Most UAE employers budget overtime at 125 per cent, and most of the overtime they actually pay is at 150. In our production data, 72 per cent of recorded overtime hours fall at the premium rate. If the number in your model is the standard rate, your model is describing the smaller part of the bill.
The measurement
All figures are from live production data on 31 August 2026, in aggregate, across four companies with active staff. No company, site or individual is identifiable.
Recorded overtime hours by rate:
Overtime appears on 5,456 of 40,970 attendance records, so 13.3 per cent of shifts carry some.
Why the split lands this way
Under Federal Decree-Law No. 33 of 2021, ordinary overtime is the basic hourly rate plus 25 per cent. It rises to plus 50 per cent in two circumstances: hours worked between 22:00 and 04:00, which does not apply to shift workers, and work on the weekly rest day where no replacement day is given.
Read as a rule, those sound like exceptions. Read against how these businesses actually run, they are the default.
The week is flat. In the same dataset, the lightest day of the week carries 12.9 per cent of punches and the heaviest 15.3. Friday and Saturday carry 14.8 and 14.6. When a site never closes, rest-day work is not an unusual event, it is the rota.
Nights are ordinary. 22:00 and 23:00 together carry 10.1 per cent of punches and midnight another 4.0. Roughly one punch in seven happens after 22:00, which is precisely the window where the rate changes.
Put those together and the 72 per cent stops being surprising. The premium rate is not attached to unusual work. It is attached to normal work in a sector that runs seven days and into the night.
The budgeting error this causes
Take an operation planning 1,000 overtime hours in a quarter, on an average basic hourly rate of AED 20.
At the headline rate, 1,000 hours at 125 per cent is AED 25,000. That is the figure that goes in the plan, because 125 per cent is the number everybody knows.
At our observed split, 720 hours at 150 per cent and 280 at 125 is AED 21,600 plus AED 7,000, which is AED 28,600.
Fourteen per cent over plan, every quarter, from a mix assumption nobody wrote down. It is not a large error per hour. It is a consistent one, it compounds across the year, and it is invisible because the overtime itself was authorised. Nobody overspent. The estimate was just built on the wrong rate.
Where it hides
It is not on the payslip as a surprise. Each individual payment is correct. The gap only exists between the plan and the total, which is a finance conversation rather than a payroll one, and those two conversations often do not meet.
It is not visible in hours. Hours worked look exactly as expected. It is the cost per hour that differs, so any report denominated in hours will show nothing wrong at all.
It moves with the roster, not with demand. Shifting a shift from 21:00 to 22:30 does not change the work, the headcount or the hours. It changes the rate on every hour after 22:00. A roster change made for operational reasons can move the cost meaningfully without anybody framing it as a cost decision.
That last one is the most useful thing in this post. Rosters are usually designed by operations against coverage, and priced by finance after the fact. The 22:00 boundary means the two are the same decision.
Ramadan makes it worse, not better
The instinct is that Ramadan reduces cost, because statutory hours are reduced by two per day. In our data the shape of the day changes in a way that pushes work further into the premium window rather than out of it.
During Ramadan 2026, punches at 01:00 rose from 0.3 per cent of traffic to 4.6, and 20:00 rose from 1.2 to 5.8, while 22:00 and 23:00 collapsed from 10.8 per cent between them to 3.2. Work did not stop at 22:00. It moved either side of it, and the side it moved to after midnight sits squarely inside the 22:00 to 04:00 band.
Shorter days, more of the remaining hours at the premium rate. Whether that nets out cheaper depends entirely on your own rota, which is the point: it is a question to compute, not to assume.
The 14.9 per cent that makes the number soft
There is an honesty problem underneath all of this, and it is ours as much as anybody's.
Overtime is computed from hours worked, and hours worked are computed from a clock-in and a clock-out. In the same dataset, 6,110 of 40,970 attendance records have a clock-in and no clock-out. That is 14.9 per cent, roughly one shift in seven, where the end of the shift was decided by a person rather than recorded by a terminal.
Which means a share of that 15,651 hours rests on inference. Not a large share, and not a random one either: missing clock-outs cluster on shifts that end at handovers and at sites with a single congested exit, which are disproportionately the shifts that run late and therefore the ones most likely to attract overtime in the first place.
We are not going to pretend to a precision the data does not support. The 72 per cent split is robust, because both sides of it come from the same records and any estimation error affects both. The absolute hours figure is softer. Anybody running this analysis on their own data should check their own incomplete-record rate first, because it sets the confidence interval on everything downstream.
A worked example of the boundary
The clearest way to see the roster-as-pricing point is with one shift.
A site needs eight hours of evening coverage. Two ways to schedule it:
14:00 to 22:00. Every hour sits outside the night window. Any overtime attached to it is priced at 125 per cent.
16:00 to 00:00. The last two hours fall after 22:00. Any overtime in those two hours is priced at 150 per cent.
Same eight hours, same person, same work, same headcount. The second pattern costs more per overtime hour, and the difference only exists because of where the shift was placed on the clock.
Neither is wrong. There are good operational reasons to choose either. The point is that the choice has a price attached, and in most companies the person making it cannot see the price at the moment they make it. That is a reporting problem, not a scheduling one, and it is fixable.
What to do about it
Work out your own split before you budget anything. Take last quarter's overtime hours and separate the ones after 22:00 and on rest days from the rest. That single ratio is worth more than any benchmark, including ours.
Price rosters, not just staffing. When a shift pattern moves across 22:00 or onto a rest day, the cost changes without the headcount changing. Somebody should be able to see that number before the rota is published.
Check who is exempt. The night uplift does not apply to staff working shift patterns. Whether a given person counts as a shift worker is a contractual and factual question, and getting it wrong is expensive in both directions: pay the premium when you need not, or fail to pay it when you must.
Watch the concentration. Overtime spread across many people is a scheduling reality. Overtime concentrated on a few is usually a vacancy being covered at 150 per cent, which is the most expensive way to be short-staffed.
Nothing here is legal advice, and a specific case should go to somebody qualified. It is arithmetic and observed data, offered because the arithmetic is usually done with the wrong rate.
Wurxa does this arithmetic on every shift rather than once a quarter: overtime is calculated from real, roster-aware punches at 125% or 150%, the night window and rest days are applied per site, and the overtime report sits next to WPS payroll in the same workspace. Punches come from the ZKTeco terminals you already own through Wurxa attendance, so the split in this post is a report you can open, not a project. Every new workspace gets every feature for 14 days with no card. Start your free trial, or work out a specific figure first.
Put the workflow into practice.
Start with every Wurxa feature for 14 days. No credit card required.
