There is a conversation that happens in most Indian companies between the 1st and the 5th of every month. Payroll produces a number of payable days. A manager or an employee disputes it. Someone opens the attendance report, counts by hand, and gets a third number. The dispute is settled by whoever is most senior or most persistent, and the same conversation happens again next month about a different employee.
It is tempting to read this as an attendance data problem — that punches are missing, or the biometric device is unreliable. Occasionally that is true. Far more often the attendance data is fine and the rules are the problem: the policy that converts attendance into payable days exists as shared understanding rather than as a written specification, and shared understanding does not survive contact with an edge case.
This article is about writing that policy precisely enough that software can apply it the same way every time, which is the only definition of "consistent" that actually holds.
Why the counts diverge
Attendance answers a factual question: was this person present, and when. Payroll answers a contractual one: how much of the month do we owe them for. These are different questions and they are not connected by arithmetic. They are connected by policy — and every policy decision left unmade becomes a place where two reasonable people count differently.
The gap widens because the two sides usually have different defaults. Attendance systems default to counting days with a record. Payroll defaults to calendar days minus deductions. Those two approaches only agree when nothing unusual happens, and something unusual happens to about one employee in ten every month.
The seven decisions that cause most disputes
1. What is the denominator?
Is a month 30 days, the actual calendar days (28 to 31), or the working days in that month? Each is defensible and each produces a different per-day rate, so the same absence costs different amounts under each. A fixed 30-day denominator makes February expensive for absent employees and January cheap. Actual calendar days make the per-day rate move month to month. Working days require you to have already decided the weekend and holiday rules below.
The decision matters less than making it once, writing it down, and never quietly changing it — a mid-year switch reprices every absence in the year and is close to impossible to explain.
2. Are week-offs paid when the surrounding days are not?
An employee absent Friday and Monday: is the Saturday and Sunday between them paid? Some organisations apply a "sandwich" rule that treats the intervening week-off as unpaid when it is bracketed by unauthorised absence. Others always pay week-offs. Others pay week-offs only if the employee worked a minimum number of days that month.
All three are legitimate. All three are frequently unwritten. And this single decision accounts for a large share of the disputes we see, because the employee's intuition ("I don't work weekends anyway") and the policy's logic point in opposite directions. If you apply a sandwich rule, it must be in the written policy and it must be visible on the payslip, or it will read as an arbitrary deduction.
3. What exactly is a half day?
Half-day rules are where precision collapses. Is it based on hours worked, on arrival time, on a manager's marking, or on the employee applying for half-day leave? What happens when someone works four hours and fifty minutes against a five-hour threshold? Does a half day consume half a leave balance, cost half a day's pay, or both?
The double-count trap is worth naming specifically: a half day that both deducts half a day of pay and deducts half a day from the leave balance is charging the employee twice for one absence, unless that is a deliberate policy choice. This is a genuinely easy mistake to make when attendance and leave are handled by different systems that each apply their own half of the rule.
4. Which record wins when several apply to one day?
A single date can carry an approved leave, a punch record, a holiday and a week-off simultaneously. Someone works on a public holiday while on approved leave — what is that day? Without a stated precedence order, two systems will resolve it two ways and both will be internally consistent.
Write the precedence explicitly, for example: holiday beats week-off, approved leave beats absence, actual attendance beats approved leave. It does not have to match ours; it has to exist.
5. When does an employee start counting?
An employee joining on the 18th has no attendance records for the 1st to the 17th. Any system that fills unmarked days as absent will manufacture seventeen days of absence and either dock the pay or generate seventeen exceptions for someone to clear by hand. The same applies in reverse to leavers, and to anyone on long approved leave.
The fix is a floor and a ceiling on the counting window — nothing before the date of joining, nothing after the last working day, nothing after today. It sounds obvious; it is one of the most common defects in attendance-to-payroll pipelines, and it is invisible until you hire someone mid-month.
6. What happens to a missing punch?
An employee who was demonstrably at work but has no punch — device down, phone dead, forgot — creates a record that says absent. If the only route to fixing it is an email to HR, the correction happens outside the system and the audit trail is a mailbox.
A defined attendance regularization flow makes this ordinary: the employee raises the correction, a manager approves it, the record updates with both the change and the reason attached. The value is not the convenience. It is that the exception becomes data instead of correspondence, and next month you can count how often it happened and to whom.
7. When does the month close?
If attendance for a period can still be edited after payroll has run, your payslips and your attendance reports will disagree permanently and nobody will be wrong. Attendance for a payroll period needs a lock, a defined cut-off, and a documented process for post-lock corrections that carries them into the following month rather than silently rewriting history.
Write the policy as a specification, not as prose
Most attendance policies are written to be read by employees, which is a legitimate goal and a poor specification. A policy that says "habitual late arrival may attract disciplinary action" cannot be implemented. A policy that says "arrival after 09:45 is marked late; three late marks in a calendar month deduct half a day" can be.
A useful test: hand your policy to someone who has never worked at your company, give them a month of raw attendance data and ask them to produce payable days. Every question they have to ask you is a hole in the policy — and each of those holes is currently being filled by an individual's judgement, differently each time.
The second test: could two different people, applying the policy to the same data, produce different answers? If yes, the policy is not yet a policy.
Then make the system apply it
Once the rules are precise, they should live in configuration rather than in anyone's head. Attendance management in AttendancePay is built around this: shift definitions, grace periods, half-day thresholds, week-off and holiday treatment, sandwich rules, overtime eligibility and the precedence between conflicting records are all settings, evaluated the same way for every employee and every day. The output is a payable-days figure with a visible derivation, so a dispute becomes a question about the rule rather than an argument about the count.
Two properties matter more than the specific feature list, whatever system you use:
- The derivation must be inspectable. If payroll says 24.5 payable days, someone should be able to see which days contributed what, and which rule produced each result. A number without a derivation cannot be defended.
- Exceptions should surface before the run, not during it. Missing punches, unapproved regularizations and employees with no records at all are all knowable on the 28th. Discovering them while payroll is running turns a routine correction into a deadline problem.
What good looks like
The realistic goal is not zero disputes. It is that disputes become rare, specific and quick: an employee questions one day, the derivation shows which rule applied, and the conversation is either "the rule is working as written" or "the rule is wrong and we should change it". Both are productive. Neither requires anyone to recount a month by hand.
Companies that reach that state usually did one unglamorous thing first: they sat down with payroll, HR and one sceptical operations manager and wrote down the seven decisions above. The software configuration took an afternoon. The agreement took considerably longer, and it was the part that mattered.
