What 274 Days of UAE Attendance Data Actually Look Like

What was measured, and what was not

Every figure below comes from live production data, read in one consistent

snapshot on 31 August 2026: 140,775 punches and 40,970 attendance records, from

47 biometric terminals across 43 sites and four companies with active staff,

spanning 274 consecutive days from 1 December 2025.

The punch count is a moving target, because the terminals are still sending. It

rose by fourteen while these queries were being written. Everything here was read

at one moment so the figures reconcile with each other. Read it again tomorrow

and the totals will be larger while the percentages will not have moved.

*Editorial disclosure: Wurxa publishes this and sells attendance software, so

read the numbers with that in mind. Everything here is an aggregate. No company,

site or individual is identifiable, no figure is small enough to single anybody

out, and nothing has been excluded to make a point look better.*

Two limits worth stating plainly. This is a specific shape of employer:

multi-site operations that already run biometric hardware. A single-site

professional services firm would produce a different histogram entirely. And 274

days is three quarters of a year, so seasonal claims are not available.

The week has no shape

If you handed a European operations manager this table they would assume it was

broken. Friday and Saturday, the UAE weekend, carry 14.8 and 14.6 per cent. That

is more traffic than Monday.

The explanation is not that nobody rests. It is that these are operations that

do not close: retail floors, facilities teams, hospitality. The staff rotate; the

site does not. Sunday being the lightest day at 12.9 per cent is the closest

thing to a weekend effect in the whole dataset, and it is a 2.4 point spread

across the entire week.

What follows from that is practical. A roster built on a Monday-to-Friday

mental model will be wrong for most of these sites, and every downstream

calculation, overtime included, inherits the error.

The day has three peaks, not two

Three things stand out.

The largest peak is at 18:00, not 08:00. Eleven per cent of every punch in

274 days lands in that one hour. A single-shift office produces its biggest spike

in the morning; this produces it in the evening, which is the signature of a

changeover, one group leaving as another arrives.

The middle of the morning is dead. From 10:00 to 11:00 the rate falls to 1.4

per cent, lower than midnight. Nothing happens between the morning arrival and

the midday movement, which is exactly what you would expect and is worth stating

because it is the baseline against which anything unusual shows up.

Nights are real, not noise. 22:00 and 23:00 together carry 10.1 per cent, and

midnight carries another 4.0. Roughly one punch in seven happens after 22:00.

Under Federal Decree-Law No. 33 of 2021, work between 22:00 and 04:00 attracts a

50 per cent overtime uplift rather than 25, and shift workers are excluded from

that specific provision. Whether a given punch in that window costs the ordinary

rate or the premium one depends entirely on whether the roster says shift, which

is a data question before it is a legal one.

The true trough is 03:00, at 0.1 per cent. One punch in a thousand.

The number that causes the most trouble

Of 40,970 attendance records, 6,110 have a clock-in and no clock-out. That is

14.9 per cent, roughly one shift in seven.

Records with the opposite problem, a clock-out and no clock-in, number zero.

That asymmetry is the entire finding. It is not that the terminals drop punches

at random; a random fault would produce both kinds in similar quantities. It is

behavioural. People clock in because clocking in is how you are recorded as

present and how the day starts. Clocking out is the last act of a shift, done

while leaving, and nothing visible happens if you skip it.

Every one of those 6,110 records is a day where the hours worked cannot be

computed from the data. Somebody has to decide what happened, and whatever they

decide becomes payroll. That is where disputes begin.

Half the records say "late"

20,296 records, 49.5 per cent, carry a late flag. The mean lateness on those

records is 271 minutes.

Four and a half hours. Nobody is four and a half hours late half the time.

That number is not measuring people. It is measuring the distance between a

schedule and the shift somebody actually worked. When a night-shift employee is

attached to a day-shift roster, every single one of their days is recorded as

hours late, and the flag stops carrying information.

The lesson we took from it, having built the thing that produces the flag: a

lateness metric is only as good as the roster underneath it, and a system that

reports half its records as exceptions has stopped reporting exceptions. Before

anybody acts on lateness, the roster has to be checked against what people are

actually doing.

Overtime is mostly the expensive kind

15,651 overtime hours were recorded in the period. The split:

Almost three quarters of paid overtime is at the premium rate. That follows

directly from the two findings above: if the week is flat, rest-day work is

routine rather than exceptional, and if a seventh of punches happen after 22:00,

night hours are normal operating conditions.

The practical consequence is a budgeting error we suspect is common. An employer

who estimates overtime at 125 per cent, because that is the number in the

handbook, is estimating with the rate that applies to the smaller share of it.

The terminals themselves

47 devices: 44 online, 3 offline at the time of measurement. 43 have sent at

least one punch. Four never have.

Traffic is heavily concentrated. The busiest terminal carries 11,805 punches and

the quietest has sent two. The top five devices account for 27.2 per cent of

all traffic and the top ten for 44.4 per cent. Half the fleet handles a small

fraction of the volume.

That matters for anyone sizing a deployment. The instinct is to treat terminals

as interchangeable and monitor them equally. In practice a handful of doors carry

the business, and an outage on one of those five is a different event from an

outage on one of the quiet forty.

Across all 274 days in the window, every single day has punches. Not one gap.

Ramadan moves the entire day

This is the clearest signal in the dataset, and it is worth its own table. The

window covers Ramadan 2026, 17 February to 19 March, which contributed 13,844

punches across 31 days.

The whole working day slides later. Activity drains out of the middle of the

afternoon and reappears after dark, with a new peak at 20:00 that barely exists

the rest of the year and a genuine early-morning shift between midnight and

02:00 where there is normally almost nothing.

The 22:00 and 23:00 collapse is the most counter-intuitive line in the table. The

rest of the year, those two hours carry 10.8 per cent of traffic between them.

During Ramadan they carry 3.2. Work has not stopped; it has moved either side of

them, to 20:00 and to after midnight.

Two consequences that cost money. Under Federal Decree-Law No. 33 of 2021, the

50 per cent overtime uplift applies between 22:00 and 04:00, so a shift pattern

that moves from 22:00 into 01:00 does not leave the premium window, it moves

deeper into it. And a roster built for the other eleven months will flag a large

share of Ramadan attendance as exceptions, because the schedule it compares

against no longer describes the day.

We had assumed, before looking, that the two-hour statutory reduction would be

absorbed into rosters and not show up in aggregate punch data. That was wrong,

and it was wrong by a wide margin. The reduction is not what shows up. The

reshaping of the day around it is.

The mobile app nobody uses

Of 140,775 punches, 140,764 came from a terminal and 11 were entered by hand.

Not one came from the mobile app, despite the app existing and working.

That is worth reporting against our own interest. The mobile punch is the feature

asked about in every sales conversation, and in this population nobody touches

it. The honest reading is that these are employers whose staff pass a fixed door

at a fixed time, so a phone adds nothing, and the demand we hear about is coming

from a different kind of company than the ones currently running on this data.

What the manual eleven tell you

Eleven punches out of 140,775 were entered by a person rather than captured by a

terminal. That is 0.008 per cent.

It is a small number doing a lot of work. An attendance record that can be edited

freely is worth very little in a dispute, because the other side can argue it was

edited. A record where hand-entry is vanishingly rare, and every instance of it

is logged as such, is a different kind of evidence.

The design principle underneath it, which we would defend in any system: manual

entry has to exist, because terminals fail and people forget, and it has to be

visibly distinct from a captured punch. Not blocked. Marked. The eleven are not

a problem; the fact that you can tell they are eleven is the point.

What we would tell an operator

Four things, in the order they would save you time.

Check your rosters before you look at any lateness report. If more than a

small minority of records are flagged, the flag is describing your schedule, not

your staff.

Treat the missing clock-out as a design problem, not a discipline problem.

One in seven is too many to solve by asking people to remember. Either the last

punch of the shift has to be unavoidable, or the system needs an explicit,

auditable way to close the day.

**Budget overtime at the premium rate if you run past 22:00 or across the

weekend**, because in a flat week the premium rate is the normal one.

Know which five terminals matter. Not all doors are equal, and the monitoring

should not pretend otherwise.

See how Wurxa reads attendance ·

Check overtime rates