Ask a manager how much overtime they authorised last month and you will usually get a confident, specific answer. Compare it with payroll and the two numbers rarely match.
The gap is not dishonesty. It is that most overtime is never a decision. It accumulates.
The three kinds, and only one is a decision
Planned overtime is a decision. Someone is short, you extend a shift or add one, and a person with the authority to spend money spent it knowingly. This is the kind everyone thinks about, and it is usually the smallest of the three.
Drift is the fifteen minutes at the end of a shift, four times a week, across thirty people. Nobody approved it because nobody was ever asked. It does not appear in anyone's mental model of overtime, and at any reasonable scale it is larger than the planned kind.
Cover is what happens when someone calls in sick and the person already on shift stays. This one gets approved, usually verbally, usually by whoever is on the floor, and usually with no record beyond the resulting hours.
Cutting overtime almost always means cutting the second and third, and almost every attempt to cut it targets the first.
When it starts is not a detail
Whether an hour counts as overtime depends on the threshold, and the threshold is rarely just "over the contracted hours". It might be daily or weekly. It might be averaged across a reference period. It may be set by a collective agreement rather than by you.
Two consequences worth being concrete about.
First, an hour over on Tuesday may not be overtime at all if the week averages out, and treating it as such overpays. Second, and more expensively, an hour that is overtime may not look like one in a system that only compares against a weekly contracted figure.
Get the threshold definition wrong and every downstream number is wrong in the same direction, quietly, for as long as it takes someone to check.
What you have to be able to show
Whatever your jurisdiction, the practical requirement is the same: hours actually worked, attributable to a person, distinguishable between ordinary and additional, and not silently editable afterwards.
That last one matters more for overtime than for anything else, because overtime is the number most likely to be adjusted after the fact - "we agreed he'd take it as time off instead", "that was a mistake, remove it". Those are legitimate adjustments. An adjustment with no trace of who made it and what it replaced is not.
The cheapest thing that actually reduces it
Not a policy. A weekly comparison.
Put scheduled hours next to actual hours, per person, every week. Drift is invisible in a monthly total and obvious in a weekly one, because five minutes a day is twenty-five minutes a week and it shows up as a pattern rather than a rounding error.
Then look at the direction of the gap. Consistently over means your shifts are longer than you think they are. Consistently under is worse - it usually means people are clocking out and then finishing up, which is unpaid work and a larger problem than the overtime you were investigating.
A short checklist
- Do you know your actual overtime threshold - daily, weekly, averaged, and from which instrument?
- Can you see drift? Scheduled versus actual, weekly, per person.
- Does cover get recorded at the moment it is agreed, or reconstructed from hours afterwards?
- Are post-hoc adjustments traceable? Who changed it, from what, and when.
- Is anyone checking the under-runs? They are the ones that turn into claims.