Device Logs Are Evidence, Not the Payroll File

A biometric reader stores identity events: this enrolled template matched at this terminal at this timestamp. That is valuable evidence of who stood at the device. It is not a payroll record. The device does not know the shift, the unpaid break, the leave calendar, the overtime threshold, or whether the person was deployed to a client site that has no reader. Exporting the device log into Tally or greytHR as if it were hours is how present-days and payable hours diverge.

Make software the system of record. The device pushes or is pulled into an attendance application that attaches employee master data, shift rules, and exceptions. Payroll then reads the application, not the terminal. If someone still emails a USB export from the device on the 28th, you have two sources. The USB file will win in a dispute because it looks official, even when it is missing half the field team.

Keep the device log as an audit attachment, not as the import. When an employee challenges a late mark, you should be able to show the raw match time from the terminal and the calculated hours from the rules engine. If those two numbers are supposed to be identical, you have not applied policy. If they are allowed to differ without an explanation field, you have not applied control. Both the raw event and the payable result belong in the period pack.

• Treat biometric events as identity timestamps, not as pay hours
• Use attendance software as the system of record payroll imports from
• Do not let a USB device dump compete with the approved hours file
• Store raw match time and payable hours together for disputes

Map Employee IDs Before the First Export

Payroll breaks first on identity, not on time arithmetic. The number stored on the device is often a short user ID, while the salary register uses an employee code, and the contractor master uses a vendor or labour ID. If device user 0047 is “S. Kumar” and payroll has two Kumars, the export will post to the wrong person or fail the import. Build a mapping table before go-live: device user ID, employee code, full name, employment type, and status.

Never map by display name. Names collide, spellings change, and contractors get reused IDs when a device is not cleaned after exit. When someone leaves, retire the device user ID the same day you exit them in HR. When someone joins, enrol the template only after the employee code exists. A week of temp enrolments at a plant gate is a month of unmatched punches at payroll. If a replacement worker inherits user 0047, last month's Kumar and this month's Sharma will share a salary line.

Validate the map on every export, not once a year. The validation file is simple: every device user ID in the period must resolve to exactly one active employee or contractor code; every paid person who was rostered at a gated site must have at least one mapped ID. Unmapped punches should fail into an exception list, not disappear. Paying only the people who matched is a quiet underpayment of everyone whose ID drifted.

• Map device user ID to employee or contractor code, never to display name
• Retire device IDs on exit the same day HR records the last working day
• Enrol new templates only after the payroll identifier exists
• Fail unmapped punches into exceptions instead of dropping them

Fix Device Clock Drift Before You Trust the Timestamps

India does not change clocks for daylight saving, so timezone conversions are rarely the biometric problem. Device clock drift is. A reader that is seven minutes fast on the 1st and twelve minutes fast on the 30th will move grace, overtime, and half-day triggers without anyone changing policy. Across 200 employees, a systematic 10-minute bias is not noise. It is a paid error that looks like discipline. IST is stable; the cheap wall clock inside the terminal is not, especially after a power cut.

Sync clocks on a schedule and record the last sync. Networked devices should use a time server. Standalone devices need a documented manual check, for example every Monday against a known IST source, with the delta written in the site log. If a device was 18 minutes slow for a week, do not silently rewrite history in payroll. Either accept the raw times and handle affected days as exceptions, or apply a dated correction with the drift amount and who authorised it.

Compare first-in and last-out patterns to the roster, not to the device’s own clock. If an entire night shift appears to check in 11 minutes late every day on one terminal and on time on the neighbouring terminal, suspect the clock, not the crew. A single employee 11 minutes late is a person problem. A whole site 11 minutes late is hardware. Payroll should not levy half-days from a drifting reader.

• Treat IST clock drift as the India-specific time risk, not daylight saving
• Sync networked readers and log manual checks for standalone devices
• Do not silently back-shift a month of punches after discovering drift
• If a whole site is uniformly late, inspect the terminal clock first

Build a CSV Payroll Can Post

A useful biometric export is not a dump of every successful and failed match. Payroll needs a person-day summary, with the punch log available as support. Minimum columns: employee code, name, work date, first in, last out, punch count, device ID, and site or department. The attendance system should then add shift name, unpaid break minutes, regular hours, overtime hours, attendance status, leave code, and exception flag. Sending only first in and last out forces payroll to invent breaks and thresholds.

Choose first-in last-out only when the shift is a single span. A split shift or a call-back needs two pairs, not one 13-hour envelope. If a cook checks in at 08:00, out at 15:00, in at 18:00, and out at 22:00, first-in last-out is 14 hours and will create false overtime. Preserve punch sequence. Pair them in order, apply the shift template, and only then total the day. Hardware that stores unordered logs without pairing is not payroll-ready until software does that work.

Agree the file shape with the payroll owner once. If greytHR wants present days and LOP, derive those from hours and the leave calendar after pairing. If Tally wants overtime hours on a pay head, send the classified overtime column, not a duration between first and last punch. Character encoding, date format, and decimal hours versus days must be written down. A file that opens looking correct in Excel can still fail an import because 08-03-2026 was read as March in one system and August in another.

• Export a person-day summary plus a punch log, not a raw match dump
• Do not use first-in last-out on split shifts or call-backs
• Add shift, break, regular, overtime, status, and leave in software
• Lock date format, ID format, and hours-versus-days with payroll before go-live

Hardware-Only Capture Fails Field and Multi-Site Staff

A wall-mounted reader only sees people who walk up to it. Delivery staff, roaming FM technicians, multi-site security supervisors, and restaurant teams who never pass the back-office device will be absent in the biometric file and present in reality. If payroll imports only the device, those people become LOP. If supervisors then type their hours into a side sheet, you have rebuilt the unofficial process the device was meant to replace. Hardware-only payroll is a plant-gate method pretending to be a workforce method.

Design capture by work pattern. Gated plants and warehouses can use biometric or face devices as the primary mark. Field and client-site staff need a mobile mark with identity and location, written into the same attendance record as the device events. A person who works Monday at a plant and Tuesday at a customer site should not have two attendance systems. They should have two capture methods and one employee day. The export to payroll then has one ID, two source types, and no missing Tuesday.

Do not fake a biometric row for field work. Creating a manual punch on Device 01 so the export looks complete destroys the audit. Use a distinct source type: device, mobile, or approved correction. Payroll does not need to know the hardware model. It does need to know that Tuesday's hours were a verified field mark, not a cloned plant punch. Mixing those without a source flag is how ghost attendance appears in a hardware-only report and how a client later disputes a billed post.

• Readers miss anyone who does not physically reach the terminal
• Use one attendance record with device and mobile capture methods
• Never clone a device punch to hide field work
• Keep source type on the row so payroll can see how the hours were captured

Integrate the Attendance System, Not the USB Stick

A real biometric-to-payroll integration is a scheduled, mapped, approved export from the attendance system of record. It is not a clerk downloading AEBIO or an Excel plugin from the terminal. Integration means employee codes already match, hours are already classified, exceptions are already reviewed, and the payroll import either succeeds or returns row-level errors you can version. If any of those are missing, you still have a USB process with extra software in the middle.

Test with one site and one pay cycle before you retire the USB habit. Compare device first-in last-out, software-calculated hours, and payroll-posted hours for 20 employees including a night shift, a leave day, a missing punch, and a field day with no reader. If those four cases do not reconcile, do not scale. Hardware vendors will say the device is accurate. Payroll will say the import failed. The gap is almost always mapping, pairing, or policy, not the fingerprint algorithm.

Attend Mitra is the connected layer between capture and pay. Face or GPS marks, and device events where you still use them, can sit with shifts, leave, and approvals so the file finance imports is already payroll-ready. Field staff are not dropped because they never touched the plant reader, and the biometric log remains evidence rather than a competing source of truth. That is the difference between exporting a device and integrating attendance with payroll. After the file leaves the device, convert timesheets into payroll and prepare attendance for payroll. Compare capture methods in biometric vs GPS vs face attendance, and keep posting in payroll software.

• Integrate from the attendance system of record, not from a device USB export
• Pilot night, leave, missing-punch, and field-day cases before scaling
• Require row-level import errors and versioned retries
• Keep hardware evidence attached, not in charge of payable hours