Two things decide whether this works smoothly:
- The coordinates. Long-press the building in Google Maps and copy the latitude and longitude exactly. Swapped or truncated numbers are the most common cause of "you are 2,000 km away" errors.
- The radius. Phone GPS indoors is accurate to roughly 10 to 50 metres. A radius of 100 to 200 metres covers the building and the car park without letting people punch in from the tea stall down the road.
Give each branch its own location and radius. Field staff and approved work-from-home days should be exempt from the geofence - but their location should still be recorded with every punch.
Shift rules that stop the edge cases
- Grace time: a check-in after shift start plus grace is marked late.
- A strict check-in window refuses punches that are far too early or far too late, which stops "check in at 6 am, leave, come back at 11".
- Missed check-outs should close automatically a few hours after the shift ends and be flagged, so a forgotten tap does not become a 20-hour working day in payroll.
Rolling it out without a revolt
- Explain why. Fair attendance protects the people who come on time. Say what is recorded (time, location, photo) and who can see it.
- Enrol photos in one session. Good light, plain background, no cap or sunglasses.
- Run a two-week trial with flags only, before you start refusing punches.
- Have a correction route. Phones die and GPS drifts. A correction request that goes to the manager keeps trust in the system.
What about privacy?
Collect only what you use: location at the moment of the punch, not all day (field staff on duty are the exception, and they should know it). Keep photos with the attendance record and restrict who can view them. Tell employees in writing, and include it in your attendance policy.
Done this way, phone attendance is fairer than a register and cheaper than a machine - and the month-end arguments mostly disappear.