An attendance policy that actually computes
Grace, lateness, half-days, regularisation, shifts: the rules every establishment improvises until they collide with payroll. Here is how to write a policy a system can execute and a workforce can trust.
Rates and thresholds re-verified 27 July 2026. Statutory figures change by notification, so always confirm against the current official source before acting.
What an attendance policy is for
An attendance policy converts presence into pay, fairly and predictably. It defines the working day (shifts, timings, breaks), how presence is captured (device, web, mobile, with location rules where relevant), what tolerance exists (grace periods, late marks), what partial presence means (half-day thresholds), how errors get corrected (regularisation with approval), and what the consequences are (late-mark accumulation, LWP). Written well, it is arithmetic: any punch pattern maps to exactly one outcome, computable by a system and predictable by the employee. Written badly (or not at all) every month ends in discretion, and discretion at pay time reads as favouritism even when it is not.
The rules that need explicit answers
Grace: how many minutes, how many times a month, and what the (n+1)th late arrival costs. Half-day: the threshold hours that split a day, and whether they measure from shift start or total presence. Missed punches: the regularisation window, the approver, and the monthly cap before it becomes a counselling conversation. Overtime: what counts (rostered? approved? any excess?), at what rate, against which statutory limits. Weekly offs and holiday work: comp-off or premium, with expiry. Shift boundaries: which day owns a night shift that crosses midnight (the roster's day — decide it once). Every undecided rule above is currently being decided ad hoc by whoever closes attendance: the policy's job is to fire that improvisation.
Writing the policy, rule by rule
-
1
Define the day
Shifts with timings, breaks and the midnight rule; weekly offs and their rotation; the holiday calendar per location. The day's definition is the base every other rule references.
-
2
Define capture
Who punches how (device, web, mobile, geo rules) and what a missing punch means by default. Capture design determines how much regularisation you will live with.
-
3
Define tolerance
Grace minutes, monthly caps, late-mark consequences, half-day thresholds. Tolerances are policy generosity made precise — decide them once, apply them mechanically.
-
4
Define correction
Regularisation windows, approvers, caps and logging; who can override what, with every override attributed. Correction design is where fairness is most visible.
-
5
Define consequence and connect to pay
Late-mark conversions, LWP for excess absence, overtime rates against captured hours; then let the attendance close feed payroll directly, so the policy's output is the pay input with no transcription between.
A worked example
A 90-person unit writes its first real policy: two shifts with 12 minutes' grace thrice monthly, half-day below 4.5 hours, regularisation within 3 days capped at 4 a month, overtime only when rostered or pre-approved at the statutory rate, night shift owned by its roster day. The first computed month surfaces exactly what improvisation had been hiding: two chronic 9:40 arrivals that had never been visible, one supervisor whose team's 'full attendance' was nightly regularisation, and 31 overtime hours that were real and had been paid flat. Three awkward conversations later, the second month closes in an hour, and the payday queue at HR's desk simply stops forming.
Figures are illustrative, for mechanism only. Verify current rates and your own structure before relying on any number.
Biometric, mobile, web: choosing capture honestly
Capture should fit the workplace, not the vendor's brochure. Fixed workforces with gates suit biometric devices — fast, fraud-resistant, no phone dependency. Distributed and field staff suit mobile punches with policy-controlled geo-stamping — capture that travels. Office and hybrid teams suit web punches inside the tools they already use. Most real establishments need two of the three, and the policy should say which roles use which, plus the fallback when a device fails, because 'the machine was down' is either a documented exception path or a monthly argument.
The trust dimension: policies employees can predict
The measurable sign of a good attendance policy is not compliance; it is the absence of surprise. An employee who can predict exactly what a 9:52 arrival costs, see their late-mark count mid-month, and watch a regularisation request move through approval treats the system as fair even when its answer is no. Opacity, not strictness, is what breeds the gaming and grievance cycles, which is why the policy belongs in self-service where everyone reads the same rules, and why every consequence should be visible before payday rather than discovered on the payslip.
How payroll software applies this
In PeopleDeck the policy is configuration the engine executes: shifts and grace rules evaluate punches as they land, half-days and late marks apply themselves, regularisations route with caps and logs, overtime computes from captured hours at your rates, and the attendance close feeds payroll with no export in between. The attendance register and the wage register reconcile because they are one record, which is what an inspection wants and what employees, in a different vocabulary, call fairness.
Go deeper: Attendance & leave · The payroll engine · Leave policy
Primary sources: Ministry of Labour & Employment
Maintained by Databus Technology Solutions against the source notifications; the verification date above is refreshed whenever a figure changes. This guide explains rules and mechanics; it is not legal or tax advice. PeopleDeck applies statutory rates and generates upload-ready files; it never files returns on your behalf, and positions on contested questions belong with your consultant.
Frequently asked questions
What is a reasonable grace-period design?
The common pattern: 10-15 minutes' grace, usable a capped number of times monthly (often three), with excess late marks converting to a half-day or leave deduction per policy. The design matters less than the cap and consequence being written and machine-applied — uncapped grace is no policy, and unwritten consequence is grievance fuel.
How should regularisation work without becoming a loophole?
Windowed (request within a few days), approved (by the manager who can verify), capped (a monthly count that flags chronic patterns), and logged (every regularisation an auditable record). The cap converts a data-quality tool back from an attendance-laundering one.
Can attendance shortfalls deduct from pay?
Absence beyond leave balances is LWP and pro-rates pay lawfully: the deduction must trace to recorded absence under a written policy. Punitive deductions beyond the wage math (fines) are tightly restricted under the wage framework; consequence design should live in leave and late-mark rules, not invented penalties.
What about field staff who can't punch at an office?
Mobile punches with optional geo-stamping per your policy, duty marking for client visits, and route-based patterns for delivery and sales staff. The policy states what capture is expected per role: the mistake is holding field roles to desk-role capture and then regularising constantly.
Does the policy need to differ by department?
Rules can vary by role class (shift workers, field staff, office staff) but should be identical within a class. Cross-class envy is manageable; within-class inconsistency is the credibility killer, and the reason policy variation should be structural, not managerial.
How do attendance records interact with inspections?
Attendance registers are core inspection artifacts under the labour framework, and they must reconcile with wage records — paid days on the payslip tracing to presence records. Systems that compute pay from attendance produce that reconciliation by construction; parallel records produce findings.
How should the policy handle device failures and power cuts?
With a written exception path: supervisor-attested manual marking for the outage window, logged as such, reconciled when the device restores. The pattern to avoid is silent bulk regularisation; it destroys the record's credibility. An outage that produces an attested, dated exception log is a non-event at any inspection.
Policies that compute, months that close.
Applied on every payslip, files generated for upload — per employee, per month.
Start free